Lack of a shared language is a common communication situation in the globalizing world. Sometimes this can be mitigated by the use of machine translation technology, such as Google translate, but there are mission-critical tasks, like in health care, where one has to be sure about the correctness of the translation. In such situations, human interpreters are the best choice, but interpreters are scarce and in urgent situations they are not always available. This calls for improved and more reliable machine translation initiatives.

The project to be presented is developing a mobile translator for ambulance personnel use. The translator uses a verifiable and controllable machine translation technology, which is based on semantics, grammars, and professional terminology. The technology has been developed in the international open source project Grammatical Framework (GF) and tested in numerous research projects as well as commercial applications. This project is the first one to apply GF in a healthcare setting. The aim is to develop a platform for a range of health care applications, provided this pilot project for ambulance/emergency care is successful.

The translator works as a mobile app, in which the user can speak and write questions and other phrases, and get them translated to speech and text in other languages. The phrases cover the concepts used in the SBAR protocol (Situation-Bakgrund-Aktuellt tillstånd-Rekommendation) for ambulance use, as gathered from available documents and a questionnaire sent out to professionals at SU Ambulans. The SBAR protocol is also made available as a dynamic phrasebook, where the user can select appropriate phrases from menus. To help translate spontaneous speech and writing, the translator will also have a facility of suggesting nearest-matching phrases and ranking them by proximity to the verified standard phrases.

The current prototype covers around 400 concepts, from which millions of phrases can be built. It will work for 7 languages and enable translation between any two of them, although the primary use case is translation from Swedish to another language and translating simple answers from the other language to Swedish. GF has potential for extending the application to over 30 languages.