There are times where AI waits for the user to say a second thing, doesn't get it, and then aggressively pushes ahead to the next bot utterance. Sometimes this is fine, but other times it is incoherent, referring to something never said. Can we do some analysis of possible event structures to determine which observations are safe to skip, and use fillers otherwise?
Example:
Whew, I'm glad that's over for another month.
ha oh yea?
<expects "how do you think it went?">
Thanks. It's always easier when the numbers are good.
Closing. This is now implemented, and works. We will have to keep our eye out in the Turk tests to decide if it is too restrictive, leading to too many fillers.
There are times where AI waits for the user to say a second thing, doesn't get it, and then aggressively pushes ahead to the next bot utterance. Sometimes this is fine, but other times it is incoherent, referring to something never said. Can we do some analysis of possible event structures to determine which observations are safe to skip, and use fillers otherwise?
Example: Whew, I'm glad that's over for another month. ha oh yea? <expects "how do you think it went?"> Thanks. It's always easier when the numbers are good.
See line 43 of: https://tools.giantotter.com/logviewer/labels/staging.Coach%20Otto.Webchat%20(Turk).ai.37-2.olivia/sessions/bb585ca8c94d4c4cbca4bce26e9ea943?env=dev