Hi mailing,
I have implemented a phrase based translation system. My model I does nothing with reordering or distortion. I would like to evaluate my model I next to the baseline moses system. As there are no human aligned phrase evaluation schemes, the only thing I can do is use BLUE/NIST as an evaluation as far as I know off. Also, because my Model I does nothing with reordering for now, I do not want the baseline moses translation model to use reordering either. I was thinking about the following strategy: Use a simple moses.ini file that references only to the language model and translation model and the standard weights and nothing else. (Exclude the reordering file and set the distortion weight to 1) My questions are: 1) Does it make sense to evaluate like this? 2) Should I exclude MERT too? 3) What about filtering? The tables are not that big in this phase Thank you, Sanne Korzec
_______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
