1

I'm new to RASA framework. I'm trying to develop a Spanish NLU model capable of classifying a user message in 4 different intents ("translation", "definition", "synonym" and "pronunciation") and recognising entities (in this case, my entities could potentially be every word or expression).

So for example, the user could give the following input:

"Cómo se traduce estación de tren al inglés" (How to translate train station into English)

So the word or expression could be any word or words and they could have any length. What's the best way to solve this problem? What entity extractor could I use and what parameter should I try?

Thanks in advance!

Alberto Sánchez Ruiz
  • 95
  • 1
  • 9

1 Answers1

1

You cannot expect that any unknown sentence or words will be properly interpreted by RASA. You are responsible to train AI with a specific Model contains exact sentences and RASA will give you some probabilities whether something similar is asked.

You should define and categorize general sentences that might be asked:

Example 1:

  • Where is the {train station}?
  • How to find a {train station}?

Results:

  • Action: location
  • Variable: {train station} (can be anything here)

Example 2:

  • How to translate {word}
  • I would like to translate {word} in {English}
  • How would you say in {English} {word}
  • Translate {word} {English}

Results:

  • Action: translation
  • Variables: {English} {word}

RASA will work for you if you will be able to categorize all possible cases of the sentences thatmight be asked

Ievgen
  • 4,261
  • 7
  • 75
  • 124
  • I've already filled my nlu.md file with several examples of each intent (and tagged the entities that appear in each examples), about 50 examples of each intent. I'm trying with DIETClassifier, but it's not really working as expected. – Alberto Sánchez Ruiz Aug 14 '20 at 09:10
  • You should cover all possible cases, otherwise it will be not presice, try to use variables also and all the recommendations from their documentation. – Ievgen Aug 14 '20 at 18:53