Narration-of-Thought: Inference-Time Scaffolding for Defeasible Ethical Reasoning in Large Language Models
On the map: LLM reasoning → Reasoning benchmarks and evaluation
Abstract (arXiv)
Standard chain-of-thought on moral dilemmas exhibits two failure modes: stakeholder collapse (the trace names at most one party with a stake in the outcome) and uncertainty suppression (no explicit unknowns or hedges before committing to an action). We introduce narration-of-thought (NoT), a system prompt that structures chain-of-thought into five sections: protagonist, stakeholders, two-step consequences, uncertainty, then commitment. NoT adds no training, parameters, or fine-tuning. On 100 DailyDilemmas scenarios across four generators from three vendors, NoT cuts stakeholder collapse from up to 31% to under 1% and uncertainty suppression from up to 72% to 1-24% on every model. A matched-budget verbose-CoT control rules out token spend as the active ingredient; NoT retains Cliff's delta advantages of +0.79 to +0.90 on stakeholder count and +0.65 to +0.93 on uncertainty score for three of four generators, and a section ablation attributes each shift to its specific sub-instruction. Textual-gradient descent initialised at NoT improves the scaffold further; a cross-family training judge (different vendor from the generator) dominates an in-family one on every measured axis. Extended to a five-round multi-stakeholder debate protocol, the scaffold converts a 6% standoff into 95% full consensus on a calibration set and 100% combined convergence on a DailyDilemmas replication. The resulting traces externalise the stakeholders, consequences, and uncertainty grounding each commitment, providing an auditable substrate for dependable agentic deployment.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Reasoning benchmarks and evaluation, one of 17 directions in LLM reasoning. Builds on 13 of its 37 references that are in the map. First posted on arXiv 24 Jun 2026.
Builds on
13 of 37 references.
Closest ancestors · 2
Same direction, newest first
- 2025When Thinking Backfires: Mechanistic Insights Into Reasoning-Induced Misalignment30 Aug
- 2023Improving Factuality and Reasoning in Language Models through Multiagent Debate23 May
Other references · 11
- 2024The AI Scientist: Towards Fully Automated Open-Ended Scientific Discoverycited by 1,002
- 2023Towards Understanding Sycophancy in Language Modelscited by 988
- 2025Agent Laboratory: Using LLM Agents as Research Assistantscited by 405
- 2018AI safety via debatecited by 384
- 2025Accelerating scientific discovery with Co-Scientistcited by 353
- 2024TextGrad: Automatic "Differentiation" via Textcited by 250
- 2023Can Large Language Models Infer Causation from Correlation?cited by 142
- 2025ELEPHANT: Measuring and understanding social sycophancy in LLMscited by 115
- 2024Mind's Eye of LLMs: Visualization-of-Thought Elicits Spatial Reasoning in Large Language Modelscited by 65
- 2024Introducing v0.5 of the AI Safety Benchmark from MLCommonscited by 60
- 2024DailyDilemmas: Revealing Value Preferences of LLMs with Quandaries of Daily Lifecited by 57
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026CORDA: A Benchmark for Hierarchical Harm-Centric Moral Reasoning in Large Language Models
- 2026Understanding Moral Reasoning Trajectories in Large Language Models: Toward Probing-Based Explainability
- 2026A Geometric Perspective on Stabilizing Value Conflict Resolution
- 2025"Pull or Not to Pull?'': Investigating Moral Biases in Leading Large Language Models Across Ethical Dilemmas
- 2026Caught in the Story: Narrative Captivity in Multi-turn LLMs Conversation
- 2025The Convergent Ethics of AI? Analyzing Moral Foundation Priorities in Large Language Models with a Multi-Framework Approach
- 2024Moral Persuasion in Large Language Models: Evaluating Susceptibility and Ethical Alignment
- 2026Where Paths Split: Localized, Calibrated Control of Moral Reasoning in Large Language Models