CorefInst: Leveraging LLMs for Multilingual Coreference Resolution

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Arslan, Tuğba Pamay, Erol, Emircan, Eryiğit, Gülşen
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866916960335298560
author Arslan, Tuğba Pamay
Erol, Emircan
Eryiğit, Gülşen
author_facet Arslan, Tuğba Pamay
Erol, Emircan
Eryiğit, Gülşen
contents Coreference Resolution (CR) is a crucial yet challenging task in natural language understanding, often constrained by task-specific architectures and encoder-based language models that demand extensive training and lack adaptability. This study introduces the first multilingual CR methodology which leverages decoder-only LLMs to handle both overt and zero mentions. The article explores how to model the CR task for LLMs via five different instruction sets using a controlled inference method. The approach is evaluated across three LLMs; Llama 3.1, Gemma 2, and Mistral 0.3. The results indicate that LLMs, when instruction-tuned with a suitable instruction set, can surpass state-of-the-art task-specific architectures. Specifically, our best model, a fully fine-tuned Llama 3.1 for multilingual CR, outperforms the leading multilingual CR model (i.e., Corpipe 24 single stage variant) by 2 pp on average across all languages in the CorefUD v1.2 dataset collection.
format Preprint
id arxiv_https___arxiv_org_abs_2509_17505
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CorefInst: Leveraging LLMs for Multilingual Coreference Resolution
Arslan, Tuğba Pamay
Erol, Emircan
Eryiğit, Gülşen
Computation and Language
Artificial Intelligence
Coreference Resolution (CR) is a crucial yet challenging task in natural language understanding, often constrained by task-specific architectures and encoder-based language models that demand extensive training and lack adaptability. This study introduces the first multilingual CR methodology which leverages decoder-only LLMs to handle both overt and zero mentions. The article explores how to model the CR task for LLMs via five different instruction sets using a controlled inference method. The approach is evaluated across three LLMs; Llama 3.1, Gemma 2, and Mistral 0.3. The results indicate that LLMs, when instruction-tuned with a suitable instruction set, can surpass state-of-the-art task-specific architectures. Specifically, our best model, a fully fine-tuned Llama 3.1 for multilingual CR, outperforms the leading multilingual CR model (i.e., Corpipe 24 single stage variant) by 2 pp on average across all languages in the CorefUD v1.2 dataset collection.
title CorefInst: Leveraging LLMs for Multilingual Coreference Resolution
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2509.17505