Automation will be a building block of the modern translation pipeline and automating quality estimation is key to this. We’ve focused our attention on identifying high-quality translation output that can be produced automatically. We have experimented with several methods and focused on two approaches — identifying non-translatable segments and identifying high-quality machine translation for which post-editing is not needed. This effectively reduces the number of words that need to go through the human translation workflow. We would like to share the results along with the performance of these approaches in specific language pairs, domains and so on.
Takeaways: Attendees will learn how automated quality estimation can be incorporated and leveraged in the localization workflow and the benefits it can bring.