Efforts are underway to teach ChatGPT Kazakh. Its adapted transcription models already reach 98% accuracy. The results of a joint project between the «Qazaq Tili» International Society and OpenAI, aimed at advancing the Kazakh language in the digital environment, were presented in Astana. The work is being carried out under an agreement signed in Washington last year. The project’s text database now contains 14 billion tokens, while its audio dataset has reached 12,000 hours of recordings.
«We are training ChatGPT. The system is designed to serve both as a translation tool and as a tool for summarizing information related to human knowledge, as well as answering questions in Kazakh. At the same time, this tool can help other users who may be interested in our history, culture and traditions,» noted Rauan Kenzhekhanuly, President of the «Qazaq Tili» International Society.
The AI Evaluation Benchmark Suite assesses the proficiency of artificial intelligence systems in Kazakh, as well as their knowledge of Kazakh traditions and culture. Special tests have been created for this purpose. Also, Document AI and NLP tools have been developed. These technologies enable the digitization and recognition of archival texts written in Kazakh using the Arabic script or the Latin alphabet. Plans also include developing a tokenizer that takes into account the semantic structure of the language. Implementing these solutions requires significant computing resources.
«We are ready to allocate
1% of the resources of the Data Center Valley planned for construction in
Ekibastuz to the development of the Kazakh language. This work requires
significant computing power, so it should not be undertaken by scientists alone
– it is not just a research effort. Kazakhtelecom is also ready to play an
active role in this work and provide the necessary resources,» said
Bagdat Mussin, Chairman of the Board of Kazakhtelecom JSC.
Related Articles
