CHALLENGES AND OPPORTUNITIES FOR LOW-RESOURCE TURKIC LANGUAGES IN THE ERA OF LARGE LANGUAGE MODELS
Abstract
Large Language Models (LLMs) have become the foundation of modern Artificial Intelligence applications, enabling significant progress in text generation, machine translation, question answering, and conversational systems. However, their performance remains uneven across languages due to substantial differences in the availability of digital resources. While high-resource languages benefit from extensive web content and large-scale linguistic datasets, many Turkic languages continue to experience limited representation in multilingual AI systems. This imbalance reduces the quality of AI-generated content and restricts access to advanced language technologies for millions of native speakers. This paper examines the major challenges affecting LLM performance in low-resource Turkic languages, including digital content scarcity, insufficient training datasets, agglutinative linguistic structures, and limited evaluation benchmarks. Recent multilingual AI initiatives are reviewed to identify current progress and remaining limitations. Based on existing research, the study proposes a conceptual framework emphasizing community-driven content creation, open linguistic resources, multilingual benchmarks, and collaborative dataset development. The findings suggest that sustainable improvement in LLM performance depends not only on larger models but also on strengthening the digital ecosystem of low-resource languages through continuous content and resource development.References
[1] OpenAI. GPT-4 Technical Report. arXiv:2303.08774, 2023.
[2] S. Tukenov et al. SozKZ: Training Efficient Small Language Models for Kazakh
from Scratch. arXiv:2603.20854, 2026.
[3] X. Xue et al. mT5: A Massively Multilingual Pre-trained Text-to-Text Transformer.
Proceedings of NAACL, 2021.
[4] S. Bender et al. On the Opportunities and Risks of Foundation Models.
arXiv:2108.07258, 2021.
[5] OpenAI. GPT-4o System Card. OpenAI, 2024.
[6] M. Joshi et al. State and Challenges of NLP for Low-Resource Languages.
Communications of the ACM, 2024.
[7] Wikipedia contributors. Turkic Languages. Available:
https://en.wikipedia.org/wiki/Turkic_languages
[8] Ethnologue. Languages of the World. SIL International, 2025.
[9] W3Techs. Usage Statistics and Market Share of Content Languages for Websites.
Available: https://w3techs.com/technologies/overview/content_language
[10] X. Song et al. Tokenization Matters: Challenges for Morphologically Rich
Languages. ACL Findings, 2023.
[11] Meta AI. No Language Left Behind: Scaling Human-Centered Machine
Translation. Nature, 2022.
[12] Hugging Face. BLOOM: A 176B-Parameter Open-Access Multilingual Language
Model. 2022.
[13] Meta AI. FLORES-200: Benchmarking Machine Translation Across 200
Languages. 2022.
[14] Mozilla Foundation. Common Voice Dataset. Available:
https://commonvoice.mozilla.org
[15] A. Conneau et al. Unsupervised Cross-lingual Representation Learning at Scale.
ACL, 2020.
[16] H. Nigatu, J. Canny, and S. Chasins. Low-Resourced Languages and Online
Knowledge Repositories: A Need-Finding Study. arXiv:2405.16669, 2024.