Transition-based Dependency Parser

About

This repository contains An arc-eager transition-based dependency parser for a master's course I took in 2017.

input

The parser takes as input CONLL06-formatted files from the Penn treebank (English) and the TIGER treebank (German).

output

The parser outputs 1 file which in CONLL06 format which contains the predictions. The files is saved as prediction-[lang].conll06 where [lang] refers to the current language being used. Example: prediction-english.conll06

Features

This parser:

• Uses an averaged perceptron to train.
• Extracts basic features (Nivre, 2008) and rich non-local features (Zhang and Nivre, 2011)

Usage

Run the python files via line commands or using any python IDE.

Training

The script can be called using terminal or shell commands with the following argument:

• < language >: language of the treebank data (either en for English or de for German).

python trainer.py <language>

Example

python trainer.py en

Parsing

The script can be called using terminal or shell commands with the following argument:

• < language >: language of the treebank data (either en for English or de for German).

python parser.py <language>

Example

python parser.py de