Elaine O’Curran, Alex Yanishevsky and Olga Beregovaya from Welocalize’s Language Tools and Automation Team presented at the 11th Biennial Conference of the Association of Machine Translation in the Americas (AMTA-2014) in Vancouver last month. Elaine O’Curran, Welocalize Training Manager and presenter at AMTA, provides her highlights and insights from the conference.
Last week, several members of the Welocalize Language Tools and Translation Automation Team went to Vancouver, Canada and participated in a number of presentations and panel discussions related to machine translation and post-editing at this year’s AMTA conference. We enjoyed four full days of presentations, demos and workshops in a very collaborative environment that brought together developers, researchers and translators from various backgrounds.
Here are my top three highlights from this year’s conference:
- MT for User-Generated Content (UGC). This was the topic of several presentations, including the opening keynote by eBay’s Hassan Sawaf. Not only in the context of post-editing, also in general use of MT for a global personalized customer experience. My own presentation, “Machine Translation and Post-Editing for User Generated Content”, also covered the key considerations for this topic. You can find the full presentation below.
- Dynamic engine adaptation methodologies and dynamic “real time” engine building were other hot topics. Welocalize’s VP of Language Tools and Automation, Olga Beregovaya, delivered a joint presentation with Alon Lavie from Safaba Translation Solutions on dynamic overnight engine training to meet the challenges of content drift. Prashant Mathur, a researcher from the Bruno Kessler Foundation, shared their work on multi-user adaptive statistical MT, which takes into consideration that several translators will often work on the same document or project and that the engine needs to adapt to the style and post-editing choices of each individual user. The presentation is posted for full view below.
- The Post-Editing Workshop provided new approaches to post-editing, including the use of monolingual post-editors and casual post-editors. The paper that resonated most with me related to the cognitive effort of MT errors for post-editors. It identifies specific categories of cognitively challenging MT errors and the goal is to understand what features of the source text are associated with cognitively demanding errors in MT output. Once this is identified, efforts can be concentrated on reducing the types of errors that require the most time and effort to fix during post-editing. This, combined with the dynamic engine adaptation mentioned above should improve productivity for post-editors.
All the Welocalize AMTA-2014 presentations are posted below. If you would like to discuss any of these topics, please contact me directly.
Training Manager, Welocalize Language Tools and Automation