Narrative Humanization in AI

• Narrative humanization is a process that restores individual subjectivity and emotional depth in storytelling through human-AI collaborative methods.
• Methodologies such as co-creation workshops, tactile expression, and persona engineering enhance narrative agency and cultural resonance.
• Empirical studies demonstrate that structured narrative design boosts user empathy and authenticity in computational storytelling.

Narrative humanization encompasses a set of theoretical and practical methodologies aimed at foregrounding the subjectivity, agency, and emotional realities of individuals in narrative contexts—especially when mediated, assisted, or synthesized by computational systems. It is primarily concerned with restoring human depth to stories that would otherwise be marginalized, fragmented, or de-personalized by age, technology, scale, or abstraction. Recent advances position narrative humanization as central to both computational storytelling and human-AI interaction, invoking a blend of narrative theory, design research, machine learning, and empirical methodology.

1. Conceptual Foundations and Definitions

Narrative humanization is defined as the process of restoring and foregrounding the subjectivity, emotional depth, and agency of individuals—particularly those marginalized by age, digital literacy, or linguistic barriers—by enabling them to tell, visualize, and materially embody their own stories (Zhan et al., 2 Jul 2025). Theoretical roots include Bruner’s performative theory of narrative (emphasizing reframing and revaluing experience) and participatory design, which prioritizes the externalization of internal states and negotiation of social meaning.

In computational contexts, narrative intelligence is the broader umbrella: the set of abilities to craft, tell, understand, and respond affectively to stories, incorporating functions of narrative understanding, generation, affective response, and narrative telling (Riedl, 2016). Humanization, in this computational sense, is operationalized not only through content but through style, agency, persona, and the intricate mapping of lived experience onto narrative form.

2. Methodologies and Human-AI Collaborative Mechanisms

Narrative humanization in human–AI systems is architected through interleaved workflows that foreground participant control, cultural symbolism, and tangible engagement.

• Co-Creation Workshops: Procedures involve oral storytelling facilitated by human mediators, transcription and dialect-sensitive paraphrasing, and symbolic mapping of narrative content onto culturally meaningful signifiers (e.g., Xiaozhuan Hanzi glyphs) (Zhan et al., 2 Jul 2025).
• Soft AI Presence: AI’s role is reconfigured from autonomous generator to backstage semantic support. LLMs operate as suggestion engines, offering symbolically evocative prompts (e.g., glyphs or narrative arcs) without direct, personified interaction—thus reducing cognitive load and performance anxiety, while preserving narrative agency.
• Tactile and Material Expression: Physical composition (using foam boards, clay, etc.) serves as an alternative or complement to verbalization, operationalizing embodied narrative and enabling expressive depth beyond linguistic or digital constraints.
• Persona Engineering in Narrative Agents: In longitudinal GenAI systems, persona design and prompt engineering ensure a consistent narrative voice (e.g., the daily “Makoto” character), with user inputs and iterative feedback loops adapting output while sustaining identity and co-authorship dynamics (Fabre et al., 20 May 2025).
• Human–AI Role Delineation: Large-scale narrative synthesis (e.g., in civic contexts) is achieved via pipelines in which AI performs scalable decomposition (scene/thematic extraction and clustering), while humans curate, edit, and validate the composite narratives for authenticity, voice, and community resonance (Overney et al., 23 Sep 2025).
• Empathic Modeling: Retrieval and generation of stories are calibrated by optimizing for empathic resonance (main events, emotional trajectories, morals/takeaways) rather than surface-level semantic or lexical similarity, using transformer-based encoders fine-tuned on high-quality affective-annotation datasets (Shen et al., 2023).

3. Character, Agency, and Emotional Trajectories

Foregrounding agency and emotional complexity is a core aim of narrative humanization.

• Agency: Emerges in systems that allow participants not simply to select from, but to reinterpret, decompose, and recombine AI-derived prompts. Artifacts—such as new Hanzi forms—become material traces of personal meaning, even where verbal storytelling falters (Zhan et al., 2 Jul 2025).
• Dual-Axis Arc Analysis: Quantitative analysis of character change in personal or testimonial narratives can disentangle “inner” (belief) from “outer” (practice or action) arcs, enabling the identification of archetypal trajectories (constant positive, oscillating, ascending, etc.) and the design of storyboards that emphasize moral depth and lived contradiction (Shizgal et al., 2024).
• Empathic Similarity: Narrative humanization is reinforced by focusing on affective-aligned retrieval and generation. Embedding-based measures integrating event, emotion, and moral similarity yield retrievals that foster stronger user empathy, associative identification, and social connectedness (Shen et al., 2023).
• Style and Empathy Pathways: The HEART framework decomposes narrative style into components (character development, plot volume, vivid emotional subjectivity, first-person POV, and setting) most strongly associated with reader transportation and resultant empathy. Empirical modeling confirms that increased vividness and plot complexity, as well as personalization to reader traits, systematically enhance empathic engagement (Shen et al., 2024).

4. Architecture, Evaluation, and Metrics

Narrative humanization necessitates both system-level architectures and robust evaluation.

• Workflow Stages: Typical pipelines include input preprocessing (scene/theme decomposition), thematic creation and clustering (with multi-pass LLM agreement to avoid drift), story scaffolding (composition based on balanced concrete/interpretive cues), and rigorous human validation. These are instantiated in systems like StoryBuilder for civic engagement (Overney et al., 23 Sep 2025).
• Metrics: Empirical validation leverages both subjective and behavioral measures. For narrative agency and presence, self-report scales targeting trust, respect, cognitive/affective/associative empathy, and sense of agency are used (e.g., Likert scales, State Empathy Scale, SoA/SoO protocols) (Hamagashira et al., 26 Dec 2025, Shen et al., 2023). Field analytics—session duration, navigation metrics, engagement with citations—provide quantitative backbone for deployment evaluation (Overney et al., 23 Sep 2025).
• Model Evaluation: Cosine similarity of fine-tuned story embeddings for empathic similarity, BLEU/ROUGE/METEOR for narrative summary quality, and cluster coverage for character-arc representational diversity are established practices (Shen et al., 2023, Shizgal et al., 2024).

Humanization Dimension	Core Method	Representative System
Agency & Materiality	Tactile co-creation	Hanzi Workshop (Zhan et al., 2 Jul 2025)
Emotional/Ethical Depth	Persona Design	More-than-Human Storytelling (Fabre et al., 20 May 2025)
Empathic Resonance	Affective Retrieval	EmpathicStories (Shen et al., 2023)
Scale & Authenticity	Hybrid Synthesis	StoryBuilder (Overney et al., 23 Sep 2025)
Character Complexity	Arc Decomposition	Trajectory Analysis (Shizgal et al., 2024)
Style-Driven Empathy	HEART Taxonomy	HEART-felt Narratives (Shen et al., 2024)

5. Empirical Case Studies and Prototypical Patterns

Real-world deployments reveal concrete patterns in narrative humanization.

• Hanzi Co-Creation: Elderly migrant participants converted fragmented migration memories into physical glyphs, leveraging AI-suggested symbols. Key outcomes included enhanced narrative coherence despite linguistic fragmentation, increased narrative agency, and the creation of hybrid artifacts that anchored abstract emotion in cultural form (Zhan et al., 2 Jul 2025).
• Longitudinal Storytelling: Repeated daily interaction with a persona-rich AI agent (“Makoto”) produced themes of interpersonal resonance, oscillating ambivalence, and gradual socio-chronological bonding. However, limitations around narrative control and transparency signaled persistent challenges in ensuring participant authorship and engagement (Fabre et al., 20 May 2025).
• Civic Narrative Synthesis: The hybrid Human–AI pipeline in StoryBuilder demonstrated that experience-heavy (scene-based) composite narratives induced significantly greater trust and respect than theme- or opinion-heavy summaries, though rapid stance change was rare (Overney et al., 23 Sep 2025).
• Character Archetype Distribution: Holocaust testimony analysis surfaced empirically that most belief arcs were constant positive, but practice often oscillated; cross-cluster analysis suggested archetypes useful for foregrounding resilience or internal conflict (Shizgal et al., 2024).
• Narrative Framing in VR: Positive narrative context for avatars increased sense of personal agency and familiarity, demonstrating that top-down narrative framing in immersive environments yields measurable effects on embodiment and perceived self-control (Hamagashira et al., 26 Dec 2025).
• Empathy by Story Style: Structural equation modeling confirmed that emotional vividness and plot richness mediated narrative transportation and empathy, establishing a predictive basis for crafting humanized stories via narrative style manipulations (Shen et al., 2024).

6. Limitations, Open Problems, and Sociotechnical Implications

Despite progress, narrative humanization faces substantive open challenges.

• Agency-Transparency Tradeoff: The delegation of narrative structure to AI agents risks obscuring authorship and undermining user empowerment. Adjustable autonomy, explainable prompt structures, and human-in-the-loop checkpoints are proposed mitigations (Fabre et al., 20 May 2025).
• Evaluation Complexity: There are no universally accepted quantitative metrics for story quality, coherence, or empathy. Most work relies on subjective scales or fine-grained behavioral analytics; generalization across domains remains problematic (Riedl, 2016).
• Corpus Diversity and Bias: Obtaining broad, representative corpora remains a bottleneck. Systems trained on narrow narrative sources risk yielding exclusionary or biased outputs; continuous curation and ethical review are critical (Riedl, 2016, Shen et al., 2024).
• Manipulation and Ethics: As empathic or style-optimized narratives can be leveraged toward persuasive or manipulative ends, transparency, participant consent, and safeguard mechanisms are necessary elements for responsible deployment (Shen et al., 2024).
• Complexity and Generalization: Computational models still fall short in handling figurative language, long-range dependencies, and genuinely novel story creativity (Riedl, 2016). Cross-modal and cross-cultural generalization, and the balancing of abstraction with individualized depth, remain important frontiers.
• Accessibility: Humanizing narrative workflows that minimize digital, cognitive, and linguistic barriers—by centering tactile/material engagement and culturally resonant symbols—demonstrate continued relevance for equitable design (Zhan et al., 2 Jul 2025).

7. Design Guidelines and Prospects

Empirical research converges on a set of design recommendations to maximize narrative humanization:

1. Center Narrative Agency: Prioritize participant choice, interpretation, and meaning-making; avoid over-scripted or one-way AI outputs (Zhan et al., 2 Jul 2025).
2. Employ Soft AI Support: Maintain AI roles as suggestive or supportive rather than substitutive; backend-only operation preserves human primacy (Zhan et al., 2 Jul 2025, Overney et al., 23 Sep 2025).
3. Integrate Multi-Modal Expression: Support tactile, visual, and sensory forms of storytelling to bridge digital literacy divides (Zhan et al., 2 Jul 2025, Shizgal et al., 2024).
4. Leverage Narrative Structure and Style: Employ frameworks such as HEART to systematically enhance empathy, trust, and transportation via stylistic modulation (Shen et al., 2024).
5. Deploy Hybrid Human–AI Editing: Clearly delineate AI and human roles in synthesis to balance scale with authenticity and trust (Overney et al., 23 Sep 2025).
6. Personalize By Empathic Similarity: Retrieve and generate narratives that prioritize affective resonance; align recommendations to user-reported emotional profiles (Shen et al., 2023).
7. Enable Transparency and Control: Surface prompt structures, enable autonomy adjustments, and provide user-facing explanations to build trust and sustain engagement (Fabre et al., 20 May 2025).
8. Culturally Anchor Symbolism: In context-rich domains (e.g., Hanzi for Chinese elders), use culturally meaningful interfaces for co-creation (Zhan et al., 2 Jul 2025).
9. Ensure Ethical Oversight: Guard against manipulative or exclusionary deployments by ongoing annotation transparency and participant control (Shen et al., 2024).

These principles scaffold the design of narrative systems—both fully automated and hybrid human–AI—that robustly center subjective experience, foster empathic connection, and preserve the diversity and agency integral to genuinely humanized storytelling.