We present BabelDr, a web-enabled spoken-input phraselator for medical domains, which has been developed at Geneva University in a collaboration between a human language technology group and a group at the University hospital. The current production version of the system translates French into Arabic, using exclusively rule-based methods, and has performed credibly in simulated triaging tests with standardised patients. We also present an experimental version which combines largevocabulary recognition with the main rule-based recogniser; offline tests on unseen data suggest that the new architecture adds robustness while more than halving the 2-best semantic error rate. The experimental version translates from spoken English into spoken French and also two sign languages.