How does the .deps file get built?

[TheDutchDevil]

I'm interested in evaluating this technique on unseen data, and therefore, I'm going through both the code and the paper to determine how the data should be formatted such that I have .txt file and a .deps file.

However, I have to confess that the structure of the .deps file eludes me. I'm not sure how the new sequence ids get determined, and how exactly the dependency information returned by for instance StanfordCoreNLP can be translated such that each sentence is converted into six arrays capturing the dependency information.

Is there any more documentation available? Or are there scripts that show how these files can be created?

Kind regards,

Nathan