Large language models improve linking of historical names with multiple comparisons
Beyond Single-Negative Preference: Multi-Negative DPO for LLM-Centric Historical Entity Linking
Computation and LanguageArtificial Intelligence
Summary
Linking names in old newspapers to the right historical people is tricky because names can be unclear or misspelled. This paper improves the way large language models learn by comparing the correct person not just against one wrong choice, but against many wrong options at once. This method helps the model do better, especially with tricky names and errors in old texts. The authors also found that finding good candidate names to compare remains a big challenge.
What this means in practice
- •For digital archives teams: Improve automatic linking of historical person names in digitized newspapers, reducing confusion from ambiguous or misspelled names.
- •For information retrieval engineers: Enhance large language model training by incorporating multiple wrong answer comparisons, improving entity linking in noisy text data.
Authors
Tien Nam Nguyen, Emanuela Boros, Ahmed Hamdi, Adam Jatowt, Mickaël Coustaty, Antoine Doucet
Abstract
Large language models (LLMs) have recently shown promise for historical entity linking, but preference optimization for this task is often formulated with only one negative candidate per training instance. This discards information from the remaining candidates retrieved for the same mention. We introduce multi-negative direct preference optimisation (MDPO), a reference-based pairwise objective that compares the correct entity with all valid rejected candidates associated with each mention. MDPO preserves the Bradley-Terry formulation of DPO while exploiting the complete candidate set through masked, length-normalised sequence scores. We evaluate MDPO on hipe-2020 and newseye, covering French, German, English, Swedish, and Finnish historical newspaper text. Experiments show that MDPO improves over supervised fine-tuning and single-negative DPO, with particularly strong gains for NIL mentions, semantic ambiguity, OCR noise, and historically difficult names. Further analyses disentangle candidate-generation and selection errors, showing that candidate retrieval remains a key bottleneck for end-to-end entity linking. These results demonstrate that incorporating all within-instance negative candidates is a simple and effective improvement for LLM-based historical entity linking.