# SpanBERT: Improving Pre-training by Representing and Predicting Spans

## TL;DR

SpanBERT enhances pre-training by focusing on span-level representations and span masking, leading to significant improvements in span-related NLP tasks like question answering and coreference resolution.

## Contribution

It introduces span masking and span boundary prediction techniques, advancing BERT's capabilities for span understanding and improving performance across multiple NLP benchmarks.

## Key findings

- Outperforms BERT on span tasks
- Achieves 94.6% F1 on SQuAD 1.1
- Sets new state-of-the-art on coreference resolution

## 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 BERT-large, 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 show gains on GLUE.