Speakers
Description
Earlier multilingual dictionaries of sports terminology that include French and Serbian were largely based on simple translations from foreign sources, without systematic terminological analysis whereas more recent works do not include French. To be able to address this gap and compile a reference dictionary, we first need to identify the optimal methodology for the compilation of this type of dictionary entries. This paper presents the evaluation of two different possible approaches — corpus-based and AI-assisted generation. First, we selected ten polysemous French sports terms with multiple Serbian equivalents across different contexts and generated dictionary entries using both corpus analysis and AI models. The corpus used in the process is a multilingual parallel corpus ParCoLab containing aligned English, French, and Serbian texts from official sports rulebooks of 14 sports, providing a solid basis for terminological extraction, whereas the AI models evaluated were GPT-5.5, Sonnet 4.6 and Gemini 3.6. The generated entries were then evaluated by human experts for accuracy, consistency, and practical usefulness. The results show that while AI significantly accelerates the compilation process, corpus-based methods remain more reliable. Combining both approaches, thus, proves to be most effective: corpus ensures data quality, while AI supports efficiency. However, final validation by a human terminologist remains essential for refining meanings, standardizing terminology, and ensuring long-term usability.