Interactive storytelling Study Guide

📖 Core Concepts Interactive storytelling – Digital narratives where the plot is not fixed; story unfolds based on user actions. Author vs. User roles – Author defines setting, characters, and situation; user drives the story through choices. Drama Manager – Central AI that monitors story variables and selects/executes coherent story beats. Agent Model – Generates possible actions for non‑player characters (NPCs) using personality + emotion models. User Model – Records player choices and play‑style to inform the Drama Manager and Agent Model. Agency vs. Structure – Successful systems balance dramatic structure (tension, conflict, resolution) with meaningful player agency. 📌 Must Remember Key variables tracked by the Drama Manager: worldwide conflict, relationship scores, tension level. Five player‑style vectors (PAST): fighter, power‑gamer, storyteller, method‑actor, tactician. Four major approaches: Environmental, Data‑driven, Language‑based, Planning‑based. Evaluation tools: Likert‑scale questionnaires (quantitative) + conversation‑centric qualitative metrics. Common AI pitfalls: mis‑interpreting sentiment (e.g., “sad” vs. clinical depression) and timing errors. 🔄 Key Processes Story Beat Selection (Drama Manager) Scan monitored variables → rank candidate beats by coherence & dramatic tension → execute highest‑ranked beat. NPC Action Generation (Agent Model) Retrieve NPC personality/emotion state → list feasible actions → filter by current story goals → present to Drama Manager. Player Style Assessment (PAST) Log recent player actions → map to the five style vectors → weight upcoming plot repairs accordingly. Planning‑Based Repair Detect potential plot hole → generate new events that restore logical continuity → insert into narrative flow. 🔍 Key Comparisons Environmental vs. Data‑driven – Environmental: emergent plot from user actions; Data‑driven: pre‑built story fragments combined on‑the‑fly. Language‑based vs. Planning‑based – Language‑based: relies on limited DSL for user input parsing; Planning‑based: anticipates and repairs plot gaps proactively. ASD vs. PAST – ASD: repairs based solely on author‑defined points; PAST: selects repairs guided by the player’s style vector. ⚠️ Common Misunderstandings “More agency = better story” – Unlimited freedom can break narrative tension; balance is essential. Assuming AI fully understands language – Systems like Façade still suffer shallow semantic grasp and timing errors. Confusing interactive storytelling with visual novels – Visual novels are a genre with fixed branching; interactive storytelling aims for open‑ended, emergent narratives. 🧠 Mental Models / Intuition “Tension curve” – Imagine the story as a roller‑coaster: tension rises, peaks at conflict, then resolves before the next climb. “Style vector as a personality lens” – Treat each player style as a colored filter that highlights certain plot repairs (e.g., a “fighter” prefers combat‑heavy fixes). 🚩 Exceptions & Edge Cases When conflict variables saturate – If worldwide conflict is already high, the Drama Manager must shift to relational tension to avoid narrative burnout. Ambiguous user input – If sentiment detection fails, fallback to generic clarification prompts rather than forcing a specific plot beat. 📍 When to Use Which Choose Environmental when you want maximal emergent storytelling and have robust NPC autonomy. Pick Data‑driven if you have a large library of reusable story fragments and need quick composition. Use Language‑based when the domain is narrow enough for a custom DSL and you want higher input coverage. Apply Planning‑based for safety‑critical narratives where plot holes must be pre‑emptively repaired. Select ASD for author‑controlled, scripted repairs; PAST when you want repairs tailored to the player’s demonstrated style. 👀 Patterns to Recognize Repeated tension spikes → likely a deliberate dramatic climax cue for the Drama Manager. Sudden NPC personality shifts → may indicate a planned plot repair (agent model adjusting actions). Player choices clustering around a style vector → triggers PAST‑specific plot branches. 🗂️ Exam Traps Distractor: “Interactive fiction = interactive storytelling” – Wrong; interactive fiction usually has limited agency and fixed branching. Mistaking “environmental approach” for “environmental storytelling” – The former refers to emergent plot generation, not the visual setting. Confusing “user model” with “agent model” – User model tracks the player; agent model handles NPCs. Assuming Likert scales measure agency – They capture overall experience; agency is better evaluated with conversation‑centric qualitative metrics. --- If any heading above seemed thin, it reflects the amount of detail provided in the source outline.