Olga Beregovaya, Welocalize VP, Language Tools, shares the buzz around machine translation (MT) that took place at LocLeaders forum + LocWorld, London.
A few years ago, case studies of successful MT deployments were sparse. Those using MT in their localization had the status of daredevils. At LocLeaders + LocWorld, many MT cases were presented and discussed in great detail. MT is here to stay. And real industry efforts are going into making the best of MT. Here’s my summary of the best (MT) nuggets from the week in London.
What’s the major development in MT?
Focus has shifted from assessing the quality of MT output – by default, MT output is always flawed and imperfect – to assessing WHAT the output will be used for.
• How exact does the translation need to be for the purpose it serves? How relevant is the MT output?
• How easy is it for the end user to understand and use the target language appropriately?
• What is required by the post-editor to bring output to the human-level quality?
This is bringing about change. Adequacy of MT output is now favored over fluency.
What are the two main topics surrounding MT now?
1. How to measure translators productivity and map to payment methods
2. The variations in acceptable post-editing quality levels
Both subjects reflect that MT is changing the translator industry landscape.
There is a shared understanding that the old quality models may no longer be the answer when applied to post-edited output used for new content delivery models.
What are challenges facing clients for MT?
Getting full vendor + translator buy-in, both in terms of post-editing and the most fair way of pricing it. The challenge is to set the right expectations around the MT engine and agree what the ultimate goals of the MT program are – reducing cost, time-to-market, fresh content to new audiences, enabling real-time customer support and more.
What are the key technical developments for MT?
Focus more on HOW MT is used in the translation industry, rather than on MT engine improvements. It’s not necessarily an additional 5 BLEU score points – although that IS important – but rather build an MT engine that delivers to the purpose and produces usable content that meets the objectives of the overall localization strategy.
Additionally, Francois Richard, HP, was asked to give a PechaKucha on MT at the LocLeaders forum in London. It’s a presentation style in which 20 slides are shown for 20 seconds each – a format which keeps presentations precise and fast-paced, powers multiple-speaker events.
[blockquote cite=”- Francois Richard, HP” align=””] Once I’d delivered my PechaKucha on HP & MT at LocLeaders, it was both a pleasure – to be part of an MT-aware audience – and a pressure – for the same reason. I knew pricing MT post-editing would be part of the debate and it was. Getting to talk to translators about their daily work is always very useful and enjoyable. And as Betty’s (Galliano, CEO, Ocean Translation) quoted, “Don’t worry … Translators will get used to MT. [/blockquote]
For information about machine translation and to watch more of Olga on MT please visit, /wemt/