Machine Translation Post-Editing

Singapore University of Social Sciences

Machine Translation Post-Editing (TNT505)

Synopsis

Machine translation post-editing (MTPE) is a special and popular use case for humans to work with an automated translation system. This course discusses fundamentals of modern machine translation (MT) technology and MT post-editing strategies from a human perspective. It first introduces the current situations of MT deployment, including user types, opportunities and challenges. Then it examines the technical aspects of MT, including the mathematical frameworks for three MT paradigms (rule-based, statistical and neural MT), automated MT evaluation metrics, quality estimation and automatic post-ending (APE). Throughout this course, students will build a connection between their existing experience with language and knowledge of MT, define a problem that is based on real-life MTPE scenarios, apply relevant methodologies covered in the course and design a study to address this problem. The coursework will be project-based so that students are able to prepare themselves for real-life MTPE implementation.

Level: 5
Credit Units: 5
Presentation Pattern: Every January
E-Learning: - Learning is done MAINLY online using interactive study materials in Canvas. Students receive guidance and support from online instructors via discussion forums and emails. This is supplemented with SOME face-to-face sessions. If the course has an exam component, this will be administered on-campus. To be confirmed

Topics

Machine Translation (MT) in the language industry
Introduction to the MT
MT Paradigms and MT tool practice
Neural MT
Machine Translation Quality Estimation (QE)
Introduction to MT post-editing (MTPE)
MT Editing Strategies
MT Customisation in specific domains: Healthcare and life science
MT Customisation in specific domains: Business and finance
MT Customisation in specific domains: Law
Human-in-the-loop MT workflow

Learning Outcome

Compare different types of MT engines
Design workflow of human-aided machine translation
Prioritise different tasks in MT post-editing
Formulate an approach to the use of MT
Predict MT output after pre-editing
Appraise MT output
Improve the efficiency of the current design and workflow