Workpackages

WP 1 Data and Adaptation

The aim of the work in this package is to ensure that we have the necessary corpora and other terminological and linguistic resources to build SMT systems, and to create adapted SMT systems for the public health domain. To do this we make maximal use of existing resources, collecting and cataloguing them in a way that is useful for other system builders. We apply state-of-the-art domain adaptation techniques to ensure that our systems are tailored to the translation of public health, and we supplement the parallel data with automatically mined multilingual terminology resources.

WP 2 Semantics

The aim of this work is to produce semantically accurate translations, focusing primarily on the preservation of semantic roles and the overall polarity (positive/negative) of the sentence but also on lexical choice by employing existing large-scale high quality dictionaries.

WP 3 Morphology

The aim of the work in this package is to ensure that we have the necessary capability to generate accurate morphology when building SMT systems, and that our morphological resources are adequately adapted to the public health domain so that high quality translation is ensured. One particular area of focus are errors of linguistic case, which can cause significant changes in the meaning of MT output that could be particularly dangerous in the domain of consumer healthcare information.

WP 4 Deployment

In this work package we deploy translation systems based on the work of WP1, WP2 and WP3, and integrate the translations created by these systems into the content management systems of COCHRANE and NHS 24.

WP 5 Evaluation

In this work package we evaluate our MT systems on an annual basis using the metrics we develop in this project, as well as standard metrics. We also participate in open evaluation campaigns, which allow us to benchmark our systems and to disseminate our research. Although we thoroughly evaluate the MT systems themselves, we are also committed to ensuring proper user evaluation of the MT deployed on live websites. To this end, both the NHS24 and COCHRANE will perform user acceptance testing of their automatic translation functionality and report web site traffic statistics and user acceptance ratings.  

WP 6 Dissemination and Exploitation 

The aim of this work package is to maximise the impact of the HimL project and to ensure that all possible angles for the exploitation of the projects results are explored.