A Hybrid Transformer-Based System for Multilingual Translation with Domain-Specific Terminology Explanation

Abstract

Communication across languages becomes particularly challenging when specialized terminology is involved because literal translations often fail to convey the precise meanings essential in professional contexts. Current translation systems have problems with translating technical terms from specific domains like law, medicine, or finance, such that they generate translations that lack explanations of what technical terms actually mean. This creates significant problems in professional settings where misunderstanding specialized vocabulary can have serious consequences. This research developed a hybrid transformer-based system that addresses these limitations by translating specialized terminology across legal, medical, and financial domains while providing contextual explanations in the target language. The system used two transformer models: Sentence-BERT to detect special terms and MarianMT to translate. Data collection involved extracting 7,100 specialized terms with definitions from authoritative sources, alongside 70,000 parallel sentence pairs for each of three language pairs: English-Spanish, English-French, and English-German. The CRISP-DM framework was used for development from problem definition to deployment. The English-Spanish model achieved 59.26 BLEU, English-French achieved 38.76 BLEU, and English-German achieved 25.68 BLEU. Term extraction achieved 92.8-96.5% average accuracy across domains, with perfect performance on exact matches and 79-94% accuracy on misspelled or incomplete terms.