SmartMATE : Online Self-Serve Access to State-ofthe-Art SMT

Access to good quality Machine Translation (MT) has never been as easy as it is today. Portals such as Google Translate and Bing Translator facilitate huge amounts of translation requests on a daily basis, for an ever increasing spectrum of language pairs. People are finding many uses for the raw MT output provided by these, and other, freely available engines on the web, including gisting, assimilation, first drafts of translations for dissemination, etc. However, each of these systems is a ’one-size-fits-all’ solution, where no customization is available to the user. One alternative is to purchase a system, which may be overly expensive, or sub-optimal for the type of documentation required to be translated. Another alternative is to install a freely available system such as Moses, but this may prove unduly onerous for the naı̈ve user. In this paper, we present a portal which facilitates selfserve MT using state-of-the-art statistical MT (SMT). This is currently free for anyone to access and personalise their system by uploading their own Translation Memory(TM) and glossaries. By means of a simple key-press, optimal training, development and test data are created on-thefly, which are then used for an automatic system build, with the results published in very acceptable amounts of time, together with automatic evaluation scores. According to user trials, this – together with builtin TM and online editing functionality – is a very exciting development, with the potential to vastly expand the user-base for self-serve SMT on a global basis.