This directory contains a tool for extracting transfer rules from
parallel MRS corpora, developed at Nanyang Technological University
(NTU) by Petter Haugereid (2010-2012). More detailed instructions on
how to use it can be found at
http://moin.delph-in.net/MtRuleExtraction

In order to make it work you need to have the LOGON system installed:

http://moin.delph-in.net/LogonInstallation

You need a parallel corpus in two files. If one of the languages is
Japanese, you will need to install the MeCab tagger:

sudo apt-get install python-yaml
sudo apt-get install mecab-ipadic-utf8 python-mecab

You will need to download the Anymalign phrase aligner:

http://perso.limsi.fr/Individu/alardill/anymalign/latest/anymalign2.5.zip

And put it in the same directory as the bash script.

To get more results, you should also use the MOSES system, but that is
not a part of this procedure.  This directory contains a shell script
that runs through the whole procedure of rule extraction. It can be
executed with:

bash extr-rule.bash SOURCE TARGET CORPUS TRANSDIR

e.g.

bash extr-rule.bash corpora/mini.ja corpora/mini.en mini jaen

TRANSDIR contains the language pair specific code.
CORPUS is a name for this particular run

The resulting transfer rule files are written into the TRANSDIR

For more information abour each program run by the script, see
comments in extr-rule.bash

To add MOSES, use it to align the mrs files and then use
phrtab-thin.py to combine its phrase table with the one from
anymalign.