Text this: Abstractive Summarization of Historical Documents: A New Dataset and Novel Method Using a Domain-Specific Pretrained Model