From Traditional Taggers to LLMs: A Comparative Study of POS Tagging for Medieval Romance Languages

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Science & Research — Research Methodology & Innovation, Mathematics & Computational Sciences, Digital Humanities · Depth: Expert, quick

Summary

A systematic empirical evaluation compares large language models (LLMs) against traditional rule-based and statistical taggers for Part-of-Speech (POS) tagging in Medieval Romance languages. Focusing on Medieval Occitan, Medieval Catalan, and Medieval French, the study addresses challenges like orthographic variation and morphological complexity. Experiments on historically grounded datasets demonstrate that LLM-based approaches consistently outperform traditional taggers. Significant improvements are observed with fine-tuning and multilingual training. Cross-lingual transfer learning particularly benefits under-resourced varieties, while targeted bilingual training can surpass broader multilingual configurations for specific target languages. The findings emphasize the importance of linguistic proximity and dataset characteristics in designing effective transfer strategies for historical Natural Language Processing.

Key takeaway

For Digital Humanities researchers or NLP engineers working with historical Romance languages, you should prioritize integrating LLM-based POS tagging pipelines. Leverage fine-tuning and cross-lingual transfer learning, especially for under-resourced varieties, as these methods consistently outperform traditional taggers. When designing transfer strategies, carefully consider linguistic proximity and the specific characteristics of your datasets to optimize performance, potentially favoring targeted bilingual training for specific languages.

Key insights

LLMs significantly improve POS tagging for Medieval Romance languages, especially with fine-tuning and cross-lingual transfer.

Principles

• Linguistic proximity is crucial for historical NLP transfer strategies.
• Dataset characteristics influence transfer strategy design.
• Targeted bilingual training can surpass broader multilingual setups.

Method

The study compares traditional taggers with LLMs using zero-shot, few-shot, monolingual fine-tuning, and cross-lingual transfer learning on historically grounded datasets for Medieval Romance languages.

In practice

• Deploy LLM-based POS tagging pipelines in digital humanities research.
• Utilize cross-lingual transfer for under-resourced historical language varieties.
• Consider targeted bilingual training for specific target languages.

Topics

• Part-of-Speech Tagging
• Large Language Models
• Medieval Romance Languages
• Digital Humanities
• Cross-lingual Transfer Learning
• Fine-tuning

Best for: AI Scientist, NLP Engineer, Research Scientist

Related on AIssential

• English is Not All You Need: Systematically Exploring the Role of Multilinguality in LLM Post-Training — Multilingual post-training significantly improves LLM performance across languages and tasks, outperforming English-only approaches.
• Scripts Through Time: A Survey of the Evolving Role of Transliteration in NLP — Transliteration bridges the "script barrier" in NLP, enhancing cross-lingual transfer and efficiency, especially for low-resource languages.
• Data Augmentation for Historical NER: A Systematic Comparison of Lexical and LLM-based Approaches — This paper systematically compares lexical and LLM-based data augmentation for Historical Named Entity Recognition.
• CzechDocs: A Multiway Parallel Dataset of Formatted Documents for Minority Languages in Czechia — A new multiway parallel dataset facilitates evaluating machine translation systems' ability to preserve document formatting, particularly for minority languages.
• A Larger Annotated Corpus of Portuguese Coreference — LLMs can annotate Portuguese coreference, but a new LLM-based system did not outperform existing rule-based methods.
• Opportunities and Challenges of Large Language Models for Low-Resource Languages in Humanities Research — LLMs offer transformative potential for low-resource language research in humanities, despite significant data and ethical challenges.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.