Personality Expression Across Contexts: Linguistic and Behavioral Variation in LLM Agents

TL;DR

This paper investigates how identical Big Five personality prompts for LLM agents produce distinct linguistic, emotional, and behavioral expressions across four social contexts. Grounded in Whole Trait Theory, the authors systematically vary context via ice-breaking, negotiation, survival decisions, and empathetic dialogue to observe context-aware modulation. They combine LIWC analysis, a pre-trained trait classifier, and an LLM-based personality evaluator with emotion inference and behavioral metrics (Agreement and Concession) to characterize expression across dimensions. Findings show that trait expression is strongest in cooperative contexts and modulated by social goals and affect, supporting a view of LLM personality as context-sensitive adaptation rather than fixed consistency.

Abstract

Large Language Models (LLMs) can be conditioned with explicit personality prompts, yet their behavioral realization often varies depending on context. This study examines how identical personality prompts lead to distinct linguistic, behavioral, and emotional outcomes across four conversational settings: ice-breaking, negotiation, group decision, and empathy tasks. Results show that contextual cues systematically influence both personality expression and emotional tone, suggesting that the same traits are expressed differently depending on social and affective demands. This raises an important question for LLM-based dialogue agents: whether such variations reflect inconsistency or context-sensitive adaptation akin to human behavior. Viewed through the lens of Whole Trait Theory, these findings highlight that LLMs exhibit context-sensitive rather than fixed personality expression, adapting flexibly to social interaction goals and affective conditions.

Figures (6)

• Figure 1: Overview of the experimental framework examining context-dependent personality expression in LLM-based dialogue agents
• Figure 2: BERT-based Pre-trained Personality Prediction Model Result kazameini2020personality
• Figure 3: LLM-based Personality Evaluation Result
• Figure 4: Distribution of Valence–Arousal across four dialogue contexts
• Figure 5: LLM-based personality difference across dialogue contexts. Higher values represent stronger trait differentiation (High vs. Low).