Jie's assessment is correct, your moses.ini is missing a value under the [weight-t] section. There should be 5 values there, but instead there are 4. I'm not familiar with the MERT implementation that is bundled with moses these days, so I can't really tell you where to look, but it should definitely be tuning all 5 of these parameters.
On Sun, May 16, 2010 at 7:55 PM, David Edelstein <[email protected]> wrote: > I used the default settings and haven't done any factored training, so > I believe it's five. (I admit I am not sure exactly where that is > determined.) It's the number of scores in the phrase table, right? > Here is a sample: > > zyy mA qlt qbl kdh ||| as i said before this ||| 1 1.80204e-10 1 > 0.00438685 2.718 > zyy mA qlt qbl ||| as i said before ||| 1 5.83355e-08 1 0.0216905 2.718 > zyy mA qlt ||| as i said ||| 0.166667 1.17013e-07 1 0.0444402 2.718 > zyy mA ||| as ||| 0.00128535 8.39518e-06 1 0.261536 2.718 > zyy ||| as big as me ||| 1 0.00205425 0.5 0.00694445 2.718 > > > > On Sun, May 16, 2010 at 4:43 PM, Jie Jiang <[email protected]> wrote: >> Hi David: >> >> It is wired that the number of scores for [ttable-file] is specified to 5, >> but the number of weights in [weight-t] is only four. How many scores do you >> have actually? >> >> Best regards! >> >> Jie Jiang >> CNGL, School of Computing, >> Dublin City University, >> Glasnevin, Dublin 9. >> Tel: +353 (0)1 700 6724 > > _______________________________________________ > Moses-support mailing list > [email protected] > http://mailman.mit.edu/mailman/listinfo/moses-support > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
