MGIZA++

Introduction

MGIZA++ is a software based on the famous word-alignment software GIZA++. Since GIZA++ is an signal-processing software and the processing of GIZA++ is time-consuming, MGIZA++ modify the structure of GIZA++ and then support the multi-thread architecture.

Support Word Alignment Model

• IBM Model 1
• IBM Model 2
• IBM Model 3
• IBM Model 4
• IBM Model 5
• IBM Model 6
• Hidden Markov Model

Install

For more information about how to install MGIZA++, please go to Install MGIZA++.

Input File Format

The input format of MGIZA++ is almost the same with GIZA++, please read the GIZA++#Input_Format part to get more idea.

The only different between these two softwares is for the cooccurence file. MGIZA++ needs cooccurence file for processing. To get the coocurence file, please input the commend

giza-pp/snt2cooc.out vocFile1 vocFile2 snt12

Parameters

The parameters for MGIZA++ is also almost the same with GIZA++, please refer to the GIZA++#Parameters part.

Since the MGIZA++ is for multi-processing GIZA++ training, the only different parameter is the number of CPUs using in the processing. Please indicate how many CPUs by using -ncpu NUM.

Run

The configure file is the same with GIZA++. Please refer to GIZA++#Run.

mgiza-pp/mgiza configure.gizacfg -ncpu NUM

Path Information on Verbs

• MGIZA++
  • /home/verbs/shared/stages/tools/mgiza-pp/

Reference

http://geek.kyloo.net/software/doku.php/mgiza:overview