I would like to know if prodigy could help to annotate a text in order to train a dependencies parser between multiple phrases.
Example sentence: " The cat is grey. But he shines"
I would like to train a model so that it recognizes that "he" refers to "the cat"
Prodigy has a mode for confirming dependency arcs (the
dep view), but doesn't currently have anything for specifying arbitrary arcs between words.
Overall it sounds like you want to do coreference resolution, or perhaps just anaphora resolution? You might find it better to look for pretrained models for these things. There's a plugin for spaCy, Huggingface's Neural Coref: https://github.com/huggingface/neuralcoref . This currently only supports an older spaCy version, but we're working on getting it updated. You might be able to find other tools as well.
We hope to provide full annotation support for coreference resolution in future, especially since I'm not aware of many tools to do this type of annotation.