Paraphrase Generation

This is a PyTorch implementation of the simple sequence-to-sequence paraphrase generator using Transformer (Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin, arxiv, 2017).

Usage instructions

Train parpahrase generation model

python train.py \
--train_source_file <file_path> \
--train_target_file <file_path> \
--valid_source_file <file_path> \
--valid_target_file <file_path> \
--spm_file <file_path>

• train_source_file one-sentence-per-line raw corpus file for training.
• train_target_file one-sentence-per-line raw corpus file for training.
• valid_source_file one-sentence-per-line raw corpus file for validation.
• valid_target_file one-sentence-per-line raw corpus file for validation.
• spm_file SentencePiece model file.

Generate paraphrase

python generate.py \
--input_file <file_path> \
--output_file <file_path> \
--spm_file <file_path> \
--search_width 3

• input_file one-sentence-per-line raw corpus file to paraphrase.
• output_file output file
• search_width beam search width.

Example

Quora Question Pairs

Download data from here. Download trained model parameters and generated paraphrases from here.