We are creating a chatbot in Spanish and it seems the utterances are not being recognizable in spanish language, we are currently developing a training for one task that is defined.
We are working with 200 utterances and is not recognizing neither of those. The same chatbot in english version, start recognizing the intention with only 60 or less utterance defined.
With patterns works fine, but your technical always mention that is better work with utterance and only use patterns for those cases where you want to be more precise.
So how many utterance do we need to keep defining until de chatbot starts recognizing the user intentions in spanish?
It is not the number of trained utterances that determines if an intent is detected. It is the accuracy. Please verify:
Is the training done in spanish language properly.
Verify utterance testing, or NL anaysis from builder ‘talk to bot’ scoring - that’ll help us clarify the reason for spanish utterance not mapping to the intent.
Please share the information from step 2 for our review.
We are seeing this issue since we start working with Kore that utterances in Spanish are not being recognize and instead we have to work with patterns and traits, but as the documentation says defined all the utterance as you can and only use patterns for tune the intelligence.
We have been testing and it recognize the utterance doing one of this two solutions:
Create the chatbot from the scratch in spanish and import the utterances
2.- Create the chatbot in english and then enabled the spanish language and start defining utterances in the platform.
Here are some other examples from the chatbot that we created in spanish and defined the utterance
If the intent is not picked by “Ranking and resolver” it is usually due to negative patterns or traits in play. We do not see traits info in the screenshots, so there may be a negative pattern that is eliminating the intent to be picked.
You can find for sure by clicking on the ‘Ranking and Resolver’ and it will show like so,
Additionally, from the last screenshot shared, we can see that training is not completely done.
You may want to train the bot as well for accurate results.
