According to Moses SMT, "MGIZA was developed by Qin Gao. It is an implementation of the popular GIZA++ word alignment toolkit to run multi-threaded on multi-core machines." The website referenced is now defunct, so don't hold your breath for updates.
I have also included the Python NLTK library on this, as I will try to do on all the machines.
vagrant ssh
First, ensure that pip, nltk and punkt are installed. If not, you can do this using the following commands:
sudo apt-get install python-pip
sudo pip install nltk
python -m nltk.downloader punkt
Then
cd ./tools
chmod +x *.sh
export INSTALL_HOME=/home/vagrant
echo "This is the part that takes a while"; ./getcorpora.sh $INSTALL_HOME && ./en-sp-align_words.sh $INSTALL_HOME /home/vagrant/tools/mgiza_configfile
Wait. A long time.
Once the script finishes running, you should be able to view the results in the corpora directory, under a filename similar to "src_trg.dict.A3.final.part000"
From the root of the git repository, cd ./mgiza && vagrant up will get it going. Go get a cup of coffee while MGIZA compiles. After a few minutes, the machine will be up and ready for you to work. Get in by using vagrant ssh. You'll notice several sub-directories in the vagrant user's home.
vagrant@vagrant-ubuntu-trusty-64:~$ ll
total 32
drwxr-xr-x 5 vagrant vagrant 4096 Nov 23 15:29 ./
drwxr-xr-x 4 root root 4096 Nov 23 15:29 ../
-rw-r--r-- 1 vagrant vagrant 220 Apr 9 2014 .bash_logout
-rw-r--r-- 1 vagrant vagrant 3637 Apr 9 2014 .bashrc
drwx------ 2 vagrant vagrant 4096 Nov 23 15:29 .cache/
-rw-r--r-- 1 vagrant vagrant 675 Apr 9 2014 .profile
drwx------ 2 vagrant vagrant 4096 Nov 23 15:29 .ssh/
drwxr-xr-x 4 vagrant vagrant 4096 Nov 23 15:33 tools/
The "tools" directory is the one you'll want to do your work out of. Note: At no time is sudo necessary to use these tools.
Looking inside the tools/ directory, here is what we find:
vagrant@vagrant-ubuntu-trusty-64:~/tools$ ll
total 28
drwxr-xr-x 4 vagrant vagrant 4096 Nov 23 15:33 ./
drwxr-xr-x 5 vagrant vagrant 4096 Nov 23 15:29 ../
drwxr-xr-x 3 root root 4096 Nov 23 15:33 bin/
drwxr-xr-x 3 353500 10000 4096 Jul 25 2008 europarl/
-rwxrwxr-x 1 vagrant vagrant 192 Nov 23 15:29 getcorpora.sh*
-rw-rw-r-- 1 vagrant vagrant 1649 Nov 23 15:29 mgiza_configfile
-rwxrwxr-x 1 vagrant vagrant 913 Nov 23 15:29 tokenize.sh*
getcorpora.sh will fetch the Europarl English-German parallel text and en-sp-align_words.sh will tokenize, transform, and run a mgiza's text classification on it. That part takes a while, so be patient. Also, a while means, "I let it run for six hours on 1024MB of memory and a single core and it didn't finish."
Update: Running this on a DigitalOcean droplet with 2 GB of memory, here is the time it took:
real 457m24.826s
user 457m10.177s
sys 0m10.712s
If you're curious as to how they're doing it, crack open the scripts and take a look. I won't tell anyone. en-sp-align_words.sh is mostly based on the work from this blog.