Vuyko Mistral: Adapting LLMs for Low-Resource Dialectal Translation

2025-06-09

AI Summary

This study addresses the critical challenge of low-resource, morphologically complex Hutsul dialect—a regional variety of Ukrainian. We present the first Hutsul–Standard Ukrainian parallel corpus and lexicon. To overcome data scarcity, we propose a RAG-enhanced synthetic data generation pipeline and fine-tune an open-source 7B LLM using LoRA. We introduce the first LLM adaptation framework specifically designed for low-resource dialects and establish an unsupervised, multi-metric evaluation suite integrating BLEU, chrF++, TER, and GPT-4o-based discriminative assessment. Experimental results demonstrate that our fine-tuned model significantly outperforms GPT-4o in zero-shot translation across all metrics. All resources—including the corpus, dictionary, trained models, and code—are publicly released, constituting foundational infrastructure and a methodological blueprint for low-resource dialect NLP research.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

In this paper we introduce the first effort to adapt large language models (LLMs) to the Ukrainian dialect (in our case Hutsul), a low-resource and morphologically complex dialect spoken in the Carpathian Highlands. We created a parallel corpus of 9852 dialect-to-standard Ukrainian sentence pairs and a dictionary of 7320 dialectal word mappings. We also addressed data shortage by proposing an advanced Retrieval-Augmented Generation (RAG) pipeline to generate synthetic parallel translation pairs, expanding the corpus with 52142 examples. We have fine-tuned multiple open-source LLMs using LoRA and evaluated them on a standard-to-dialect translation task, also comparing with few-shot GPT-4o translation. In the absence of human annotators, we adopt a multi-metric evaluation strategy combining BLEU, chrF++, TER, and LLM-based judgment (GPT-4o). The results show that even small(7B) finetuned models outperform zero-shot baselines such as GPT-4o across both automatic and LLM-evaluated metrics. All data, models, and code are publicly released at: https://github.com/woters/vuyko-hutsul

Problem

• Adapting LLMs for low-resource Ukrainian dialect translation
• Addressing data shortage with synthetic parallel corpus generation
• Evaluating fine-tuned models against zero-shot baselines

Innovation

• Created parallel corpus for Ukrainian dialect translation
• Used Retrieval-Augmented Generation for synthetic data
• Fine-tuned LLMs with LoRA for dialect translation