Travatar

This is the home of Travatar, a tree-to-string statistical machine translation system. In other words, when performing translation, you first parse the input using a syntactic parser, then Travatar will perform translation for you. This is particularly effective for language pairs that require a large amount of reordering, such as English-Japanese.

Download/Install

Preparation

Travatar is mainly developed on linux and Mac OSX, but should also work on Windows (via cygwin). In order to install Travatar, you must have git, Boost and autotools installed. The easiest way to install these tools on (Ubuntu/Debian) Linux is: 

sudo apt-get install git g++ libboost-all-dev libz-dev autoconf automake autotools-dev libtool

Download/Compilation

The code of Travatar is distributed according to the LGPL, and can be distributed freely according to the license. You can download the latest version of Travatar by going to the github page, or directly by using the following git command. 

git clone https://github.com/neubig/travatar.git

Next, compile travatar with the following commands: 

cd travatar
autoreconf -i
./configure
make
src/bin/travatar --help

If the final command prints the Travatar help, everything is working properly.

Program Documentation

Overview

The following paper gives an overview of the motivation behind Travatar, and it would be great if you could cite it if you use Travatar in your work:

Travatar: A Forest-to-String Machine Translation Engine based on Tree Transducers
Graham Neubig. In Proceedings of the ACL Demonstration Track. 2013. 

@inproceedings{neubig13travatar,
  title = {Travatar: A Forest-to-String Machine Translation Engine based on Tree Transducers},
  author = {Graham Neubig},
  booktitle = {Proceedings of the ACL Demonstration Track},
  address = {Sofia, Bulgaria},
  month = {August},
  year = {2013}
}

Training

In order to run Travatar, you will have to train a model. Please take a look at the step-by-step guide to creating a model for Travatar. There are also a number of training options that will affect how training works.

Translating

In order to translate with Travatar, you use the model defined in the previous step. In the most simple of situations, you must simply run the following command, substituting travatar.ini with the appropriate configuration file: 

cat input-parsed.txt | travatar -config_file travatar.ini > output-words.txt

Here, input-parsed.txt is a file containing one syntactic parse tree per line.

Development/Support

Contributors

If you have any questions about Travatar, please send them to the travatar-users mailing list, which you can join by sending mail to travatar-users+subscribe with the domain googlegroups.com. (If you don't get a reply, you can also send questions to neubig at gmail.com.)