MST-parses of Gutenberg Children corpus, with word-pair weights obtained from BERT-like DNN, different ways to calculate weights are presented as different GuCHMI versions (the folder names reflect the experiment version as GuChMI-v?).
For more info about the method of weight extraction per version, consult shalyapin@naint.ru

From BERTRAM, we obtain directional weights, i.e. each pair of words-instances is related to two different weights, depending which word is considered as the head of the link.
Different methods are used to determine each word instance-pair for our directionless parser. Each parses folder is labeled with the method used to produce it:
sum:	both directions are summed
uni: 	only the left-right direction is considered
max:	only the max of both weights is used
maxabs:	only the max of the absolute value of both weights is used
sumabs:	the sum of the absolute of both weights is used

For a selected number of experiments, the MST-parse distance mode is used in two modalities (folder names reflect this):
dist:		A shift of +1/d (where d is the separation between two word instances  in the sentence) is added to the weight
distMult: 	A factor of 1/d (where d is the separation between two word instances in the sentence) is multiplied to the weight