Post-Editing Tool with Dynamic Active Learning ‘PEDAL’ increases ROI on machine translation systems by reducing post-editing time and improving productivity with a more ‘enjoyable’ working environment for post-editors.
PEDAL combines cutting-edge technology for online retraining and active learning into a web-based Computer Aided Translation (CAT) tool. The tool enables machine translation systems to quickly and dynamically auto-correct repetitive errors, thereby ensuring that the system improves as early as possible. PEDAL identifies the optimal review path through a document for the post-editor, guiding the post-editing process to make industrial translation workflows more competitive. This can make post-editing work more enjoyable as the post-editor can instantly see their changes reshaping the behaviour of the machine translation system. The PEDAL application is computationally low-cost and language independent, and it can be easily integrated into translation workflows. During testing, PEDAL’s active learning approach decreased post-editing times by an average of roughly 10% for English/German and 4% for English/French.