Nested Named Entity Recognition as Single-Pass Sequence Labeling

TL;DR

This paper presents a novel approach to nested named entity recognition by transforming it into a sequence labeling task, enabling efficient and effective recognition of nested entities using standard encoders and linearization techniques.

Contribution

It introduces a single-pass sequence labeling method for nested NER that simplifies the problem and achieves competitive performance with existing complex systems.

Findings

Achieves competitive results on nested NER benchmarks.

Reduces complexity by linearizing constituency structures.

Enables training with off-the-shelf sequence labeling tools.

Abstract

We cast nested named entity recognition (NNER) as a sequence labeling task by leveraging prior work that linearizes constituency structures, effectively reducing the complexity of this structured prediction problem to straightforward token classification. By combining these constituency linearizations with pretrained encoders, our method captures nested entities while performing exactly n tagging actions. Our approach achieves competitive performance compared to less efficient systems, and it can be trained using any off-the-shelf sequence labeling library.