Text Simplification to Specific Readability Levels

The ability to read a document depends on the reader’s skills and text’s readability level. In this paper, we propose system that uses deep learning techniques simplify texts in order match We use novel approach with reinforcement loop contains classifier. classifier’s output is used decide if more simplification needed, until desired level reached. models are trained data annotated levels from Newsela corpus. Our perform at sentence level, each meet specified version of corpus aligned also produce an augmented dataset by automatically annotating pairs sentences using readability-level text achieve better performance than state-of-the-art for task.