Big one this, very exciting. To make it more approachable I've made a proposal based around a small-scale interaction (so we have somewhere to start), and then below I've made some points about dialogue in general to keep in mind where we want to go
Starting Scenario
A dialogue is created in a similar way to building to a scene (#36), except with an additional step to generate the "social context" of the dialogue. At a basic level this should label the Participants and a history of messages between them
Take for example that I initiate a dialogue by interrupting a street preacher. A Dialogue is then created between myself and the preacher, and we both start with one goal:
I have chosen my goal by selecting an option for dialogue
The street preacher, interrupted, wants to talk to me about a radical new sect of Piscalism
The preacher may reply differently depending on my choice:
I asked about Piscalism - the happy preacher can pursue our mutual goal immediately
I asked the preacher's name - the preacher will accommodate my goal because it doesn't contradict their goal, but next time they may decide to pursue its own goal instead - "well met sir, I am Ivan of the Godly Parish"
I said something strange that the preacher didn't understand, or that contradicted their goal, for example asking them to a game of football - "the preacher stares at you, bewildered". They decide to pursue their goal anyway. "Have you heard the word of God?"
The "goals" here are the agents' short-term goals in narration and are different from a larger scale goal like world domination
Notes on AI
the game state may change such that the AI chooses to adopt a goal of avoiding dialogue, for example if I offended them
the use of goals could be useful in Machine Learning, but we may find that the use case of that is more in tune with developing scripts for the various stages (e.g. goal selection)
The complexity of "what to say and how to say it" can be heavily reduced by choosing a narration of what is being said, not what is literally being said. E.g. "The character says hello and introduces themselves as Hercules"
Limiting the dialogue to two participants
Only the player is able to initiate dialogue. I think that the NPC initiating a dialogue isn't very different from other forms of asynchronous events
Character building with dialogue
Later, we want dialogues to have "side-effects" on the game state
Relationships and memories
When the preacher presents me with the question, I could choose to make my character a convert or staunchly atheist. Perhaps the player poses a question about the NPC that it doesn't know, so needs to make a quick choice and write that to its character
Contradiction will be a difficult challenge, both in the sense of the player picking up on the AI contradicting itself (maybe after it's written some new development to itself) and in terms of the player saying contradictory things to the AI
Narration
I think that we could expand the social context in the dialogue to encompass other things, for example world state (this is a dangerous situation) and narrative goals which can steer the direction of conversation
Big one this, very exciting. To make it more approachable I've made a proposal based around a small-scale interaction (so we have somewhere to start), and then below I've made some points about dialogue in general to keep in mind where we want to go
Starting Scenario
A dialogue is created in a similar way to building to a scene (#36), except with an additional step to generate the "social context" of the dialogue. At a basic level this should label the Participants and a history of messages between them
Take for example that I initiate a dialogue by interrupting a street preacher. A Dialogue is then created between myself and the preacher, and we both start with one goal:
The preacher may reply differently depending on my choice:
The "goals" here are the agents' short-term goals in narration and are different from a larger scale goal like world domination
Notes on AI
Notes on reducing complexity
Character building with dialogue
Later, we want dialogues to have "side-effects" on the game state
Contradiction will be a difficult challenge, both in the sense of the player picking up on the AI contradicting itself (maybe after it's written some new development to itself) and in terms of the player saying contradictory things to the AI
Narration
I think that we could expand the social context in the dialogue to encompass other things, for example world state (this is a dangerous situation) and narrative goals which can steer the direction of conversation