I have looked at the pipeline, and I understand that the token2vec is used to obtain the embedding, and then it is passed into the TransitionBased Parser component. Finally, a good result will be obtained. I'm not sure why TransitionBased Parser can be used for ner tasks. Could you explain that or provide some reference links? thank you
It feels like spaCy discussion board is better suited for your question.
This seems to be a relevant thread: spacy.TransitionBasedParser.v2 for NER: What is the Formal Description? · explosion spaCy · Discussion #11405 · GitHub
Another resource worth mentioning could be this video by Matt, where he explains the internals of spaCy's entity recognition model.