Instruction Tuning on Public Government and Cultural Data for Low-Resource Language: a Case Study in Kazakh

TL;DR

This paper presents a large-scale instruction-following dataset for Kazakh, enhancing low-resource language understanding through LLM fine-tuning and demonstrating improved performance in various tasks.

Contribution

Introduces and open-sources a 10,600-sample instruction dataset for Kazakh, employing LLM-assisted data generation and manual verification to improve language model performance.

Findings

Fine-tuning models on the dataset improves task performance.

LLM-assisted data generation is effective for low-resource languages.

High-quality dataset enhances understanding of governance topics.

Abstract

Instruction tuning in low-resource languages remains underexplored due to limited text data, particularly in government and cultural domains. To address this, we introduce and open-source a large-scale (10,600 samples) instruction-following (IFT) dataset, covering key institutional and cultural knowledge relevant to Kazakhstan. Our dataset enhances LLMs' understanding of procedural, legal, and structural governance topics. We employ LLM-assisted data generation, comparing open-weight and closed-weight models for dataset construction, and select GPT-4o as the backbone. Each entity of our dataset undergoes full manual verification to ensure high quality. We also show that fine-tuning Qwen, Falcon, and Gemma on our dataset leads to consistent performance improvements in both multiple-choice and generative tasks, demonstrating the potential of LLM-assisted instruction tuning for low-resource languages.