Prediction: Sentence or words?

Given the INTENT and all other STATE information how does RASA predict the response …

Does it predict full sentence as a single unit OR as separate words that combined create a sentence ?

Can you elaborate ?

Rasa doesn’t produce new sentences, it chooses the best one out of the predefined responses. If you want to read in more detail check out [1811.11707] Few-Shot Generalization Across Dialogue Tasks and [1910.00486] Dialogue Transformers

thanks thats what i was looking for… i.e. how does Rasa work … if you have more details on that let me know

As I’m reading the papers a question comes to mind … How would I go about if I want to featurize along with the intent and entities also POS tags.

In my case Subject-Verb-Object. Let say I have a way to programmatically extract them via spacy OR pattern.en library