El dl 23 de 04 de 2012 a les 10:45 +0200, en/na Per Tunedal va escriure: > Hi again Francis, > > Pronouns are important in a language. In my French grammars the pronouns > have at least 50 % more pages than the verbs. The same applies to my > English grammars. All the same, most of the education is spent on verbs. > The same applies to translation systems? A great effort on getting the > verb conjugations right and leaving the pronouns aside?
That could be just an indication of what linguists find more interesting as opposed to what is educationally more important. Also, the relative ease of being exhaustive with pronouns (or any closed category), as opposed to verbs (or any other open category). > On Sun, Apr 22, 2012, at 18:47, Francis Tyers wrote: > > --snip-- > > > > I think it would be really great if it could be fixed. But in the > > meantime, we work with what we've got. Leave the repetitive bit up to > > the computer, and leave the human being to work out the hard bit. > > You've got a point. The hard bits are the most interesting. You wouldn't > want to take the fun out of translation, would you? :-) I think it's a matter of where to put effort, and how much return you get on your time invested. You're right that it is probably the most linguistically interesting, but in terms of "getting your MT system working" interesting, it is lower on the scale. > > > > > B. > > > > > > ... The translation of the personal pronoun "lui" will be either > > > > > > "honom" (masculine = him) or "henne" (feminine = her). > -- snip-- > > > > Well, it might be that I'm used to listening to non-native English, but > > often Romance language speakers make his/her mistakes in English. When > > the referent is obvious it doesn't really cause a problem. > > > > "My mum got stranded last week when his car broke." > > > > Unless there is a preceeding part of the discourse where another > > person's car is introduced, it's quite clear who the car belongs to. > > That is, the gender mistake does not effect the intelligibility of the > > final translation. > > > > Your right. :D > --snip-- > > > > > > > > > > > > D. > > > > > >...it makes a difference if a possessive pronoun refers to the > > > > > > subject or refers to someone else: > > > > > > "Kalle älskar sina hundar. Niklas älskar hans hundar." = "Kalle > > > > > > aimes > > > > > > ses chiens. Niklas aimes ses chiens." (Kalle loves his dogs. Niklas > > > > > > loves his dogs. It's the same dogs, that are kept by Kalle, not by > > > > > > Niklas! Same problem in English.) > > > > > > > > > > Someone might have encountered this problem working on e.g. the pair > > > > > IS-EN or DA-EN. > > > > > > > > Icelandic->English we just translate 'sin' and 'hans' to 'his' I think. > > > > > > Yes, but the problem is the other way around: En - SV. How to translate > > > "his dogs"? How to know if the dogs belongs to Kalle or to Niklas? The > > > Swedish translation will imply that one of them is the owner. How do you > > > tackle it in the pair Icelandic->English? > > > > We don't translate from English->Icelandic, only from > > Icelandic->English. > > > > The possibility of building a one way system never has occurred to me. > It would significantly simplify the work if I decided to build only a > Swedish to English or Swedish to French system. > I will consider your idea! Or use different technology for different directions... > --snip-- > > > > > > > > Sorry not to be more positive, but most of what you're describing are > > > > general problems with translation (and machine translation), not just > > > > with Apertium. You can approximate a fix, but you're never going to get > > > > it right all the time, the best you can do is get the closest > > > > approximation possible. > > > > > > True. The reason I am interested in Apertium, is that I expected to get > > > a more consistent translation than with statistical translation. I also > > > expected to overcome some common errors by somehow creating a rule. But, > > > as I said before, I am not sure it's the right tool for SV-FR. (SV-EN > > > are closer, but there still remains some issues, as you can see.) > > > > Well, it will be a more consistent translation, whether it is more > > consistently wrong or not is another matter ;) > > > > On a purely linguistic and technical level, it is not the best tool for > > the job of translating Swedish<->French. > > > > There are other systems which would linguistically be better choices. > > You could for example try setting up OpenLogos, or LOGON, or Matxin > > which give you a deeper analysis. Or training Moses, which will give you > > state-of-the-art SMT results. You could also try buying an off-the-shelf > > system from someone like SYSTRAN or Gramtrans. > > I have checked the alternatives. OpenLogos has nothing for Swedish, but > LOGON has both a Swedish and a French dictionary (in an early stage, I > suppose). I will have to dig deeper into possible solutions for SE - FR. Ok :) F. ------------------------------------------------------------------------------ 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
