Mitigating Data Scarceness through Data Synthesis, Augmentation and Curriculum for Abstractive Summarization

This paper explores three simple data manipulation techniques (synthesis,\naugmentation, curriculum) for improving abstractive summarization models\nwithout the need for any additional data. We introduce a method of data\nsynthesis with paraphrasing, a data augmentation technique with sample mixing,\nand curriculum learning with two new difficulty metrics based on specificity\nand abstractiveness. We conduct experiments to show that these three techniques\ncan help improve abstractive summarization across two summarization models and\ntwo different small datasets. Furthermore, we show that these techniques can\nimprove performance when applied in isolation and when combined.\n