project to preprocess spancat datasets into .spacy

[kadarakos]

This is a project that is meant to help with experimenting with the spancat and ner components. Currently it has:

1. ConLL (English, German, Spanish, Dutch)
2. WiniNeuRal (English, German, Spanish, Dutch)
3. Dutch Archeology data set
4. AnEM
5. Wnut 2017

For the archeology data set it creates a random train/dev/test split and for all other data sets it uses the standard splits. It has a small utility view_spans.py to show a couple of random examples on the command line from a data set and runs debug data with the a default spancat config to show the span-characteristics.

The idea is to have a home for the spancat/ner data sets to be able to easily benchmark components on them. Plus others can use these data sets perhaps for pre-training.

[kadarakos]

The project currently lives here https://github.com/explosion/spancat-datasets/, but there is no reason not to make it public and its nicer if its in the projects in any case as we discussed.