Dialog as a Game

Peter Border's paper, presented at the 2005 DiGRA conference, proposes treating in-game conversation as a rule-governed game in which speakers make moves by transitioning between short dialog "chunks" rather than traversing ever-larger decision trees. The central claim is that this framework yields more socially and emotionally coherent non-player characters, and that stronger NPC dialog is critical to broadening games' appeal and undermining players' meta-gaming strategies, such as exhausting a dialog tree by repeatedly restarting conversations.

The proposed system centers on internal state and relationship variables. Characters carry simplified emotional states (happy/sad, afraid/confident, angry/peaceful, excited/calm, drunk/sober), while relationship variables track closeness, face, liking or irritation, and trust, with the conversational "ball" passing between speakers as face is gained or lost. These values decay toward zero over time unless refreshed. Each character evaluates progress through a payoff function, and chunks are offered only when their requirements are met. A simple knowledge mechanism, based on keyword-value pairs and test strings, lets earlier dialog gate later lines.

The work is explicitly preliminary. The rules were derived by annotating several short plays in a spreadsheet, and the framework was tested through index-card improvisation trials on a single scenario, "The Fighter and the Barmaid," involving roughly 50 chunks. The author reports dozens of exchanges proceeding naturally but offers no quantitative evaluation of naturalness, and a Java implementation was still future work at the time. The paper also suggests applications beyond games, including online sales, help desks, and tutoring.