Wordplay: When Language Meets Games @ EMNLP 2025

In our team at Microsoft Research, we have been exploring how generative AI can transform narrative design games. In the first part of the talk, I will introduce GENEVA, a prototype tool for generating and visualizing branching narratives using large language models (LLMs). This tool allows narrative designers to develop and iterate narratives much faster than traditional tools. In the second part, I will present a case study on how generative AI can evolve game narratives in real-time based on player feedback. The goal of this work is to understand how players, game creators and generative AI can collaboratively create and evolve a game’s narrative over time. Learn more about our team’s work on our website – https://msr-emergence.com/.

Recent advancements in Large Language Models (LLMs) have showcased their impressive capabilities in various language tasks. However, traditional evaluation methods often fall short in capturing the full spectrum of LLMs’ abilities, particularly in interactive and dynamic contexts. This talk explores the limitations of conventional benchmarks and proposes a shift towards evaluating LLM agents through interactive collaboration and competition. By engaging LLMs in multi-agent environments, we can better assess their adaptability, problem-solving skills, and social intelligence. This approach not only provides a more comprehensive understanding of LLM capabilities but also paves the way for developing more robust and versatile AI systems.