Leveraging Large Language Models in Low-resourced Language NLP: A spaCy Implementation for Modern Tibetan

Large Language Models (LLMs) are transforming the possibilities for developing Natural Language Processing (NLP) tools for low-resource languages. While languages like Modern Tibetan have historically faced significant challenges in computational linguistics due to limited digital resources and annotated datasets, LLMs offer a promising solution. This paper describes how we leveraged Google's Gemini Pro 1.5 to generate training data for developing a basic spaCy language model for Modern Tibetan, focusing particularly on Part-of-Speech (POS) tagging. Combining traditional rule-based approaches with LLM-assisted data annotation, we demonstrate a novel methodology for creating NLP tools for languages with limited computational resources. Our findings contribute to the broader effort to enhance digital accessibility for low-resource languages while offering practical insights for similar projects in computational linguistics.