The code to transform SoNar to Conll format neglected important information about dutchcoref HOT 1 CLOSED

The conversion script focuses on the task of producing the column with the coreference chains. This last column, together with the tokens, is the only information needed by an end-to-end model. Other kinds of models do rely on parse features.

If you need the other information, you could extract parts of it from SoNaR data, although that has no speaker information AFAIK. Another possibility is to run a parser (Alpino or spaCy). The addparsebits.py script can be used to add such information to the CoNLL file. However, IIRC, neuralcoref uses spaCy to parse sentences and does not read the parses from the CoNLL file.

Also, how can I use the code from this repo to let's say, predict Coref given a couple of Dutch sentences?

The example in the README contains instructions for that. You first run the Alpino parser and then pass its output to coref.py.

from dutchcoref.