prepare-switchboard

This repository updates the originally written classes for The Switchboard Dialog Act Corpus and uses it to freely parse the dataset.

Instructions to run

1. Extract swda.zip dataset inside data/raw/
2. Run python main.py to create train/val/test splits. The *_sentences.tsv files generated could be used to train an autoencoder while *_dialog.tsv files could used to train a simple sequence-to-sequence model for dialog generation.