Overview of Discontinuous Name Entity Recognition (DNER)

Named Entity Recognition (NER) is a foundational task in natural language processing that identifies and classifies entity mentions in text. While standard NER assumes entities occupy contiguous token spans, a significant class of entities particularly in biomedical and clinical text are discontinuous, spanning multiple non-adjacent fragments within a sentence. Discontinuous NER (DNER) poses unique challenges: conventional sequence labeling frameworks cannot represent fragmented spans, and manual annotation of discontinuous entities is costly and requires specialized domain expertise.

This presentation provides a comprehensive overview of methods developed for DNER. We survey the major paradigms, including sequence labeling extensions, hypergraph-based, graph-based, transition-based, span-based and relation-based, grid-tagging and word-pair, and generative (sequence-to-sequence) approaches, analyzing their representational assumptions, strengths, and limitations. We further examine few-shot supervised NER as a low-resource supervised setting, where limited labeled examples constrain model generalization, and distantly supervised NER (DS-NER), which leverages external knowledge bases to automatically generate noisy training labels, reducing annotation cost at the expense of label quality.

Despite independent progress in these areas, a critical gap remains in the literature: the intersection of distant supervision, few-shot learning, and discontinuous entity recognition has received little attention. Existing DS-NER methods are designed for contiguous entities and fail to capture fragmented spans, while few-shot methods similarly lack mechanisms for discontinuous structure. We identify this gap as an underexplored and practically important research direction, particularly for low-resource domains. We discuss potential solutions that could support progress in this area.

Committee: Qi Li (major professor), Tichakorn Wongpiromsarn, Mengdi Huai, Hongyang Gao, and Ali Jannesari