El dl 23 de 04 de 2012 a les 14:44 +0200, en/na Per Tunedal va escriure: > Hi Francis, > > Personally, I am always interested in what is not working. Things > working all right are totally uninteresting to me!
Well, in Apertium at the moment, for Swedish--French, and for free/open-source rule-based MT systems for Swedish--French, nothing is working! > In fact I have made several studies on projects that wasn't completed > or didn't meet the expectations. I've learnt a lot from that. Getting your expectations right is a good thing to do when working with machine translation. If your expectations are right, then you are more likely to meet them ;) > Now, I am very interested in the shortcomings of MT. When I let someone > review a machine translated text they laugh and ridicule it. This can have several reasons: 1) They are impolite -- polite people don't laugh at language learners, or disabled people, a machine translation system is basically that. 2) They don't know how machine translation works -- there could be an ambiguity which is not context resolvable. 3) The translation is actually funny -- there is some mistake somewhere along the line that results in something actually humorous. I think the outcome is pretty evenly split between the three categories. > Once, I was an agent of a very nice piece of software and suddenly got a copy > that > was machine translated to Swedish. My wife never has laughed as much as > when she tried out that program. She laughed until her tears came. She > collapsed on the chair and I wasn't talk-able. At the course in Russia earlier this year there were some moments like that, so trust me, even MT developers can laugh at our work. ;) Ooh, this one is good: http://wiki.apertium.org/wiki/Talk:Welsh_to_English/Archive_1#.281.3.36.29_Preverbal_particles_-_affirmative If you scroll down to the parts about the "dog" (a disambiguation error), it still makes me laugh even to this day "Dog rust going" LOL (incidentally we now get the right translation) > Why put any effort on what's already working? This is why I try and work with languages that don't already have a working machine translation system. Languages which do are not so interesting. > At this time, machine > translation like Google Translate and Bing Translator make fairly good > translations. But they often fail on the same problems. Like pronouns, > for instance. :-) And people laugh at them. They aren't free software. People laugh at a lot of things, some funny some not (see above) :) > That's why I am looking for better alternatives. The rule based/transfer > systems are for the moment overtaken by the statistical translation > systems, but I believe that might change. It's not going to change without quite a bit of work, and quite a bit of work doing things which "already work" with other systems. Fran ------------------------------------------------------------------------------ For Developers, A Lot Can Happen In A Second. Boundary is the first to Know...and Tell You. Monitor Your Applications in Ultra-Fine Resolution. Try it FREE! http://p.sf.net/sfu/Boundary-d2dvs2 _______________________________________________ Apertium-stuff mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/apertium-stuff
