The Elephant in the Coreference Room: Resolving Coreference in Full-Length French Fiction Works
Antoine Bourgois, Thierry Poibeau
TL;DR
This work tackles coreference resolution in long-form literature by releasing Long-LitBank-fr, an annotated corpus of three full-length French novels (~$285{,}176$ tokens) focused on character mentions. It introduces a modular, resource-efficient CR pipeline that scales to long documents, integrates a document-wide propagation strategy for proper nouns, and demonstrates performance gains relative to a French baseline while revealing the impact of document length on CR. The study provides a detailed annotation protocol with high inter-annotator agreement, merges the new resource with LitBank-fr to form a large Long-LitBank-fr dataset, and validates the approach through a gender-inference case study, showing substantial recall improvements over rule-based methods. Together, these contributions advance in-domain CR for literary texts and enable robust downstream analyses in literary studies and NLP.
Abstract
While coreference resolution is attracting more interest than ever from computational literature researchers, representative datasets of fully annotated long documents remain surprisingly scarce. In this paper, we introduce a new annotated corpus of three full-length French novels, totaling over 285,000 tokens. Unlike previous datasets focused on shorter texts, our corpus addresses the challenges posed by long, complex literary works, enabling evaluation of coreference models in the context of long reference chains. We present a modular coreference resolution pipeline that allows for fine-grained error analysis. We show that our approach is competitive and scales effectively to long documents. Finally, we demonstrate its usefulness to infer the gender of fictional characters, showcasing its relevance for both literary analysis and downstream NLP tasks.