Hi Loic, bare in mind that there are two options:
-output-search-graph -output-word-graph. I'm not sure about it, but I *think* that -output-search-graph does match the information listed in the web page. The option -output-word-graph generates a word-graph, which is a weighted acyclic graph in which each node represents a partial translation hypothesis and each edge is labelled with a word (in this case phrase) of the target sentence and is weighted according to the scores given by the SMT model. Theoretically, it contains all hypotheses which have been considered during the decoding procedure. The format is "HTK-similar" The HTK-format is a well documented format, very used in Speech and Text recognition. The HTK book, which contains documentation about the HTK format, can be accessed here: http://htk.eng.cam.ac.uk/docs/docs.shtml In this case, the "a" tag is used to output the translation model scores (direct, inverse translation models, lexical weights and phrase penalty, weighted according to moses.ini). The "r" tag is equivalently used for the distortion model. "l" is the language model score. However, I wonder why you get all those zeroes... Regards and best luck, Germán Sanchis-Trilles Quoting Loic Barrault <[EMAIL PROTECTED]>: > Hi, > > I wonder what is the format of the file produced by the > -output-search-graph option ? > > The details in the documentation > (http://www.statmt.org/moses/?n=Moses.AdvancedFeatures#ntoc10) do not > match the format I obtain. > > Here is a sample (actually first lines) of the output-search-graph file > I obtain : > > VERSION=1.0 > UTTERANCE=0 > J=0 S=0 E=60 a=0, 0, 0, 0, 0 l=0 r=-50, 0, 0, 0, 0, 0 > w=, pC=0, c=0 > J=1 S=0 E=59 a=0, 0, 0, 0, 0 l=0 r=-40, 0, 0, 0, 0, 0 > w=, pC=0, c=0 > J=2 S=0 E=5 a=-2.94444, -3.38989, -2.85263, -3.8712, > 0.999896 l=-6.21875 r=0-0.251314, 0, 0, 0, 0, 0w=I am , > pC=-0.401291, c=-0.98555 > J=3 S=0 E=7 a=-5.273, -5.38024, -3.95124, -5.66297, > 0.999896 l=-5.60606 r=0-0.510826, 0, 0, 0, 0, 0w=I 'll , > pC=-0.681765, c=-1.15916 > J=4 S=0 E=1 a=-3.01408, -3.83185, -0.655406, -2.19723, > 0.999896 l=-3.82366 r=0-0.0727589, 0, 0, 0, 0, 0 > w=I 'm , pC=-0.33072, c=-0.739751 > J=5 S=0 E=2 a=-4.59728, -4.82598, -1.31219, -0.693147, > 0.999896 l=-2.39167 r=0-0.215111, 0, 0, 0, 0, 0w=I , > pC=-0.417593, c=-0.77641 > J=6 S=60 E=119 a=0, 0, 0, 0, 0 l=0 r=-70, 0, 0, 0, 0, 0 > w=, pC=0, c=0 > J=7 S=59 E=178 a=0, 0, 0, 0, 0 l=0 r=-40, 0, 0, 0, 0, 0 > w=, pC=0, c=0 > J=8 S=60 E=65 a=-2.94444, -3.38989, -2.85263, -3.8712, > 0.999896 l=-6.21875 r=-110, 0, -2.19723, 0, 0, 0 > w=I am , pC=-0.401291, c=-0.98555 > J=9 S=5 E=227 a=-2.94444, -3.38989, -2.85263, -3.8712, > 0.999896 l=-6.21875 r=-4-0.251314, 0, 0, 0, 0, - > 2.19723 w=, pC=0, c=0 > J=10 S=59 E=124 a=-2.94444, -3.38989, -2.85263, -3.8712, > 0.999896 l=-6.21875 r=-90, 0, -2.19723, 0, 0, 0w=I am , > pC=-0.401291, c=-0.98555 > > Thank you. > > Loïc > > _______________________________________________ > Moses-support mailing list > [email protected] > http://mailman.mit.edu/mailman/listinfo/moses-support > ---------------------------------------------------------------- This message was sent using IMP, the Internet Messaging Program. _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
