Leaders: Heidi Depraetere (Cross Language), Hannah Grap (Language Weaver), Marie-Laure Poëte (Cross Language), Hubert Schlarb (Siemens), Jaap van der Meer (TAUS), Kirti Vashee (Asia Online)
Synopsis: Following 50 years of research and development, machine translation (MT) is now arriving at the workspace of localization and customer support professionals. New promises in translation automation are made on the basis of statistical approaches. Translation memories, accumulated over more than a decade of localization activities, represent a new value as training data for MT. Statistical machine translation (SMT) is making serious inroads in everyday applications such as search, multilingual support and localization. The list of providers includes big names such as Microsoft, IBM and Google but also dedicated technology providers such as Language Weaver and Asia Online.
In this full-day workshop we compare theory and practice. Are the promises of SMT true and applicable to the localization industry? Involving developers and early adopters of the new technologies, we will debate many of the relevant questions and as many answers as possible.
- What are the basic working principles of SMT?
- How much data do we need to train SMT? Is it true, the more data the better? How do we ever get enough clean and trustworthy data?
- How do we measure the quality and what is acceptable to the end-user?
- What is the best practice in post-editing SMT output? How is post-editing SMT different from post-editing output from rule-based engines?
- Can the SMT system learn from corrections made during post-editing?
- How do we integrate SMT in a typical localization workflow?
- How do we customize SMT and improve quality on an ongoing basis?
- What is the cost of deploying SMT?
The format of the workshop is a mix of presentations, group discussions and hands-on exercises.
9.00 Translation automation in a market perspective
• Changing market requirements
• Different applications of MT: Jaap van der Meer
9.30 Evaluating quality of MT – Different approaches: Heidi Depraetere
10.00 SMT in a historic perspectives: Marie Laure Poëte
10.30 Coffee Break
11.00 Fundamentals of SMT. How it works, how it’s different, how high quality translations are generated and why data is important.: Hannah Grap
11.30 SMT and developing a continuous improvement cycle: Kirti Vashee
12.00 Introduction to the evaluation benchmarking methodology: Marie Laure Poëte
12.15 Evaluating MT – Hands-on exercise using an industry-standard evaluation metrics participants will score the quality of machine translated texts: Participants
12.45 Lunch break
13.30 User cases of SMT: Simens, Hubert Schlarb
14.30 Report on the scores from the evaluation: Heidi Depraetere
15.30 Discussion on implementation scenarios and deployment models
Speakers: Jaap van der Meer (TAUS), Marie Laure Poëte (Cross Language), Heidi Depraetere (Cross Language), Hannah Grap (Language Weaver)