==========================================================================================

     Introduction to Natural Laguage Processing Assignment 3

==========================================================================================

1. In this assignment, you will be asked to:

From Incrementality in Deterministic Dependency Parsing(2004, Nivre)

• implement the arc-standard algorithm

From A Fast and Accurate Dependency Parser using Neural Networks(2014, Danqi and Manning)

• implement feature extraction
• implement the neural network architecture including activation function
• implement loss function

• You will need to copy your embeddings to this folder, and name it as "word2vec.model" Also, change "embedding_size" in Config.py to match your model.

2. This package contains several files:

• DependencyParser.py: This file is the main script for training your dependency parser.

• DependencyTree.py The dependency tree class file.

• ParsingSystem.py This file contains the class for a transition-based parsing framework for dependency parsing.

• Configuration.py The configuration class file.

• Config.py This file contains all hyper parameters.

• Util.py This file contains functions for reading and writing CONLL file.

• data/ train.conll - train set, labeld dev.conll - dev set, labeld test.conll - test set, unlabeled

3. What you should do:
4. Implement the arc-standard algorithm in ParsingSystem.py
5. Implement feature extraction in DependencyParser.py: getFeatures(...)
6. Implement neural network architecture including activation function: forward_pass(...)
7. Implement loss function and calculate loss value: in DependencyParserModel.build_graph(...)
8. Try different number of hidden layers
9. Try different non-linear activation functions
10. Train a parser with Fixed embeddings -- by setting trainable=False in tf.Variable