Rasa X interface for labeling only

I have a lot of utterances that have been collected using rasa-nlu, and fit that format. I want to use the rasa X interface to correct annotations, and then download the resultant nlu.md file.

I have tried a number of different ways of achieving this with mixed results. I found that if I put them all in the /data/nlu.md and then correct them in the training data tab, I have to label all of them for the approach to work for instance.

Is there an endpoint I could hit where I could get them to show up as if I’d added them as a new utterance under “Annotate New Data” tab?

I also tried putting the utterances into various tables in the backend sqlite db’s, but found that error-prone.

Thanks for your help!

For anyone interesting in replicating, the only way I found was this:

  • create a minimal nlu.md containing one of each entity and one of each intent from your dataset. Put it in data/nlu.md
  • put your data into the structure expected by the rasa.db table message_log, and insert it (sqlalchemy makes this easy)
  • write a script that does a ‘smart’ diff on the nlu.md that rasa x created and your minimal version from step 1, to get only the newly labeled training data.

It’d be cool if it were easier to:

  1. upload new data for annotation
  2. specify the intents and entities somewhere other than the training data.