SPASM: Stable Persona-driven Agent Simulation for Multi-turn Dialogue Generation
2026-04-10 • Computation and Language
Computation and LanguageMultiagent Systems
AI summaryⓘ
The authors address the problem of large language models losing track of their assigned roles and personalities during long conversations, especially when used to create synthetic dialogues. They introduce SPASM, a framework that carefully manages persona creation, dialogue generation, and proper conversation ending to keep interactions stable. To keep each model focused on its own perspective, they propose a method called Egocentric Context Projection, which helps reduce identity mix-ups and repeated mirroring between conversation partners. Their experiments with multiple models and many generated dialogues show improved consistency and less confusion in the agents' behaviors.
large language modelsmulti-turn dialoguepersona driftrole confusionsynthetic dialoguespersona creationEgocentric Context Projectiondialogue generationconversation termination
Authors
Han Luo, Guy Laban
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. This requirement becomes critical when LLMs are used to generate synthetic dialogues for training and evaluation, since LLM--LLM conversations can accumulate identity-related failures such as persona drift, role confusion, and "echoing", where one agent gradually mirrors its partner. We introduce SPASM (Stable Persona-driven Agent Simulation for Multi-turn dialogue generation), a modular, stability-first framework that decomposes simulation into (i) persona creation via schema sampling, plausibility validation, and natural-language persona crafting, (ii) Client--Responder dialogue generation, and (iii) termination detection for coherent stopping. To improve long-horizon stability without changing model weights, we propose Egocentric Context Projection (ECP): dialogue history is stored in a perspective-agnostic representation and deterministically projected into each agent's egocentric view before generation. Across three LLM backbones (GPT-4o-mini, DeepSeek-V3.2, Qwen-Plus) and nine Client--Responder pairings, we construct a dataset of 4,500 personas and 45,000 conversations (500 personas X 10 conversations per pairing). Ablations show ECP substantially reduces persona drift and, under human validation, eliminates echoing; embedding analyses recover persona structure and reveal strong responder-driven interaction geometry. Our code is available at https://github.com/lhannnn/SPASM.