11 March 2014

MT update, take #2

Hard on the heels of my update of 9 March, here's another that takes a different tack and suggests that things might be moving faster -- or not -- than David Grunwald suspects.

Take a look at the article entitled: A world of seamless and immediate language translation gets closer on the The Posse List, a blog that "pretty much everything across the legal employment field".

Quote:
We are going to have much, much, much more on this in March. In late February the founder of The Posse List will be meeting again with IBM  Research’s Cognitive Systems team  and Watson team at the Mobile World Congress with language translation at the top of agenda, and then attending a “global imperialism of English” workshop in Paris.
Coming back myself on the opposite tack, it is also true that it must be at least forty years since we first heard claims that under certain circumstances a certain MT system could translate a certain language pair with an accuracy of 90%. In the Posse List article the claim reads:
Despite its simplicity, the method is surprisingly effective:  they achieved an almost 90% precision for translation of words between English and Spanish. (Check the article to get the full context.) 
As one colleague pointed out, if you're getting ten words out of every hundred wrong and occasionally one of those words is a negative -- a simple word like 'no' or any equivalent -- (and not to mention any of a myriad of other mysteries of language), then you're not doing very well at all.
As another suggests, what do you say to a cook who says "well 90% of the dishes I serve are delicious and no more than 10% are likely to cause nausea, colic or occasionally even death"?

ChatGPT, a drafting aid for translation by emulation

On 17 October 2011, I published the first of two posts summarising my general approach to the type of translation/adaptation services I was ...