I am trying to implement a multilingual chatbot using RASA NLU. Concretly, I'm designing a separate chatbot for each language (with its own translated and adapted training data). However, it would convenient for the application side if the names of the intents could stay the same between all the chatbots. In that sense, my question is:
Is the name of the intent important for the intent classification when using DIET Classifier ? In other words, is the name of the intent used though the word embedding network or is it just a "label" that has no signification ?
Thank you for your help
It does not matter what the name of the label is when it comes to accuracy.
If you rename the greetings intent to omega_jabberwocky consistently in all your configuration files you'll only change the name of the intent.
That said, it's highly recommended to come up with meaningful intent names because otherwise collaboration and maintenance become much harder.
In your use case in the multi-language setting ... you could use the intent to cause a custom action to trigger, which then responds with a custom message depending on a detected language (which may be saved in a slot, let's say). If you're interested in multi-language bots you may enjoy this talk on Youtube. It demonstrates a workflow for a multi-language setting.