Transition-based (Arc Eager) Dependency Parsing

Author: Touhidul Alam

Usage:

python3 parser.py [train/dev/test] [en/de] file_path_location

Example:

python3 parser.py train en /mount/studenten/dependency-parsing/data/english/train/wsj_train.only-projective.conll06 

Dependencies:

• Numpy
• pickle

Components

• parser.py

  • State - class initialize initial stack, buffer, arcs, left-dependency and right-dependency
  • Instance - class saves the transition label with its corresponding features
  • Parser - class does oracle parsing during training and predicting and making a parser

• reader.py

  • Corpus - class to read sentences from dataset and build them into sentence
  • Sentence - holds the datasets content

• model.py

  • FeatureMapper - takes the unique feature and add it to the feature dictionary. Feature contains baseline feature, and extended features from Nivre (2011)
  • Model - saves the FeatureMapper and weights, implemented with Averaged perceptron

Output File

pred_en.conll06 - English
pred_de.conll06 - German