Where to look for a solution
The first problem is simply one of building a system rather than designing a series of puzzles and clues. The second is a problem of gameplay-narrative divide. We want there to be an actual content to the conversation, and that requires it to be written down in advance, because AI just isn't good enough. The trick is to find rules that fit the content.
Thankfully, all the real work has been done for us. By rhetorics, conversation analysis, politeness theory and various incarnations of interactionism. These people have recorded, studied, classified the already highly game-like interaction that is argumentation and conversation.
As it turns out, it has a clear set of mechanics:
- a few modes of action, such as the classical ethos (authority), pathos (emotion) and logos (logic) [Aristotle on the SEP], or in a more, er, free-spirited interpretation, Revelation, Imagination and Explanation.
- a few distinct resources: the positive and negative face [Brown and Levinson], the mental and affective state of the speakers and listeners, their knowledge of each other's beliefs,
- codified rules for the interaction: politeness codes, turn-taking [Hirsch], conversational maxims [Grice].
Sure, we will have to leave out a lot of the meat of these theories as we turn them into game rules. But at least, we have a battle plan: write a map containing many possible ideas and arguments relating to a central topic, which will be the territory on which the conversation is "fought", and then implement mechanics for all three aspects above.
In brief, the map will be the semantic content, and what we do on it will have rules, enabling control and strategy.