Another factor that defines a context unit of a conversation, often underrated, is the TIMING of turn-taking.
That's especially true for a real-word real-time assistant as (your) Chris navigator.
I think TIMING (mainly the human utterances time distribution) is an important attribute to be added to the list. For example if user say
Call Alicia
The bot could reply:
Shall I call Alicia A on her mobile?
Afterward a possible user silence could be managed with a reprompt or a default choice (the mobile phone) if the user do not answer in X seconds.
Or the conversation (and the domain contect) could terminate after Y seconds of silence in the middle of a turn-taking.
What do you think about it?
Thanks
Thanks Duygu for the so interesting article!
Where I would read/know more about the concept of "Semanting Frames" you introduced?
Any suggestion is welcome!
BTW, that approach is very similar to what I call "dialog units" in my NaifJs opensource dialog manager:
https://github.com/solyarisoftware/naifjs/blob/master/doc/concepts.md
Looking forward the context encoder architecture. A topic I thought a lot in past and I still not implemented in NaifJs.
Thanks
giorgio