The ever-evolving landscape of translation and localization demands continuous adaptation to emerging technologies, such as machine translation (MT). However, the quality of MT outputs can vary, necessitating human intervention in the form of post-editing (PE) to achieve accuracy and fluency.
Recognizing the significance of PE in the translation industry, there is a growing consensus in academia for its integration into translation curricula, providing students with specialized PE training and enhancing their skills to meet industry demands. To fully support the acquisition of PE skills, translation trainers who are familiar with MT and PE techniques and capable of evaluating post-edited texts and providing students with structured feedback are needed.
The Post-Edit Me! project, funded by UCLouvain’s Fonds de Développement Pédagogique for a two-year period (2021–2023), seeks to support machine translation post-editing (MTPE) training and learning in translator education, as Marie-Aude Lefer, Romane Bodart, Adam Obrusník, and Justine Piette from UCLouvain mentioned in their 2023 paper.
The project has two main objectives: (i) assisting lecturers in becoming familiar with PE, devising PE tasks, and assessing students’ work to boost their PE skills and (ii) developing innovative PE training practices.
Train the Translation Trainers
To achieve these goals, the project provides various forms of support for translation lecturers. It includes conferences and team meetings to introduce new pedagogical resources and promote the sharing of best practices. Additionally, on-demand individual coaching sessions guide lecturers in planning PE tasks for their specific courses, such as selecting source texts and MT engines, providing PE instructions, and evaluating students’ post-edited texts.
One critical aspect of PE training is ensuring the quality evaluation of students’ post-edited texts. For this purpose, the project introduces the MTPEAS annotation system (Machine Translation Post- Editing Annotation System), specifically designed for pedagogical purposes. The system is user-friendly and includes seven transparent categories — value-adding edit, successful edit, unnecessary edit, incomplete edit, error-introducing edit, unsuccessful edit, and missing edit — to facilitate lecturers in providing precise feedback.
Furthermore, by combining MTPEAS categories with tags from the Translation-oriented Annotation System (TAS) taxonomy, the system can identify specific errors in post-edited texts across various linguistic and contextual dimensions.
To streamline the evaluation process, the postedit.me app is utilized, which consists of a teacher interface and a student interface. The app automates the entire workflow of post-editing, from source-text selection to feedback sharing. It also provides valuable metrics like TER and expansion rates, as well as an automatic grade based on lecturers’ annotations of students’ post-edited texts. The postedit.me app is now being developed and beta-tested at UCLouvain and the open source app will be released in October 2023.
Benefit the Translation Students
In addition to supporting lecturers, the Post-Edit Me! project aims to benefit translation students directly preparing them for the future of the industry. Under the guidance of the project’s pedagogical assistant, lecturers have begun integrating post-editing tasks into domain- and language-specific translation courses.
As a result, students now have numerous opportunities to practice PE across different language pairs and domains. Thanks to the MTPEAS standardized taxonomy and the postedit.me app, students receive clear, relevant, and detailed feedback, fostering continuous improvement in their PE skills.