O‘ZBEK VA TURKMAN TILLARI UCHUN TRANSFORMER MODELLARINI QIYOSIY TAHLIL QILISH VA MOS MODEL TANLASH MEZONLARI
Rajabova S. T. & Raximboyev X. J.
Annotatsiya
Ushbu tadqiqotda turkiy tillar oilasiga mansub o‘zbek va turkman tillari uchun Tabiiy tilni qayta ishlash (NLP) modellarini tanlash prinsiplari tizimli tahlil qilingan. Agglutinativ morfologiya, boy derivatsiya tizimi va raqamli resurslarning cheklanganligi sharoitida zamonaviy NLP arxitekturalari (Encoder-only, Decoder-only, Encoder-Decoder) qiyosiy o‘rganilgan. Har bir model tipining lingvistik to‘siqlarni yengishdagi imkoniyatlari solishtirilib, o‘zbek va turkman tillaridagi ham tahliliy (NLU), ham generativ (NLG) vazifalar uchun Google kompaniyasining mT5 (Multilingual Text-to-Text Transfer Transformer) modeli eng maqbul va universal yechim ekanligi ilmiy jihatdan asoslab berilgan hamda tegishli adabiyotlar tahlili orqali isbotlangan.
Adabiyotlar
- Kuriyozov, E., Sanakulov, N., Matlatipov, S., & Gómez-Rodríguez, C. (2022). Building NLP Resources for Uzbek: Tokenization, POS Tagging, and Morphological Analysis using Deep Learning. Language Resources and Evaluation, 56(4), pp. 1125–1151.
- Niyazov, M., & Atayev, S. (2024). Cross-Lingual Knowledge Transfer for Low-Resource Oghuz Languages: Enhancing Turkmen NLP through Deep Transformers. Turkic Languages Journal, 28(1), pp. 84–99.
- Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., ... & Liu, P. J. (2020). Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer. Journal of Machine Learning Research (JMLR), 21(140), pp. 1–67.
- Toraman, C., Yilmaz, E. H., Şahinuç, F., & Ozcelik, O. (2023). Impact of Tokenization on Language Models for Morphologically Rich Languages: A Case Study on Turkish and Turkic Languages. Information Processing & Management, 60(3), 103294.
- Xue, L., Constant, N., Roberts, A., Kale, M., Al-Rfou, R., Siddhant, A., Barua, A., & Raffel, C. (2021). mT5: A Massively Multilingual Pre-trained Text-to-Text Transformer. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (NAACL-HLT 2021), pp. 483–498.