With the recent introduction of neural machine translation (MT) we are observing a major breakthrough in the quality of MT and subsequently new opportunities for reinventing the localization process with respect to turnaround times, workflows and means of achieving the highest quality output. However, the bigger question is how does an organization select an MT engine that is best suited for the content that is being translated to see the maximum productivity and quality gains. In this study, two companies have jointly evaluated several neural MT engines using sophisticated automation tools and human evaluation approaches and compared these engines’ performance both to each other and to traditional statistical MT engines. We’ll be sharing our findings and recommendations on the MT engine selection process, alongside with other considerations such as engineering aspects of engine performance and interconnectivity between platforms.
Takeaways: The attendees will learn about methodologies and processes of selecting a modern MT engine that is best suited for translation of enterprise global content within the context of an agile localization process with demanding deadlines, continuous workflow and high-quality standards. The attendees will also learn about performance of different neural MT engines with Asian and European languages evaluated through automated metrics and human evaluation processes.