Is there no way to customize or the messages in two-stage fallback?

I have gone through the two-stage fallback part in the documentations of the latest version of Rasa 2.0, but it doesn’t touch upon how we can customize the fallback messages to be more user-friendly.

By default, the bot says something like this during two-stage fallback:

Did you mean "intent_greet"?
Yes (/intent_greet)
No (/out_of_scope)

Which is not user-friendly at all. It should be customizable to say something else like:

"Did you mean one of the following intents?:
Greetings (/intent_greet)
Something else (/out_of_scope)

There was a blog post and a video covering about this customization and how the bot fails gracefully:

However that blog post was for a very older version of Rasa, and the links it leads to have been deprecated.

I tried working my ways around to customize it, but almost 70% of the time, the customization is not able to overwrite the default fallback messages and continues to send the non user-friendly messages.

Could someone from the Rasa Team please look into this and maybe create a video or update the documentations covering this topic? Because being able to customize a two-stage fallback policy message is very essential part of Conversational AI.

Thanks!!

@koaning

1 Like

There’s no tutorial written for it (yet), but I have a repo that demonstrates this idea using the normal fallback classifier. You can find it, with a demo-project here. Does that suffice?

2 Likes

Thank you so much! That was exactly what I was looking for :smiley:

1 Like