Automated Dialogue Generation for Video Games

Automated Dialogue Generation for Video Games (ADGVG),” addresses the need for dynamic and context-aware dialogue in modern video games. Traditional dialogue systems, which are manually scripted and rule-based, are often repetitive, costly to produce, and lack the flexibility to adapt to player actions. This report presents a framework that integrates large, pre-trained language models (LLMs) with a retrieval-augmented generation (RAG) pipeline and prompt engineering to create coherent, engaging, and character-specific dialogues for non-player characters (NPCs). By grounding generative models in curated knowledge bases and using structured dialogue turn-taking, this approach overcomes key challenges such as maintaining personality consistency, long-term narrative coherence, and avoiding “hallucinated” outputs. Evaluation, which combines automatic metrics with human-cantered testing, demonstrates that this hybrid system produces more consistent and engaging dialogues compared to traditional methods. The ADGVG framework provides a scalable and versatile solution with broad applications in interactive storytelling, training simulations, and conversational AI beyond the gaming industry.