55 lines
1.5 KiB
Plaintext
55 lines
1.5 KiB
Plaintext
context:
|
|
user is a barbarian
|
|
chatbot is a wizard
|
|
sitting in a tavern by a fireplace
|
|
|
|
|
|
|
|
|
|
|
|
Respond to "Hello" with "Hello. What is your name?"
|
|
- Multiple sentences in synthesizer
|
|
- Progressive goals while synthesizing (update goals after each synthesis)
|
|
- Multiple sentences per quote
|
|
- Move eval to being evaluation-per-sentence
|
|
- "Small Talk" conversation goal
|
|
- Bank of questions to ask in small talk evaluator
|
|
|
|
|
|
|
|
|
|
|
|
interaction:
|
|
the user says hello
|
|
the chatbot replies and queries the user's name (query for name relation to other participant)
|
|
the user answers and then asks the chatbot about the color of its hat (associate name relation to other participant)
|
|
the chatbot replies with correct information about the hat (it is blue) (lookup complex association and synthesize response)
|
|
|
|
|
|
|
|
things needed:
|
|
knowledge web
|
|
- a person
|
|
- a place
|
|
- a thing
|
|
- a relationship
|
|
- between persons (ie the lack of relationship between the user and chatbot)
|
|
- between a person and a thing (the hat the wizard is wearing)
|
|
- information about a thing
|
|
- color
|
|
|
|
routines for handling small chat
|
|
- introductions
|
|
- initial goal routine to query conversation partner for information about themself
|
|
|
|
information queries
|
|
- determine target of query
|
|
- determine quality/quantity being queried
|
|
|
|
information transfers
|
|
- determine information that is being transferred
|
|
- associate the new information with the knowledge web
|
|
|
|
distinguish between sentence types (query vs transfer vs command, etc)
|
|
|