kaz29Вестник Казахстанско-Британского технического университетаHerald of the Kazakh-British Technical University1998-66882959-8109Казахстанско-Британский Технический Университет10.55452/1998-6688-2026-23-3-233-242kaz29-3188Research ArticleКОМПЬЮТЕРНЫЕ НАУКИCOMPUTER SCIENCEМОРФОЛОГИЧЕСКАЯ ДИЗАМБИГУАЦИЯ КАЗАХСКОГО ЯЗЫКА С ИСПОЛЬЗОВАНИЕМ ТРАНСФОРМЕРНЫХ МОДЕЛЕЙMORPHOLOGICAL DISAMBIGUATION FOR THE KAZAKH LANGUAGE USING TRANSFORMER-BASED MODELShttps://orcid.org/0000-0003-2982-214XЭйпмЭ. Х.AitimA. К.

PhD, ассоциированный профессор

Алматы

PhD, associate professor

Almaty

a.aitim@iitu.edu.kzМеждународный университет информационных технологийКазахстанInternational Information Technology UniversityKazakhstan202625092026233233242Copyright © Эйпм Э.Х., 20262026Эйпм Э.Х.Aitim A.К.Данная работа распространяется под лицензией Creative Commons Attribution 4.0.This work is licensed under a Creative Commons Attribution 4.0 License.https://vestnik.kbtu.edu.kz/jour/article/view/3188

Морфологическая неоднозначность представляет собой значимую проблему для обработки естествен­ного языка в агглютинативных языках, поскольку одна и та же словоформа может включать множество грамматических категорий. Казахский язык характеризуется продуктивной суффиксацией, сингармонизмом и сложными морфофонологическими закономерностями, что существенно затрудняет автоматический мор­фологический анализ. В данной работе представлена трансформер-ориентированная методология морфо­логической дизамбигуации казахских текстов, направленная на выбор корректной морфологической интер­претации словоформы в контексте. Контекстная языковая модель, адаптированная для казахского языка, до­обучается для токен-уровневой морфологической разметки с использованием вручную верифицированного аннотированного корпуса новостных статей. Предложенный метод применяет механизмы самовнимания (self-attention) для захвата дальних контекстных зависимостей, которые сложно эффективно представить традиционными правилами или рекуррентными нейронными подходами. Экспериментальная оценка пока­зывает, что трансформерная модель достигает более высокой точности и Fl-меры по сравнению с правилами морфологическими анализаторами и базовыми моделями на основе BiLSTM. Результаты демонстрируют, что контекстуализированные эмбеддинги заметно улучшают разрешение морфологической неоднозначно­сти, особенно в случаях омонимичных суффиксов и редких грамматических структур. Полученные выводы подтверждают эффективность трансформерных архитектур для низкоресурсных агглютинативных языков и формируют практическую основу для интеграции морфологически чувствительных моделей в комплексные системы обработки казахского языка.

Morphological ambiguity constitutes a significant challenge for natural language processing in agglutinative languages, as a single word form might include many grammatical categories. The Kazakh language features productive suffixation, vowel harmony, and intricate morphophonological patterns, which considerably hinder automatic morphological analysis. This work presents a transformer-based methodology for morphological disambiguation in Kazakh texts, with the objective of identifying the appropriate morphological interpretation of word forms within context. A contextual language model tailored for Kazakh is refined for token-level morphological tagging utilizing a manually validated annotated corpus of news articles. The suggested method utilizes self-attention mechanisms to capture long-range contextual dependencies that are challenging to represent with conventional rule­based or recurrent neural techniques. The experimental assessment reveals that the transformer-based model attains superior accuracy and F1-score relative to rule-based morphological analyzers and BiLSTM-based benchmarks. The findings demonstrate that contextualized embeddings significantly enhance the resolution of morphological ambiguity, especially with homonymous suffixes and infrequent grammatical structures. The results validate the efficacy of transformer topologies for low-resource agglutinative languages and establish a feasible basis for incorporating morphology-aware models into comprehensive Kazakh natural language processing frameworks.

казахский языкморфологическая дизамбигуациятрансформерные моделиобработка естественного языкаагглютинативные языкиKazakh languagemorphological disambiguationtransformer-based modelsnatural language processingagglutinative languagesThis research was funded by the Grant No. AP32522233 “QNLPAI an open-source scientific system for intelligent processing of Kazakh-language texts” of Ministry of Science and Higher Education of the Republic of KazakhstanReferencesBach, M.P., Topalovic, A., Krstic, Z., and Ivec, A. Predictive maintenance in industry 4.0 for the SMEs: A decision support system case study using open-source software. Designs, 7, 98 (2023). https://doi.org/10.3390/designs7040098Bach, M.P., Topalovic, A., Krstic, Z., and Ivec, A. Predictive maintenance in industry 4.0 for the SMEs: A decision support system case study using open-source software. Designs, 7, 98 (2023). https://doi.org/10.3390/designs7040098Aitim, A., and Abdulla, M. Data Processing and Analysing Techniques in UX Research. Procedia Computer Science, 251, 591–596 (2024). https://doi.org/10.1016/j.procs.2024.11.154Aitim, A., and Abdulla, M. Data Processing and Analysing Techniques in UX Research. Procedia Computer Science, 251, 591–596 (2024). https://doi.org/10.1016/j.procs.2024.11.154Aitim, A., Sattarkhuzhayeva, D., and Khairullayeva, A. Development of a hybrid CNN-RNN model for enhanced recognition of dynamic gestures in Kazakh Sign Language. Eastern-European Journal of Enterprise Technologies, 2 (2 (134)), 58–67 (2025). https://doi.org/10.15587/1729-4061.2025.315834Aitim, A., Sattarkhuzhayeva, D., and Khairullayeva, A. Development of a hybrid CNN-RNN model for enhanced recognition of dynamic gestures in Kazakh Sign Language. Eastern-European Journal of Enterprise Technologies, 2 (2 (134)), 58–67 (2025). https://doi.org/10.15587/1729-4061.2025.315834Aitim, A. Building a high-quality annotated corpus for Kazakh NLP: a pipeline approach. Bulletin KazUTB, 4 (29) (2025). https://doi.org/10.58805/kazutb.v.4.29-1092Aitim, A. Building a high-quality annotated corpus for Kazakh NLP: a pipeline approach. Bulletin KazUTB, 4 (29) (2025). https://doi.org/10.58805/kazutb.v.4.29-1092Aitim, A., and Satybaldiyeva, R. A comparison of Kazakh language processing models for improving semantic search results. Eastern-European Journal of Enterprise Technologies, 1 (2 (133)), 66–75 (2025). https://doi.org/10.15587/1729-4061.2025.315954Aitim, A., and Satybaldiyeva, R. A comparison of Kazakh language processing models for improving semantic search results. Eastern-European Journal of Enterprise Technologies, 1 (2 (133)), 66–75 (2025). https://doi.org/10.15587/1729-4061.2025.315954Aitim, A. Developing methods for automatic processing systems of Kazakh language. KazATC Bulletin, 133 (4), 254–265 (2024). https://doi.org/10.52167/1609-1817-2024-133-4-254-265Aitim, A. Developing methods for automatic processing systems of Kazakh language. KazATC Bulletin, 133 (4), 254–265 (2024). https://doi.org/10.52167/1609-1817-2024-133-4-254-265QNLP – Full Kazakh NLP Suite GitHub repository. https://github.com/Aigerimhub/qnlpQNLP – Full Kazakh NLP Suite GitHub repository. https://github.com/Aigerimhub/qnlpAli, A., and Gravino, C. Improving software effort estimation using bio-inspired algorithms to select relevant features: an empirical study. Science of Computer Programming, 205, 102621 (2021). https://doi.org/10.1016/j.scico.2021.102621Ali, A., and Gravino, C. Improving software effort estimation using bio-inspired algorithms to select relevant features: an empirical study. Science of Computer Programming, 205, 102621 (2021). https://doi.org/10.1016/j.scico.2021.102621Singh, K., and Gupta, P. Explainable artificial intelligence for software effort estimation: a survey and future directions. Information and Software Technology, 140, 106748 (2021). https://doi.org/10.1016/j.infsof.2021.106748Singh, K., and Gupta, P. Explainable artificial intelligence for software effort estimation: a survey and future directions. Information and Software Technology, 140, 106748 (2021). https://doi.org/10.1016/j.infsof.2021.106748Khan, J.A., and Khan, S.U.R. Empirical investigation about the factors affecting the cost estimation in global software development context. IEEE Access, 9, 22274–22294 (2021). https://doi.org/10.1109/ACCESS.2021.3055858Khan, J.A., and Khan, S.U.R. Empirical investigation about the factors affecting the cost estimation in global software development context. IEEE Access, 9, 22274–22294 (2021). https://doi.org/10.1109/ACCESS.2021.3055858Srivastava, D.K., Sharma, A.K., and Choudhary, D. Software development effort estimation using machine learning techniques: multi-linear regression versus random forest. 2021 International Conference on Computing, Communication and Green Engineering (CCGE) (2021), pp. 1–5. https://doi.org/10.1109/CCGE50943.2021.9776394Srivastava, D.K., Sharma, A.K., and Choudhary, D. Software development effort estimation using machine learning techniques: multi-linear regression versus random forest. 2021 International Conference on Computing, Communication and Green Engineering (CCGE) (2021), pp. 1–5. https://doi.org/10.1109/CCGE50943.2021.9776394Alsaadi, M., and Saeedi, K. Agile effort estimation based on user stories: a systematic literature review. Artificial Intelligence Review, 55 (7), 5485–5516 (2022). https://doi.org/10.1007/s10462-021-10132-xAlsaadi, M., and Saeedi, K. Agile effort estimation based on user stories: a systematic literature review. Artificial Intelligence Review, 55 (7), 5485–5516 (2022). https://doi.org/10.1007/s10462-021-10132-xMatsubara, P.G.F. SEXTAMT: a systematic map to navigate the wide seas of factors affecting expert judgment software estimates. Journal of Systems and Software, 185, 111148 (2022). https://doi.org/10.1016/j.jss.2021.111148Matsubara, P.G.F. SEXTAMT: a systematic map to navigate the wide seas of factors affecting expert judgment software estimates. Journal of Systems and Software, 185, 111148 (2022). https://doi.org/10.1016/j.jss.2021.111148Fávero, E.M.D.B. SE3M: a model for software effort estimation using pre-trained embedding models. Information and Software Technology, 147, 106886 (2022). https://doi.org/10.1016/j.infsof.2022.106886Fávero, E.M.D.B. SE3M: a model for software effort estimation using pre-trained embedding models. Information and Software Technology, 147, 106886 (2022). https://doi.org/10.1016/j.infsof.2022.106886Li, X., Zhao, H., and Yu, M. Hybrid deep learning models for software cost prediction using CNN and LSTM. Journal of Systems and Software, 188, 111282 (2022). https://doi.org/10.1016/j.jss.2022.111282Li, X., Zhao, H., and Yu, M. Hybrid deep learning models for software cost prediction using CNN and LSTM. Journal of Systems and Software, 188, 111282 (2022). https://doi.org/10.1016/j.jss.2022.111282Rosa, C.C., and Jardine, D.A. Data-driven agile software cost estimation models for DHS and DoD. Journal of Systems and Software, 203, 111739 (2023). https://doi.org/10.1016/j.jss.2023.111739Rosa, C.C., and Jardine, D.A. Data-driven agile software cost estimation models for DHS and DoD. Journal of Systems and Software, 203, 111739 (2023). https://doi.org/10.1016/j.jss.2023.111739

The authors declare that there are no conflicts of interest present.