The industrial process for quality machine translation

Abstract

Machine translation (MT) is not a tool. Machine translation is an industrial process. Selecting the right MT engine – rules-based (RBMT), statistical (SMT) or hybrid – is just one part of a process that, if correctly managed, is capable of lowering translation costs, increasing productivity and even improving quality and consistency. To reach this goal, the MT process must pass through consultation, content, customisation, piloting, processing, post-editing, metrics and maintenance. This article looks at the first three stages in the MT process – consultation, content and customisation – and how the virtuous circle of post-editing feedback supports quality MT output. This is important because if the MT process is badly managed, it is inevitably the post-editor who pays the price. A system that is based on post-editors cleaning up bad MT is just not sustainable. With quality as the goal, the question is not so much what engine to choose but what engine and what process will give the best results. To determine what activities are most effective in achieving MT quality, a LexWorks study showed that a well-trained engine with an ongoing cycle of improvements from post-editing feedback is essential for MT quality.