Efficient Comprehension: PMTSS for Simplifying and Summarizing Scientific Literature

Text summarization aims to extract key information from lengthy texts, facilitating rapid comprehension for readers. Additionally, text simplification seeks to render complex terminology into more simple language for a broader audience. This study delves into the realm of lay summarization, which involves generating simplified summaries of scientific articles to make them more digestible. We present the Pre-trained Model for Text Simplified Summarization (PMTSS), a systematic framework devised for this purpose. Initially, the backbone model was determined by comparing the state-of-the-art pretrained models for summarization since the models could generate simplified text to some extent. Extensive experiments and comprehensive evaluation metrics were utilized and the impact of different input-output lengths on model performance was also investigated. Our research focused on the relevance, readability, factual accuracy, and simplicity of the generated text. Additionally, we improved the performance of the summarization model by employing data augmentation and intermediate pretraining techniques. The results have demonstrated that our framework shows superior performance compared to reasonable baseline models across all evaluation metrics, achieving a score of 35.85 on the comprehensive evaluation metric CSS for this task.