Design of an integrated training data generation pipeline for machine translation for less-resourced languages
Системи обробки інформації · 30.06.2026 · (2(185)) · 181-192
DOI: 10.30748/soi.2026.185.21 ↗Kurz gefasst
Fachartikel in der Zeitschrift Системи обробки інформації, online erschienen am 30.06.2026. Verfasst von D. Lysenko und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
This paper describes a five-stage workflow designed for automated parallel corpora creation for direct neural machine translation systems between low-resource languages. The solution’s architecture combines multilingual sentence representation information extraction, large language model synthetic generation, and unsupervised machine translation, alongside iterative scaling via self-learning mechanisms. Quality control is maintained through neural semantic proximity assessment and hierarchical clustering tools. The proposed toolkit can generate the training datasets needed for historical text and extinct language analysis.
Zitieren
APA
Lysenko, D., Kobylin, O., Putiatina, O., & Lysenko, O. (2026). Design of an integrated training data generation pipeline for machine translation for less-resourced languages. Системи обробки інформації, 181-192. https://doi.org/10.30748/soi.2026.185.21
Vancouver
Lysenko D, Kobylin O, Putiatina O, Lysenko O. Design of an integrated training data generation pipeline for machine translation for less-resourced languages. Системи обробки інформації. 2026:181-192. doi:10.30748/soi.2026.185.21
BibTeX
@article{Lysenko2026,
author = {Lysenko, D. and Kobylin, O. and Putiatina, O. and Lysenko, O.},
title = {Design of an integrated training data generation pipeline for machine translation for less-resourced languages},
journal = {Системи обробки інформації},
year = {2026},
number = {2(185)},
pages = {181--192},
doi = {10.30748/soi.2026.185.21}
}
EUVIMED