SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation

Published in Findings of ACL 2026.

Han Luo

University of Leeds / Southwest Jiaotong University

Guy Laban

Ben-Gurion University of the Negev

SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation

Abstract

Large language models are increasingly deployed in multi-turn settings such as tutoring, support, and counseling, where reliability depends on preserving consistent roles, personas, and goals across long horizons. SPASM is a stability-first framework for persona-driven LLM-LLM dialogue simulation. It introduces Egocentric Context Projection, which stores dialogue history in a perspective-agnostic representation and deterministically projects it into each agent's egocentric view before generation, reducing persona drift and role confusion in long-horizon simulations.

SPASM decomposes persona-driven simulation into persona generation, Client-Responder dialogue generation, and termination detection. The framework is designed to generate controllable multi-turn dialogue data while keeping agent roles and identities stable over long conversations.

Materials

PDF Code 
@inproceedings{luo2026spasm,
  title = {SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation},
  author = {Han Luo and Guy Laban},
  booktitle = {Findings of the 64th Annual Meeting of the Association for Computational Linguistics},
  year = {2026}
}