Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From We to Me: Theory Informed Narrative Shift with Abductive Reasoning

Jaikrishna Manojkumar Patil, Divyagna Bavikadi, Kaustuv Mukherji, Ashby Steward-Nolan, Peggy-Jean Allin, Tumininu Awonuga, Joshua Garland, Paulo Shakarian

TL;DR

This work proposes a neurosymbolic approach grounded in social science theory and abductive reasoning that automatically extracts rules to abduce the specific story elements needed to guide an LLM through a consistent and targeted narrative transformation.

Abstract

Effective communication often relies on aligning a message with an audience's narrative and worldview. Narrative shift involves transforming text to reflect a different narrative framework while preserving its original core message--a task we demonstrate is significantly challenging for current Large Language Models (LLMs). To address this, we propose a neurosymbolic approach grounded in social science theory and abductive reasoning. Our method automatically extracts rules to abduce the specific story elements needed to guide an LLM through a consistent and targeted narrative transformation. Across multiple LLMs, abduction-guided transformed stories shifted the narrative while maintaining the fidelity with the original story. For example, with GPT-4o we outperform the zero-shot LLM baseline by 55.88% for collectivistic to individualistic narrative shift while maintaining superior semantic similarity with the original stories (40.4% improvement in KL divergence). For individualistic to collectivistic transformation, we achieve comparable improvements. We show similar performance across both directions for Llama-4, and Grok-4 and competitive performance for Deepseek-R1.

From We to Me: Theory Informed Narrative Shift with Abductive Reasoning

TL;DR

This work proposes a neurosymbolic approach grounded in social science theory and abductive reasoning that automatically extracts rules to abduce the specific story elements needed to guide an LLM through a consistent and targeted narrative transformation.

Abstract

Effective communication often relies on aligning a message with an audience's narrative and worldview. Narrative shift involves transforming text to reflect a different narrative framework while preserving its original core message--a task we demonstrate is significantly challenging for current Large Language Models (LLMs). To address this, we propose a neurosymbolic approach grounded in social science theory and abductive reasoning. Our method automatically extracts rules to abduce the specific story elements needed to guide an LLM through a consistent and targeted narrative transformation. Across multiple LLMs, abduction-guided transformed stories shifted the narrative while maintaining the fidelity with the original story. For example, with GPT-4o we outperform the zero-shot LLM baseline by 55.88% for collectivistic to individualistic narrative shift while maintaining superior semantic similarity with the original stories (40.4% improvement in KL divergence). For individualistic to collectivistic transformation, we achieve comparable improvements. We show similar performance across both directions for Llama-4, and Grok-4 and competitive performance for Deepseek-R1.
Paper Structure (23 sections, 2 theorems, 21 equations, 24 figures, 2 tables, 1 algorithm)

This paper contains 23 sections, 2 theorems, 21 equations, 24 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background
  3. Related Work
  4. Technical Preliminaries
  5. Abduction.
  6. Methodology
  7. Phase 1: Rule learning.
  8. Phase 2: Iterative Abductive Transformation.
  9. Baseline.
  10. Cost Analysis
  11. Experimental Results
  12. Setup.
  13. Narrative Diagnosis.
  14. Evaluation Metrics.
  15. Implementation Details.
  16. ...and 8 more sections

Key Result

Proposition 5.1

Let $s$ be a story represented as a sequence of tokens and $\mathcal{C}_c^s$ denote the set of chunks from $s$. For each chunk $c \in \mathcal{C}_c^s$, let $\text{size}(c)$ denote the number of tokens in $c$. Then total number of LLM transformations is bounded by following quantity.

Figures (24)

  • Figure 1: Abduction-guided LLM transformation correctly identifies narrative element makes the target narrative shift.
  • Figure 2: Abduction-guided Transformation (Our Approach).
  • Figure 3: Stability of Social-Science LLM diagnostics over 10 runs for each story.
  • Figure 4: Abduction-guided LLM transformation successfully shifts narrative from I$\rightarrow$C across multiple story segments.
  • Figure 5: The KL divergence across the stories for C$\rightarrow$I narrative shift. Top left: GPT-4o, Top right: Grok-4, Bottom left: Deepseek-R1, Bottom right: Llama-4.
  • ...and 19 more figures

Theorems & Definitions (8)

  • Example 4.1: Language
  • Definition 4.1: Observations ($\mathcal{O}$)
  • Definition 4.2: Hypothesis ($\mathcal{H}$)
  • Definition 4.3: Logic Program ($\Pi$)
  • Example 4.2: Abduction
  • Proposition 5.1
  • Proposition A.1
  • proof