A Study of Temporal Fusion Strategies for Named Entity Recognition in Historical Texts

2026-06-26

AI Summary

This work addresses the challenge of named entity recognition in historical texts, where entity forms and salience shift over time, posing difficulties for contemporary language models to effectively capture diachronic contextual cues. To tackle this, the authors propose a lightweight temporal metadata integration approach that systematically explores early and late fusion mechanisms of absolute and relative time representations within a Transformer architecture. Time-aware modeling is achieved through techniques such as cross-attention, adapter modules, and feature concatenation. Experimental results on French and German historical datasets demonstrate that the late fusion strategy significantly enhances named entity recognition performance in cross-temporal settings, exhibiting superior robustness and temporal generalization—particularly on noisy early-period data.

Abstract

Temporal variation poses a unique challenge for named entity recognition (NER) in historical texts, where entities drift in surface form and salience across time. While language models (LMs) have made progress in various NLP tasks, their ability to reason about temporality, especially in diachronic contexts, remains limited or at least, questionable. In this paper, we systematically study how temporal metadata can be structurally embedded into NER models using a range of lightweight fusion strategies. We experiment with both absolute and relative temporal representations, injected into Transformer-based architectures via early or late fusion mechanisms such as cross-attention, adapters, and concatenation. Our evaluations on French and German historical datasets reveal that late fusion strategies yield more robust and temporally generalisable performance, particularly in early and noisy periods.

Problem

• Named Entity Recognition
• Historical Texts
• Temporal Variation
• Diachronic Contexts
• Entity Drift

Innovation

• temporal fusion
• named entity recognition
• historical texts
• late fusion
• diachronic NLP