CORPUS AND AI TECHNOLOGIES FOR TERMINOLOGY MANAGEMENT IN LOCALISATION PROJECTS

The article explores terminology management for localization projects, with a special focus on extraction technologies, emphasizing corpus-based and AI-driven approaches. Given the complexity of localization, the automation of terminology management plays a crucial role in ensuring high-quality linguistic services. The study examines various terminology extraction strategies applied in Ukraine's language services market. The research objectives include defining the concept of «terminology extraction» systematizing principles of automated extraction tools, and conducting an experimental extraction process using SketchEngine, OpenAir (Plus), Fodina TermCatch. To assess the effectiveness of these tools, the authors built a specialized corpus, comprising 29,374 words and 36,912 tokens. The extracted terminology was analyzed based on key criteria: stability, precision, indivisibility, and definitional clarity. SketchEngine tool employs statistical keyword analysis, extracting 400 term candidates. However, only 25% (123) were deemed valid terminological units due to issues with phrase segmentation. Utilizing AI-driven models, OpenAir (Plus) tool extracted 81 term candidates, with 38% (31 terms) meeting the established criteria. However, it tended to overgenerate paraphrased variants. Fodina TermCatch tool extracted 1,460 candidates, but only 15.93% (233 terms) were valid. It identified redundant terms variations, which could be useful for refining source content. The study underscores the hybrid nature of terminology extraction today, combining statistical, AI-based, and hybrid approaches. While statistical methods provide greater precision, they may miss rare terms, whereas AI-driven tools offer high recall but require filtering. In conclusion, automated terminology extraction enhances efficiency in localization. However, manual validation remains essential for ensuring accuracy. The research confirms that combining multiple extraction tools yields the best results, making terminology extraction – an essential part of terminology management – a crucial step in localization project management.