MANY - Open Source Machine Translation System Combination
	
AUTHORS	
	Loïc BARRAULT - LIUM, University of Le Mans

THANKS	

	CMU and LIUM for having done a great job with the Open Source Sphinx library.
	More particularly Paul Deléglise and Yannick Estève for their help using it.
	M. Snover for having released the TERp software as Open Source.


PACKAGE CONTENT

  - Main software : MANY is written in Java and make use of TERp and Sphinx4
  - Scripts :
  	MANY.sh : The main script.
	add_id.sh : add id to 1-best output files (must be in the PATH or set in MANY.sh)
	genSphinxConfig.sh : to generate the config file (must be in the PATH or set in MANY.sh)

INSTALL

  - Compiling MANY :

To compile this software, a config file 'build.xml' for ant is provided.
Just run 'ant' (or 'ant -f build.xml') in the 'many' directory.
The library 'MANY.jar' and 'MANY_svn-revision-number' will be available in the 'lib' directory.


USE
  - Try the 'test' experiment which is provided in the test/ directory.
  - Prepare experiment : Several variables have to be set in order to run an experiment with MANY into EMS. They are all in the EMS config file.

 1. base-dir
 2. working-dir: where all the differents steps will be stored (steps, tuning, evaluation)
 3. lmdir, decodlm, lm-order and vocab: this setup the language model for TERp. Also let lmserver-port to -1 to use local lm, change it to use lm server.  
 4. moses-src-dir: to have access to mert 
 5. many-src-dir: where many is installed
 6. many: which version of MANY to use 
 7. shift-constraint, wordnet, paraphrases, shift-word-stop-list: constraint and resources for TERp 

  - Also, the 'score' scripts (inside scripts directory) must be modified in order to target the NIST score.rb scoring script.

 To run experiments : use the following command : 
 MANY_INSTALL_DIR/ems/experiment.perl --config config-file --no-graph --exec

 To continue a previous experiment run : 
 MANY_INSTALL_DIR/ems/experiment.perl --config config-file --no-graph --exec --continue run_number




 WARNING : if you try to use experiment.perl without --exec, then several errors will appear because some steps need some files created in a previous step (but they are created only when running those previous steps)



KNOWN BUGS
 - Empty lines in hypotheses files provoke an error. -> This is now fixed ! (02/2010)


RELATED WORK

	Sphinx4 	:	http://cmusphinx.sourceforge.net/sphinx4/  
	TERp		:	http://www.umiacs.umd.edu/~snover/terp/ 
	SRILM		:	http://www.speech.sri.com/projects/srilm/ 

COPYING / LICENSE
	See LICENSE.txt file provided along with this software.

BUGS	
	For help or more information, contact loic [dot] barrault [at] lium [dot] univ-lemans [dot] fr