Under-resourced studies of under-resourced languages: lemmatization and POS-tagging with LLM annotators for historical Armenian, Georgian, Greek and Syriac
Résumé
Low-resource languages pose persistent challenges for Natural Language Processing tasks such as lemmatization and part-of-speech (POS) tagging. This paper investigates the capacity of recent large language models (LLMs), including GPT-4 variants and open-weight Mistral models, to address these tasks in fewshot and zero-shot settings for four historically and linguistically diverse under-resourced languages: Greek, Armenian, Georgian, and Syriac. Using a novel benchmark comprising aligned training and out-of-domain test corpora, we evaluate the performance of foundation models across lemmatization and POStagging, and compare them with PIE, a taskspecific RNN baseline. Our results show that LLMs, even without fine-tuning, can achieve competitive or superior performance in POStagging and lemmatization for most languages even with very few shots, but still struggle more for languages with complex morphologies and non-Latin scripts.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |