Boundary Smoothing for Named Entity Recognition

TL;DR

This paper introduces boundary smoothing, a regularization technique for span-based neural NER models that improves performance, calibration, and reduces over-confidence by re-assigning entity probabilities to surrounding spans.

Contribution

It proposes boundary smoothing, inspired by label smoothing, to address over-confidence and boundary ambiguity in neural NER models, achieving state-of-the-art results.

Findings

Boundary smoothing improves NER performance on multiple benchmarks.

It enhances model calibration and reduces over-confidence.

The technique leads to flatter minima and smoother loss landscapes.

Abstract

Neural named entity recognition (NER) models may easily encounter the over-confidence issue, which degrades the performance and calibration. Inspired by label smoothing and driven by the ambiguity of boundary annotation in NER engineering, we propose boundary smoothing as a regularization technique for span-based neural NER models. It re-assigns entity probabilities from annotated spans to the surrounding ones. Built on a simple but strong baseline, our model achieves results better than or competitive with previous state-of-the-art systems on eight well-known NER benchmarks. Further empirical analysis suggests that boundary smoothing effectively mitigates over-confidence, improves model calibration, and brings flatter neural minima and more smoothed loss landscapes.