I am currently in the process of creating a danish chatbot using the Rasa NLU framework, I am trying to avoid having to generate a super large training set, thus, the “tensorflow_embedding” pipeline is not really an option. Which leaves me with the “spacy_sklearn” pipeline, however, this requires a custom spacy model for danish, my question is then how much of the spacy functionality is used in the “spacy_sklearn” pipeline? From reading some of the Rasa NLU codebase i gather that the pipeline only uses the word vectors and the tokenizer from spacy?
Can someone confirm this? It would be nice if I didn’t need to figure out how to train a POS tagger and a parser as well.