Improving Norwegian Translation of Bicycle Terminology Using Custom Named-Entity Recognition and Neural Machine Translation

The Norwegian business-to-business (B2B) market for bicycles consists mainly of international brands, such as Shimano, Trek, Cannondale, and Specialized. product descriptions these brands are usually in English need local translation. However, include bicycle-specific terminologies that challenging online translators, Google. For this reason, companies outsource translation or translate manually, which is cumbersome. In light the B2B bicycle industry, paper explores transfer learning to improve machine terminology from Norwegian, including generic text. Firstly, we trained a custom Named-Entity Recognition (NER) model identify cycling-specific then adapted MarianMT neural process. Due lack publicly available bicycle-terminology-related datasets train proposed models, created our dataset by collecting corpus cycling-related texts. We evaluated performance compared its with Google Translate. Our outperformed Translate on test set, SacreBleu score 45.099 against 36.615 average. also web application where user can input text related terminologies, it will return detected words addition