In this repository, you can find the datasets used to train and evaluate ROOT9, as described in Santus et al. (2016c).
It includes the 9600 pairs (and the 12800 ones, which also include switched hypernyms: i.e., dog RANDOM fruit) randomly extracted from EVALution (Santus et al., 2015), Lenci/Benotto (Benotto, 2015) and BLESS (Baroni and Lenci, 2010).
On top of our datasets, we also provide the subsets of the WN_Hyper, WN_Coord, BL_Hyper and BL_Coord from Weeds et al. (2014).
Please, read the attached paper (Santus et al., 2016 - LREC) for more information.
Acknowledgements
We are very grateful to Julie Weeds for having helped us, recalculating the results of Weeds et al. (2014) models also for our subsets of their datasets. Thanks also to Aristotelis Kostopoulos for the precious suggestions.