Hi, the file includes information about each input word in the test set.
The six fields are: - precision by word: how often the translations of input word match the reference - deleted by word: how often the input word was not translated (un-aligned) - length by word: how many words the input word was translated into - total number of occurrences of the input word in the test set. - how many times the input word occurs in the corpus - the surface form of the word Note that the first three numbers are aggregates. If you want to get ratios (i.e., what percentage was deleted), you need to divide by the fourth number (total occurrences in the test set). -phi On Mon, Jul 23, 2012 at 3:01 PM, John Morgan <[email protected]> wrote: > Hello, > I'm using ems with the following settings under the {EVALUATION] stanza: > analysis = $moses-scripts-dir/ems/support/analysis.perl > analyze-coverage = yes > report-segmentation = yes > report-precision-by-coverage = yes > > In the file > precision-by-input-word > 6 fields are written per line. > Could someone explain what each field contains? > > > > > -- > Regards, > John J Morgan > _______________________________________________ > 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
