Moses – I

by prashantmathur

I consider myself pretty new @ Machine Translation. I am going to start writing on Moses an opensource toolkit for Machine Translation that I’ll be using a lot from now on. I’ll be writing about the problems I faced and how I solved them after banging my head several times! 

Recently I was trying to get the word2word alignments from moses. 

Step 1 : Make sure that you’ve installed the latest version of Moses because some versions of Moses does not support word2word alignments. The version I was working with was “moses-20101201″. 

Step 2 : During training supply train-model.perl with “-phrase-word-alignment” option

Step 3 : During compilation of the translation model use processPhraseTable with “-alignment-info” option

Step 4 : During decoding use these decoder flags ”-use-alignment-info true -print-alignment-info -print-alignment-info-in-n-best -include-alignment-in-n-best -n-best-list NBEST.log 1″ 

In Step 4 if you use -print-alignment-info implies the usage of -use-alignment-info

AFAIK moses does not prints alignment info in the output file but instead it prints the alignments in NBEST.log. 1 represents that I want only 1 best translation.

PS: 1. Please do correct me if I am wrong.

2. Buzz me if you face problems, I forward your requests