An Analysis of Crowdsourced Text Simplifications

We present a study on the text simplifica-tion operations undertaken collaboratively by Simple English Wikipedia contribu-tors. The aim is to understand whether a complex-simple parallel corpus involv-ing this version of Wikipedia is appropri-ate as data source to induce simplifica-tion rules, and whether we can automat-ically categorise the different operations performed by humans. A subset of the cor-pus was first manually analysed to iden-tify its transformation operations. We then built machine learning models to attempt to automatically classify segments based on such transformations. This classifica-tion could be used, e.g., to filter out po-tentially noisy transformations. Our re-sults show that the most common transfor-mation operations performed by humans are paraphrasing (39.80%) and drop of in-formation (26.76%), which are some of the most difficult operations to generalise from data. They are also the most diffi-cult operations to identify automatically, with the lowest overall classifier accuracy among all operations (73 % and 59%, re-spectively). 1