What’s in a Bridge?: A Descriptive, Multi-Genre Analysis of the GUMBridge Corpus for Varieties of Bridging Anaphora

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Natural Language Processing · Depth: Expert, medium

Summary

Lauren Levine and Amir Zeldes conducted a descriptive corpus analysis of bridging anaphora across 16 English genres using the multi-genre GUMBridge corpus. Their investigation revealed that spoken genres contain fewer bridging instances compared to written ones. The study examined linguistic environments of bridging anaphora and their associative antecedents, considering categorical features like entity type and syntactic dependency relations, alongside numeric features such as mention length and salience. Key findings indicate that bridging anaphora are typically shorter and more definite, while their antecedents exhibit higher salience. The analysis also identified consistent genre-specific patterns in how numeric features of bridging environments vary. This research was presented at CODI-CRAC 2026 in San Diego, California.

Key takeaway

For NLP Engineers developing discourse analysis systems, understanding bridging anaphora characteristics is crucial. You should account for the finding that spoken genres have fewer bridging instances than written ones, potentially adjusting model training data or weighting. When designing anaphora resolution algorithms, prioritize identifying shorter, definite anaphora and consider the higher salience of their antecedents to improve accuracy, especially across diverse text types.

Key insights

The GUMBridge corpus analysis reveals bridging anaphora are shorter, more definite, and less frequent in spoken genres.

Principles

• Spoken genres exhibit less bridging anaphora.
• Bridging anaphora tend to be shorter.
• Bridging antecedents are often more salient.

Method

The study performed a descriptive corpus analysis, examining categorical features (e.g., entity type, POS) and numeric features (e.g., mention length, salience) of bridging anaphora and their antecedents across 16 genres.

In practice

• Prioritize written texts for bridging anaphora detection.
• Focus on shorter, definite anaphora in NLP tasks.

Topics

• Bridging Anaphora
• Corpus Linguistics
• GUMBridge Corpus
• Discourse Analysis
• Natural Language Processing
• Anaphora Resolution

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• A Cross-Genre Analysis of Discourse Relation Signaling in the GUM Corpus — The paper analyzes discourse relation signaling across diverse genres in the GUM Corpus.
• AfriSUD: A Dependency Treebank Collection for Evaluating Models on African Languages — Models exhibit a significant "syntax gap" in processing African languages, struggling with their structural diversity.
• CoDEl-BR: An Electoral Debate Corpus from Brazilian Municipal Elections — CoDEl-BR offers a high-quality corpus of Brazilian electoral debates for NLP research.
• TalkTag: Fine-Grained Morphosyntactic Error Annotation for Transcribed Speech — TalkTag automates fine-grained morphosyntactic error annotation in spoken transcripts using an LLM, even with scarce data.
• Characterizing Narrative Content in Web-scale LLM Pretraining Data — Narrative structure in web-scale LLM pretraining data is measurable and unevenly distributed, impacting model reasoning.
• Floating or Suggesting Ideas? A Large-Scale Contrastive Analysis of Metaphorical and Literal Verb-Object Constructions — Metaphorical and literal language differences are largely construction-specific, not universally consistent.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.