Spanbert: Improving pre-training by representing and predicting spans

Mandar Joshi, Danqi Chen, Yinhan Liu, Daniel S. Weld, Luke Zettlemoyer, Omer Levy

Research output: Contribution to journalArticlepeer-review

158 Scopus citations

Abstract

We present SpanBERT, a pre-training method that is designed to better represent and predict spans of text. Our approach extends BERT by (1) masking contiguous random spans, rather than random tokens, and (2) training the span boundary representations to predict the entire content of the masked span, without relying on the individual token representations within it. SpanBERT consistently outperforms BERT and our better-tuned baselines, with substantial gains on span selection tasks such as question answering and coreference resolution. In particular, with the same training data and model size as BERTlarge, our single model obtains 94.6% and 88.7% F1 on SQuAD 1.1 and 2.0 respectively. We also achieve a new state of the art on the OntoNotes coreference resolution task (79.6% F1), strong performance on the TACRED relation extraction benchmark, and even gains on GLUE.1

Original languageEnglish (US)
Pages (from-to)64-77
Number of pages14
JournalTransactions of the Association for Computational Linguistics
Volume8
DOIs
StatePublished - 2020

All Science Journal Classification (ASJC) codes

  • Artificial Intelligence
  • Human-Computer Interaction
  • Computer Science Applications
  • Linguistics and Language
  • Communication

Fingerprint

Dive into the research topics of 'Spanbert: Improving pre-training by representing and predicting spans'. Together they form a unique fingerprint.

Cite this