Leveraging Faithfulness in Abstractive Text Summarization with Elementary Discourse Units

Abstract

Abstractive text summarization uses the summarizer's own words to capture the main information of a source document in a summary. While it is more challenging to automate than extractive text summarization, recent advancements in deep learning approaches and pre-trained language models have improved its performance. However, abstractive text summarization still has issues such as hallucination and unfaithfulness. To address these problems, we propose a new approach that utilizes important Elementary Discourse Units (EDUs) to guide BART-based text summarization. We compare our approach with some previous approaches that have improved the faithfulness of the summary. Our approach was compared and tested on the CNN/Daily Mail dataset and showed an improvement in truthfulness and source document coverage.