Python scripts:
* jolli.py - recreates the experiments from the paper
* fix.py - converts the format of the early Grefenstette & Sadrzadeh datasets to match that of the Kartsaklis et al dataset (run this prior to jolli.py)
Datasets available from:
* G&S 2011: http://compling.eecs.qmul.ac.uk/wp-content/uploads/2015/07/GS2011data.txt
* G&S 2012 (original - not used in these experiments): http://compling.eecs.qmul.ac.uk/wp-content/uploads/2015/07/GS2012data.txt
* G&S 2012 (extended & re-annotated by G&S 2015): http://www.cs.ox.ac.uk/activities/compdistmeaning/GS2013data.txt
* KSP 2013: http://compling.eecs.qmul.ac.uk/wp-content/uploads/2015/07/KS2013-CoNLL.txt
Vectors & tensors available from:
* noun vectors: https://ln.sync.com/dl/e93a0a060/xm5njmyd-hqvbgw5d-cdmviv9q-6ye8yrsb
* verb tensors: https://ln.sync.com/dl/88a20c1c0/uh9friau-7tvb4mmu-nukuty3f-yzw92yuh
* (and see https://github.com/gijswijnholds/compdisteval-ellipsis for more options to experiment with!)