SIMSUM: Document-level Text Simplification via Simultaneous Summarization

This repo is the codes for Paper SIMSUM: Document-level Text Simplification via Simultaneous Summarization (ACL2023)

Installation

The required packages can be installed by

pip install -r requirements.txt

Processed Dataset

The datasets used for document-level simplification are listed in /SimSum/data , named D-Wiki for D-Wikipedia and wiki_doc for WikiDoc.

Training

To train the model:

python main.py

Bart2.py and T5_2.py means our SimSum model with BART and T5 as the backbone. For the single model, use the Bart_baseline_finetuned.py and T5_baseline_finetuned.py.

Automatic Evaluation

To evaluate the model,

python evaluate.py

which will compute the SARI, D-SARI, BLEU and FKGL score.

Human Evaluation