Thrax uses Apache hadoop (an open-source implementation of MapReduce) to efficiently extract a synchronous context-free grammar translation model for use in modern machine translation systems.

Thrax currently has support for both Hiero-style grammars (with a single non-terminal symbol) and SAMT-style grammars (where non-terminal symbols are calculated by projecting onto the span from a target-side parse tree).

COMPILING:

First, you need to set two environment variables: $HADOOP should point to the directory where Hadoop is installed. $AWS_SDK should point to the directory where the Amazon Web Services SDK is installed.

To compile, type

ant

This will compile all classes and package them into a jar for use on a Hadoop cluster.

At the end of the compilation, ant should report that the build was successful.

RUNNING THRAX: Thrax can be invoked with

hadoop jar $THRAX/bin/thrax.jar <configuration file>

Some example configuration files have been included with this distribution:

example/hiero.conf
example/samt.conf

COPYRIGHT AND LICENSE: Copyright (c) 2010-13 by the Thrax team: Jonny Weese jonny@cs.jhu.edu Juri Ganitkevitch juri@cs.jhu.edu

See LICENSE.txt (included with this distribution) for the complete terms.