GapDNER: A Gap-Aware Grid Tagging Model for Discontinuous Named Entity Recognition

TL;DR

Discontinuous NER in biomedicine faces non-adjacent and overlapping entity spans, leading to decoding ambiguity and error propagation. GapDNER introduces a gap-aware grid tagging approach that treats context gaps as spans, combines intra-span regularity (biaffine boundary features and linear attention) with inter-span relations (criss-cross attention), and employs BFS-based decoding over a token-pair grid. The method achieves state-of-the-art F1 on CADEC, ShARe13, and ShARe14, with ablations validating the contributions of gap modeling, intra-span regularity, and inter-span relations. This work offers a decoding-friendly, scalable framework for modeling complex biomedical entities and provides insights for grid-based NER on discontinuous mentions.

Abstract

In biomedical fields, one named entity may consist of a series of non-adjacent tokens and overlap with other entities. Previous methods recognize discontinuous entities by connecting entity fragments or internal tokens, which face challenges of error propagation and decoding ambiguity due to the wide variety of span or word combinations. To address these issues, we deeply explore discontinuous entity structures and propose an effective Gap-aware grid tagging model for Discontinuous Named Entity Recognition, named GapDNER. Our GapDNER innovatively applies representation learning on the context gaps between entity fragments to resolve decoding ambiguity and enhance discontinuous NER performance. Specifically, we treat the context gap as an additional type of span and convert span classification into a token-pair grid tagging task. Subsequently, we design two interactive components to comprehensively model token-pair grid features from both intra- and inter-span perspectives. The intra-span regularity extraction module employs the biaffine mechanism along with linear attention to capture the internal regularity of each span, while the inter-span relation enhancement module utilizes criss-cross attention to obtain semantic relations among different spans. At the inference stage of entity decoding, we assign a directed edge to each entity fragment and context gap, then use the BFS algorithm to search for all valid paths from the head to tail of grids with entity tags. Experimental results on three datasets demonstrate that our GapDNER achieves new state-of-the-art performance on discontinuous NER and exhibits remarkable advantages in recognizing complex entity structures.

Figures (9)

• Figure 1: Example of three discontinuous entities ("severe joint pain", "severe shoulder pain" and "severe upper body pain") within the same sentence. The blue and red dashed boxes represent the fragments of discontinuous entities and context gaps respectively.
• Figure 2: An example of our gap-aware grid tagging scheme. Blue and red fonts represent fragment and gap labels respectively, while green fonts denote entity type labels. The black dotted line covers the diagonal area of the grid.
• Figure 3: The overall architecture of proposed GapDNER model. $\otimes$ denotes vector concatenation of the last dimension. $\oplus$ represents the element-wise addition.
• Figure 4: Unified decoding process of continuous and discontinuous entities, including path searching and fragment connection. The end nodes are uniformly moved to the next token to match the start nodes of the next edge.
• Figure 5: Results comparisons on sentences containing discontinuous entities.