SelectiveGeneration

Code base for Mei et al. NAACL 2016 paper

This codebase is revised from my original version and still under-tested, so I'd share my original code and data files through this Google drive:

https://drive.google.com/drive/folders/0BwqmV0EcoUc8OHpLd3NaU19Hb0E?usp=sharing

Dependencies

• Anaconda - Anaconda includes all the Python-related dependencies
• Theano - Computational graphs are built on Theano
• NLTK - Natural Language Toolkit
• JPype - Call Java in Python
• JDK - Java Development Toolkit
• ArgParse - Command line parsing in Python

Instructions

Here are the instructions to use the code base.

Prepare Data

Creat a folder called 'data' and copy data files (to be added) to it.

Prepare Eval Code

Creat a folder called 'dist'.

Download and compile the Java code from Gabor Angeli's 2010 EMNLP paper.

Copy the generation.jar file to the dist folder.

Train Models

To train the model with options, use the command line

python train_models.py --options

For the details of options, please check

python train_models.py --help

Test Models

Choose a model to evaluate on dev or test set, with the command line:

python test_models.py --options

For the details of options, please check

python test_models.py --help

License

This project is licensed under the MIT License - see the LICENSE.md file for details