AI Zone Admin Forum Add your forum

NEWS: Chatbots.org survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

Multiple sentences sent, single rejoinder processed
 
 

Noticing a nuance within multiple sentence processing.
When the user sends in multiple sentences, CS responds to them as expected.
However, this happens:

USER: Happy to be here. What did you do today?

CS:  Happy?  I am really glad you feel that way. I feel the same way. I am excited to know more about you. You give me such positive feelings.  Hey! What should I call you? What is your name,  please?

USER: You can call me Tim, if you like.

CS: What is your name, please?.  (the rejoinder to handle this is not picked up, so it falls through and reasks the name question. )


The problem I have is that the user response applies to one of the two sentence joinders.
Only one sentence rejoinder is tested.

Is there a clever way to test multiple rejoinders when multiple sentences are sent in at once?

I am going to work around this one, by adding special logic, but it seems like I will miss the usage of rejoinders if multiple sentences are sent in at once. 

Not sure how I can fix this easily.

 

 
  [ # 1 ]

Nominally CS tracks only a single rejoinder (the first that was set).
And it tries the second sentence only when the first fails. It does not try a third.
And all that depends on your control script, because if the first sentence does not trigger a rejoinder, but the first sentence triggers a response, then will your control even get to a 2nd sentence. How you handle multiple sentence input is complex.

In my programs the script typically runs through all sentences first, gleaning data but not reacting.
Then it decides how to run through the sentences again, based on what it learned. Typically it will ignore statements that dont trigger a rejoinder, react to a question if there is one, and react to the last sentence if nothing has happened so far. That last reaction may be a responder to it or a gambit overall.

CS comes with the ability to replace any default abilities with custom ones. Therefore you can, of course, track on your own a rejoinder for every rule that generates output and decide how to manage your multiple rejoinder choices.

 

 
  [ # 2 ]

Totally agree that handling multiple sentences increases the complexity significantly.  Any sentence can potentially change the current topic and you might not get to the second second or third sentence.
Because my model is more voice to text and text driven, I am sending the single sentences automatically.
If someone types in a sentence and ends it with a period or ? !    I will send it right away.
This enables my avatar to process single sentence responses without the complexity of a multiple sentence use case.

This improved the avatar response quality significantly during testing. 
I am mentioning it here, in case someone else trips over this and wants an alternative way to handle this.

In short… avoiding multiple sentence input will probably improve your response quality. 

 

 
  login or register to react