This paper presents a method to automatically translate a large class of terms in the biomedical domain from one language to another; it is evaluated on translations between French and English. It relies on a machine-learning technique that infers transducers from examples of bilingual word pairs; no additional resource or knowledge is needed. Then, these transducers, making the most of the high regularity of translation discovered in the examples, can be used to translate unseen French terms into English or vice versa. We report evaluations that show that this technique achieves high precision, reaching up to 85% of correct translations for both French to English and English to French tasks.