The Galactic Dependencies Treebanks 1.0

Implementation of the paper "The Galactic Dependencies Treebanks: Getting More Data by Synthesizing New Languages" by Dingquan Wang and Jason Eisner. TACL 2016

Download

The data release is here, this yields 37 × 38 × 38 = 53,428 languages in 70+G compressed files (700+G after extraction) in total.

• To download and extract the entire dataset (Please make sure you have enough disk space):

    GALACTIC_ROOT=$(pwd) bin/gd-fetch 
  

• If you want to download only a subset, for example, synthetic languages substrated by English or German without extraction:

    GALACTIC_ROOT=$(pwd) bin/gd-fetch --substrate GD_English GD_German --pipeline download
  

• For more options, please use:

    bin/gd-fetch --help
  

Build

• Compile the code from the command line:

    mvn compile
  

• To build a single jar with all the dependencies included:

    mvn compile assembly:single
  

Run

• To train a permutation model of, for example, NOUN from a given treebank (toy/sample.conllu):

    GALACTIC_ROOT=$(pwd) bin/gd-train-permute --input toy/sample.conllu --node N 
  

• To permute a given treebank (toy/sample.collu) using the given permuatation models to a synthetic language, for example, enfr@Nhi@V:

    GALACTIC_ROOT=$(pwd) bin/gd-translate --input toy/sample.conllu --spec en~fr@N~hi@V
  

• For more options, please use:

    bin/gd-train-permute --help
    bin/gd-translate --help
  

Note: The given model files are generated from a slightly older Pacaya version, which is no longer used in the current release. So the models reproduced from the current version might be slightly different from what are given.

Reference

@article{galactic16,
    author = {Dingquan Wang and Jason Eisner},
    title = {The {G}alactic {D}ependencies Treebanks: Getting More Data by Synthesizing New Languages},
    journal = {Transactions of the ACL},
    year = {2016},
    note = {In review}
}