Help needed for creating a dependency gold set

Hi. I want to create an English language gold data set for dependency parsing. On the demo, I can only mark the relation labels as correct or incorrect, but cannot change them. Does Prodigy have this functionality to add, delete, or modify relations?

We don’t have that currently, no. We’re still working on a “manual” dependency annotation mode that we’re happ y with, that would allow you to annotate a whole tree. We don’t find the interface in e.g. Brat very efficient, but if the whole tree is being annotated, it’s hard to do much else.

One solution for annotating individual arcs is to set it up so that you double click on two words, and annotate a relation between them. This works reasonably well, but it means annotating each arc separately. If you can have the model suggest the word-pairs, you can also have a choice interface to select the label.

For English, the parsers are already pretty good, so I’m optimistic about a workflow that focusses on providing binary corrections, and then recalculates the parse based on those corrections. We haven’t run a large-scale test of this workflow yet, though.

1 Like