Closing the Gap at CRAC 2026: Two-Stage Adaptation for LLM-Based Multilingual Coreference Resolution

Antoine Bourgois, Olga Seminck, Thierry Poibeau · 2026

Read Closing the Gap at CRAC 2026: Two-Stage Adaptation for LLM-Based Multilingual Coreference Resolution by Antoine Bourgois, Olga Seminck, Thierry Poibeau (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-16984.txt.

Description

Bourgois, Seminck and Poibeau report their winning submission to the language-model track of the 2026 CRAC shared task on coreference resolution. Their system, built on Gemma-3-27b and fine-tuned in two stages, scored 74.32 CoNLL F1 on the official test set, ranking first in its track and third overall. A multilingual base adapter comes first, then dataset-specific adapters, with mention spans marked by headword in an XML-style format with local reindexing and documents annotated iteratively. The design held up across languages, document lengths, and annotation guidelines.

Catalog record

CREATORAntoine Bourgois, Olga Seminck, Thierry Poibeau
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE46181
LICENSEhttp://creativecommons.org/licenses/by-sa/4.0/
RIGHTSCC BY SA
SOURCEarXiv

Source: https://arxiv.org/abs/2605.16984

Browse the full library · About the Tower of Babel