Adam Lopez wrote: > There were also some experiments during the CLSP workshop that built Moses > (Fig. 6.8): > http://www.clsp.jhu.edu/ws2006/groups/ossmt/documents/jhu-summer-workshop-final-report.pdf > > Since MERT is only tuning a handful of parameters, it would be somewhat > surprising if a drastically large tuning set made a difference.
Contrary to Figure 6.8, I've run some experiments where test set performance continues to significantly improve as I increase MERT set size - I stopped at 10,000 tuning pairs. But this was for a very different domain from that of the training data, namely scientific abstracts. - John Burger MITRE
smime.p7s
Description: S/MIME cryptographic signature
_______________________________________________ Moses-support mailing list Moses-support@mit.edu http://mailman.mit.edu/mailman/listinfo/moses-support