Hi,indeed, there were some inconsistencies which I tried to fix, partially at least.
I don't really remember much, but about the hypothesis recombination issue, I think that it would be a good idea to know when a given hypothesis stems from a recombination, so perhaps it could be a good thing to unify both osgx and osg, but in the osgx case add a further field that specifies this.
About the initial hypothesis, I didn't really realize the issue, but sounds good to me :)
Cheers, Germán On Thu, 26 Jul 2012, Lane Schwartz wrote:
I'm doing some work using output-search-graph and output-search-graph-extended, and I've noticed some inconsistencies in the formats that I'd like to reconcile. It looks like Germán did some work towards this end in October (http://thread.gmane.org/gmane.comp.nlp.moses.user/5393). Before I make the changes, I just want to make sure that these items that look like bugs to me are not in fact features. In Manager.cpp, lines 653 & 671-674: When producing -osg, the hypothesis ID is always printed for "hyp=" (line 653) When producing -osgx, the hypothesis ID is normally printed for "hyp=" (lines 673-674), but when there is hypothesis recombination, the superior hypothesis ID is printed here instead (lines 671-672). This is odd, and I propose to delete lines 671-673, always printing out the actual hypothesis ID for "hyp=". Does anyone object? In Manager.cpp, lines 640-644: When producing -osg, the forward pointer and fscore are printed for the initial hypothesis When producing -osgx, the forward pointer and fscore are not printed for the initial hypothesis I propose to print these items for both -osg and -osgx Please let me know if anyone has any comments or objections. Thanks, Lane _______________________________________________ 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
