We provide human-written summaries and human-evaluated ratings to the summaries in this data.

Summaries are written by humans in the following three conditions: Manual, AI post-edit, Human post-edit. Each participant summarized 10 distinct documents and there were 12 participants in each condition.

The human-written summaries, together with the generated summaries, references (provided by the original dataset), and random summaries (created randomly using the opposite dataset) are evaluated by another set of humans. Each annotator rated six summaries (Manual, AI post-edit, Human post-edit, AI-generated, Human reference, Random) written for the same document.

There are a total of 720 rows for each dataset, XSum and Reddit-TIFU. The data is stored in a DataFrame and saved in a pickle.