Netflix uses machine translation (MT) as an assistive technology to localize a range of content types across language pairs. Our technology allows us to test MT suggestions at the segment level on the critical path and pivot between MT engines without impacting timelines. Test data sets are generated at translation request, and target data are captured at request completion. Analysis is automated. Empirical testing of multiple MT engines, language pairs and content types enables us to identify and exploit the best combination for greater accuracy, lower cost and faster turnaround.
Takeaways: Attendees will learn about how machine translation at Netflix is agnostic, empirical and effective.