Narrative Landscape: Mapping Narrative Dispositions Across LLMs

TL;DR

This paper introduces a quantitative framework to profile LLM dispositions using a structured narrative task, revealing a rigidity-exploration spectrum and how instruction types influence model behavior.

Contribution

It presents a novel PCA-based visualization called Narrative Landscape to compare model selection profiles and uncovers how instructions alter model dispositions.

Findings

Models exhibit a rigidity-exploration spectrum.

Instruction types shift selection space geometry.

Scalar metrics can mask qualitative differences.

Abstract

This study proposes a quantitative framework for profiling LLM dispositions as stable, model-specific regularities in output under repeated, controlled elicitation. Using a structured narrative constraint-selection task administered across six frontier models and three instruction types, we operationalize disposition through two dimensions: "consistency", measured as cross-replication selection overlap via Jaccard similarity, and "diversity", measured as dispersion across options via the inverse Simpson index. We further introduce Narrative Landscape, a PCA-based visualization that maps each model's selection profile into a shared space for direct comparison. Results reveal a clear rigidity-exploration spectrum across model families and show that instruction types shift the geometry of selection spaces even when scalar metrics appear similar, indicating that comparable scores can mask qualitatively distinct selection topologies.