On Tue, Apr 5, 2011 at 2:44 AM, Andrei Zene <[email protected]> wrote: > Dear Kevin, > > Thank you for your reply. If I understood correctly my job would be to play > with the smoothing parameter (alpha) for every language until I get best > results? If so, then it is ok for me. Maybe the fact that I speak Romanian > and German could be an advantage for me when testing, because both languages > have diacritics. > > I would have two other questions: > 1. I see on apertium ideas page that a preferable requirement is knowledge of > FSTs. What does FST stand for? > > 2. I do not understand exactly what you mean when you say: "Your proposal > should include code for evaluation of your system in any case (the charlifter > perl code does 10-fold cross validation on the training data)." >
1. Finite state transducer 2. 10-fold cross validation means that you divide up the training data into ten parts. Then set aside one part, train on the other 9/10, and evaluate on the 1/10 you set aside. Repeat for the other nine tenths, and average. It's important that your port to C++ have the ability to this kind of evaluation - only a tiny amount of additional work to do so. Once you have this, then tuning smoothing parameters is trivial - just lots of CPU time. Kevin ------------------------------------------------------------------------------ Xperia(TM) PLAY It's a major breakthrough. An authentic gaming smartphone on the nation's most reliable network. And it wants your games. http://p.sf.net/sfu/verizon-sfdev _______________________________________________ Apertium-stuff mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/apertium-stuff
