I’m just getting started with Rasa, and wondering whether it is possible to connect Wit.ai directly with Rasa. Meaning, every message that comes from fb messenger is already parsed to entities by built-in wit.ai as NLP. Can Rasa simply use wit.ai extracted data dictionary for intents and entities? I would like to use Rasa just for flow and persistency management basically. I’m aware I can export wit.ai data into Rasa’s NLU, but that’s not what aiming for. I found a wit.ai emulator in code, but not sure of its purpose.
Hi @scotbond welcome to the Rasa community! Yes I think it should be possible to use wit.ai as your NLU and only use Rasa for dialog management. It’s nothing to do with the wit.ai emulator though. You can take a look at some of our API endpoints on how to do this. E.g. you can:
- Add a message to the tracker with the info from wit with this endpoint
- Then have Rasa Core predict the next action with this endpoint
Can I ask what the reason for sticking to wit’s NLU is though?
Hi Akela, thanks for your detailed answer! I’ve decided to give Rasa NLU a go, and after few days feels like finally i’m on track filling the requirements I have. Specifically it was quite difficult to understand how to extract quantity of things - i.e “i’m looking for a five bedroom apartment with 2 bathrooms” I’m using Duckling to extract numbers but it was unclear how to train the model to understand a certain number is related to certain entity (wit.ai calls it Roles). by looking at your Restaurant bot example (with number of people extraction), I could finally figure this out. Definitely worth explanation in the Docs. I saw another thread in the forums asking about Roles.