APE-QUEST: Integrating Quality Estimation (QE) and Automated Post-Editing (APE) in eTranslation

2020-04-07

The Automated Post-editing and Quality Estimation (APE-QUEST) project funded by the EC’s CEF Telecom programme provides a first quality gate and crowdsourcing workflow for the CEF eTranslation machine translation (MT) system. The quality gate is implemented by injecting quality estimation (QE) and automated post-editing (APE) into the existing translation workflow. QE and APE may both be applied to the output of eTranslation or to the crowdsourced translation. This process has two major benefits: (1) it improves MT quality with additional linguistic services and (2) it creates data aggregation opportunities by making translations and post-edits “locally owned” as data is generated and curated at the end user’s site. The Figure below illustrates the APE-QUEST architecture:

As shown in the Figure above, the workflow consists of three tiers: (1) MT output with acceptable quality flows directly to the end user or connected system, (2) moderate-quality MT is enhanced through APE, and (3) low-quality MT is sent to a workflow for manual PE (crowdsourcing).

APE-QUEST contains a portal-style front end (an adaptation of MateCat) and will be packaged in an easily deployable form for DSIs and public administrations in all EU Member States, Norway and Iceland. All components have been successfully integrated and the workflow is fully operational. The APE-QUEST project will be presented as part of the EAMT Conference and the APE system will be presented as part of LREC.