Health in my Language: health domain adapted machine translation as a means to tackle the resource issues for translation

Tags: Oral
Ried J1, For the HimL consortium N2
1Cochrane Central Executive, Translations Co-ordinator, Germany, 2University of Edinburgh, Charles University Prague, Ludwig Maximilian University of Munich, NHS 24, Cochrane, and Lingea, NA

Background: To an ever-increasing extent, web-based services are providing a frontline for healthcare information in Europe. They help citizens find answers to their questions, and they help them understand and find the available local services. However, due to the number of languages spoken in Europe, and the mobility of its population, there is a high demand for these services to be available in many languages. In order to satisfy this demand, we need to rely on automatic translation, as it is infeasible and unaffordable to translate manually into all languages requested. Health in my Language (HimL) is an EU-funded, three-year project, aiming at addressing this need. Health information produced by Cochrane and NHS24 (health information and self care advice for the people of Scotland) will serve as the test case for this project.

Objectives: To create usable, reliable, fully automatic translation of public health information, initially testing with translation from English into Czech, Polish, Romanian and German.

Methods: We will use recent advances in machine translation (MT) to create a system for the automatic translation of public health information, with a focus on preservation of meaning. We will include recent work on domain adaptation, translation into morphologically rich languages, terminology management, and semantically enhanced MT to build a reliable system for the health domain. We will iterate cycles of incorporating improvements into the MT systems annually, with careful evaluation and user acceptance testing. We will develop metrics to evaluate the quality and measure the impact on post-editing of the obtained results in each cycle. Cochrane and NHS24 content will also be translated in each cycle and published on their websites, and users invited to participate in a survey to give feedback on the quality and usability of the translations. Web usage statistics will be analysed.

Results and Conclusions: The first version of the MT system is to be deployed in Sept 2015. Preliminary results and conclusions will be presented at the Colloquium as available. The focus will be on process and the Cochrane use case, rather than technological details.