What feature representation does the Prodigy artificial neural network use for annotation?

I trained a named entity recognition (NER) module using Prodigy to annotate text with labels, and I have been told on this forum that it uses an artificial neural network (ANN). (There was a training-testing split.) So here's a question I was asked and need help with - What kind of feature representation does that ANN use? I assume this means something related to the features of text that the ANN depends on to assign labels, but I'm not experienced with the term "feature representation" as it applied to ANNs, so I'm not sure.

This video is a bit dated, but it does explain how (some of the) spaCy models works internally.

Does this help?