This is a dataset that covers monotonicity reasoning.
It is based on MED by Hitomi Yanaka, Koji Mineshima, Daisuke Bekki, Kentaro Inui, Satoshi Sekine, Lasha Abzianidze, and Johan Bos.
A few corrections have been made to entries from MED, and 544 additional entries were added.
Where applicable, an additional annotation was added to the genre field indicating whether the left (first) or right (second) argument of the determiner was the one that changed between the first and second sentence.
The purpose of these modifications is to better understand how well models can learn the monotonicity of determiners and to increase the balance among up+left/up+right/down+left/down+right examples.
A part of the MED dataset is collected from the published works referred to in Yanaka et al. (2019), and copyright (where applicable) remains with the original authors or publishers.