When extracting translation rules, only the most frequent word alignment
is stored. This is then used to calculate the lexical feature score.
I haven't heard of anyone doing anything different, if you like to try,
please tell us what you find
On 25/04/2014 10:17, Juan Pino wrote:
Hello,
I was wondering how Moses deals with multiple alignments
for a phrase pair. Are all alignments stored in the phrase table, or
only the most frequent one, or only the one that gives the highest
lexical feature score (because presumably the lexical feature weight
will be positive after tuning) ? Have there been analyses about which
strategy is best ?
Thanks very much,
Juan
_______________________________________________
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
