Span-based Named Entity Recognition by Generating and Compressing Information

TL;DR

This paper introduces a novel span-based NER system that combines generative and information compression IB models, leading to improved performance across multiple datasets.

Contribution

It is the first to integrate both generative and compression IB models into a unified span-based NER framework, enhancing entity recognition accuracy.

Findings

Joint training of IB models improves NER performance

Incorporating span reconstruction maintains span information

Synonym generation enhances representation robustness

Abstract

The information bottleneck (IB) principle has been proven effective in various NLP applications. The existing work, however, only used either generative or information compression models to improve the performance of the target task. In this paper, we propose to combine the two types of IB models into one system to enhance Named Entity Recognition (NER). For one type of IB model, we incorporate two unsupervised generative components, span reconstruction and synonym generation, into a span-based NER system. The span reconstruction ensures that the contextualised span representation keeps the span information, while the synonym generation makes synonyms have similar representations even in different contexts. For the other type of IB model, we add a supervised IB layer that performs information compression into the system to preserve useful features for NER in the resulting span representations. Experiments on five different corpora indicate that jointly training both generative and information compression models can enhance the performance of the baseline span-based NER system. Our source code is publicly available at https://github.com/nguyennth/joint-ib-models.