This leads to the classic Interactive Fiction problem – the actual space of possible interactions is quite small (at least by comparison). I can type a lot before striking upon something that “works”.

Is this really a problem, classic or otherwise? When I play IF or see other people play IF, it seems to me that about 90% of the things that are typed “work” (are valid commands), and most of the other 10% are typos. However, in IF, you can tell if your action is being understood (but perhaps isn’t effective) or isn’t being understood: >LOOK UNDER BED results in “There is nothing but dust there” whereas >FLIP MATTRESS OVER results in “That’s not a verb I recognize.” In Façade, it may be more difficult to tell which is the case. This may or may not be a problem, as we’re not trying to manipulate and explore the world in the same way as in conventional text-based IF. Another difference: people learn IF conventions by interacting for a while, and this carries over to other IF interactions.

The coach concept is a good one – Eric Solomon discusses the computer as a coach or “second” in his 1984 Games Programming, but I would guess that the idea hasn’t been very widely explored in this way outside of hint systems and chess programs that dispense advice. However, I don’t think this approach is right as the framework for Façade, and do I think there are many nice things about an interaction that is symmetrical in some ways: I reply in English to what characters tell me in English. Instead of offering a “langauge panel” full of word-buttons as an interface, why not use the interface of language that we are already pretty skilled at using when we interact with people?