340320 VO Machine translation (2022W)

- This course will be taught in a hybrid format –

Goals:
Students will understand the evolution of the main machine translation (MT) paradigms (rule-based, statistical, and neural) and the extent to which each of them (or the tools used to enable them) is/are still applicable today for a range of language processing tasks. The analysis of the detailed process of building, evaluating and fine-tuning an MT system will focus on the neural paradigm.
Using examples from both production and experimental engines, students will be introduced to the variety of scenarios in which machine translation proves to be effective, as well as specifically how MT engines can be integrated in localization workflows and computer-assisted translation (CAT)/translation environment tools (TEnT). These scenarios will also cover a range of current challenges including, but not limited to, using MT for technical v creative texts, or text-to-text v speech-to-speech translation. The topics of and techniques for machine translation evaluation, as well as machine translation quality estimation will also be illustrated with examples.
However, this is not a module focusing just on the technology behind and around machine translation. The ethics of using machine translation will also be discussed, alongside bias in MT and challenges of identifying and correcting it. International standards which govern the use of machine translation and translation technologies will be introduced and discussed. Post-editing best practices will be discussed and productivity myths regarding the effectiveness of MT will be challenged. Cognitive ergonomics and alternative ways of working with MT besides the current post-edit-inside-CAT-tools will be discussed.
Content:
• Rule-based (RBMT), statistical (SMT), and neural machine translation (NMT): history and applications
• NMT architectures
• MT evaluation metrics
• MT quality estimation (QE) practices
• MT in professional workflows
• MT in more complex systems (e.g. speech-to-speech translation)
• Controlled language; pre-editing
• Post-editing machine translation (PEMT) standards and best practices
• Ethics of using MT and impact of MT on freelance linguists, the language services industry, natural languages, and the environment.
Didactic approach:
The content of this course is acquired by students in an interactive and blended way between home assignments studying the research literature, reflecting on concrete scenarios of applying machine translation systems to a variety of business problems, and the joint discussion and assessment of their applicability. The course is held in English with, whenever possible, simultaneous (but automatic, machine-generated) translation into German which, although not perfect, should still give students broad access to the contents of the course and so that they have the live experience of machine translation applied in the course itself.