Webchatbot does not detect language/not recognize text analyzer rule
Hi all,
I am trying to configure a webchatbot to let it detect the language of the user typing in the chat window.
In order to do so, i have created the webchatbot channel and then configured the Text Analyzer section in the behavior tab, adding a text analyzer of type iNLP.
Within the Prediction Studio, I have also created a Text Categorization model in which i have defined topics for english and italian languages.
Then, switching back in Dev Studio, i referenced this model within the Topic Detection area of the aforementioned Text Analyzer rule.
At this point, i tested the chatbot from the Preview Console and i receive always the "Sorry I did not understand that" answer.
Is there something i am missing to configure?
Can you please help?
Thanks,
Vince
SoftServe
US
Hey Vince,
Did you run the localization wizard first?
I created a discussion with a similar issue to yours here:
https://collaborate.pega.com/discussion/pega-chatbot-localization-and-mashup-embed-testing
Regards
-Jose
Updated: 15 Jun 2020 3:46 EDT
Accenture
IT
Hi Jose,
i ran the localization wizard but this has not solved the issue.
By the way, i have noticed that, if we use iNLP type, Pega recognize language and does the other stuffs only if you configure other rule like data flows, strategies and so on ( please look @ https://community.pega.com/video-library/pega-83-deep-dive-whats-new-decisioning-text-analytics)
Otherwise, you may use
- Pega NLP -> which works but it seems that you cannot record traning data
- Pega NLP Generated -> which works and let you record traning data but Pega Help reports this type as Deprecated
Now i started using Pega NLP Generated but i have found other issues
1) Reviewing Training Data -> You can review training data but the count of "Reviewed" stands still to zero and i am not able to rebuild the model (the voice in the menu appears always disabled)
2) I do not know if training data for italian language, reviewed within the webchatbot channel UI, is recorded automatically in the right language model, as we do not have any field where to specify for which language the records apply to.
I have seen your discussion but no one else contributed. Have you discovered something new?
Thanks,
Vince
SoftServe
US
I agree with number 2:
In the Chatbot's Behavior tab, I opened the text analyzer rule and then the Topic model. Once there (and using Prediction Studio) I added a Keywords based model for Spanish language. However when I return to the chatbot preview for Spanish, and when the chatbout could not recognize the topic, I help it trying to define it, but it does not show the topics I defined for it, only the English ones... why it didn't recognize the topic in the first place :/
Thanks
Pegasystems Inc.
IN
Even though a chatbot can respond in different languages, provided the language does not change in the middle of the conversation, Webchat channel does not have an explicit multi language support. From within Channels tab, you cannot train topics in multiple language. While manually adding language to taxonomy, following are some things you should keep in mind -
- Keep topic names same in all languages. Topic name is an internal construct, not exposed to user so topic name - Order pizza will be same in both English and Spanish for instance.
- Add training data for each language separately -
- For English, you will be able to add matching words and training data from within Channels tab
- For other languages, add models for each. Again, keep topic name same, words and sentences can be in different languages.