ICCCI - Sentence Compression with deletion

Dataset

The dataset is available at: https://github.com/google-research-datasets/sentence-compression. Download and store the *.gz files in data/ directory.

Requirements

This project requires python3.6+ and pytorch1.1+. It used the models and embeddings from FLAIR framework:

pip install flair

Preprocess data

In order to train a sequence tagging model, the original data need to be align into sequence tagging format. To align the downloaded data:

export PRJ_HOME=<path/to/this/project>
bash $PRJ_HOME/runs/preprocess.sh

Training

Different training configs for each settings are available in runs/. To start training:

export PRJ_HOME=<path/to/this/project>
bash $PRJ_HOME/runs/train_<config_name>.sh