Language model
90.54%
+4.47 over Gemma 3 27B
- 01Airun LLM90.54
- 02Gemma 3 27B86.07
- 03DeepSeek-R1 32B74.99
- 04Qwen3 30B69.56
Answer accuracy, %. Tasks in Kyrgyz.
Five benchmarks, five first places.
Compared againstGemma 3 27BDeepSeek-R1 32BQwen3 30BKani TTSAitil.kgClaude Opus 4.6Aitil.kg TranslateGPT-5.4Grok 4.20
90.54%
+4.47 over Gemma 3 27B
Answer accuracy, %. Tasks in Kyrgyz.
6.39%
3.2× fewer errors than the closest competitor
WER — errors per 100 words. Cyrillic only.
52.9
+6.9 over Claude Opus 4.6
BLEU — closeness to the reference translation.
We study how AI understands Kyrgyz and the other languages of Central Asia.
VenuesMeLLMMRLLoResMTClimateNLParXiv
MeLLM 2026·ACL Anthology2026.mellm-1.5
Zarina Uvalieva, Bektemir Kumarbai Uulu, Adilet Metinov, Tynchtykbek Tashbaltaev, Nurtilek Alibekov
Abstract
Text normalization, the task of converting noisy, informal text into a standardized form - is a fundamental preprocessing step for many NLP applications. Despite the growing need for Kyrgyz language processing tools, to the best of our knowledge, no prior work has addressed automatic text normalization for Kyrgyz, a morphologically rich, low-resource Turkic language. In this paper, we present the first systematic study of Kyrgyz text normalization. We collect a dataset of 1.67 million noisy–clean text pairs sourced from YouTube comments, Instagram posts, and Telegram channels, where users frequently write without punctuation, capitalization, or standard spelling. Pairs were annotated with Gemini 3 Pro; the 1,000-example test set was fully verified by two native Kyrgyz speakers with adjudication, and a random subset of the training data was spot-checked, while the full 1.67M training set was not verified exhaustively. For continual pre-training, we additionally use a 538 MB Kyrgyz corpus compiled from news portals and books. We evaluate five systems: a rule-based baseline, zero-shot mT5, a fine-tuned mT5-small model, a continually pre-trained mT5-small followed by fine-tuning, and zero-shot Gemma 4. Our experiments show that fine-tuned mT5-small achieves a CER of 0.0796, outperforming the rule-based baseline (CER 0.2029), zero-shot mT5 (CER 0.9887), and zero-shot Gemma 4 (CER 0.1620), a roughly 32× larger model in a fine-tuned vs. zero-shot setting. Human evaluation by two native Kyrgyz speakers confirms these results, with fine-tuned mT5-small rated as correct in 99.8% of cases. We further analyze why continual pre-training with span corruption does not improve over direct fine-tuning, finding hallucination in 35/40 of the inspected failure cases (87.5%, 95% Wilson CI [74%, 95%]).
Proceedings of the 1st Workshop on Multilinguality in the Era of Large Language Models (MeLLM 2026), pages 52–62July 2026 ©2026 Association for Computational Linguistics52
The first systematic study, on 1.67M social media pairs: the neural model makes 2.5 times fewer errors than rules.
Roads, shopping, documents and laws: everything you need in the city.
ProAirun Pro subscription only
The whole country is connected to Airun
In the listBanks 2Government 3Education 8Business 9
Banks
Bakai Bank
Banks
Берекет Банк
Government
Жогорку Кенеш
Government
Министерство юстиции
Government
Мэрия Бишкека
Education
Университет Манас
Education
Университет Ала-Тоо
Education
КГМА
Education
КРСУ
Education
МУК
Education
AUCA
Education
Compass College
Education
Nomad College
Business
mTicket.kg
Business
Etnomedia
Business
1.1 studio
Business
OnClinic
Business
ArtWin
Business
Devolution
Business
Profile.kg
Business
Bakai Store
Business
Bakai Travel
Featured in
The Times of Central AsiaSputnik КыргызстанToday.kgEconomist.kgBanks.kgReporter.kgKaktus.mediaTHE TECHАкчабарРамблер/новости
For you: in the browser and on your phone. For your product: Airun models via API.