Practical Models for Simulating Texts in Distinguished, Signature Styles, Under the Use of AI Tools in Resistance Literature

Compiled by Hassan Beheshtipour
Translated by Fazel Shirzad

2026-2-18

Introduction
Emerging technologies, particularly generative artificial intelligence, have in recent years transformed not only scientific and industrial fields but also sensitive domains of culture, historical memory, and literature. One of the most important of these domains is resistance literature that narrates the suffering, resilience, and lived experiences of people who have endured war, captivity, oppression, and collective crises,[1] and who resist the enemy by any means possible.

Resistance literature is not merely a report of events; it is the artistic formulation of a historical experience of resilience by people who refuse to accept foreign domination or live under tyranny. Many of these experiences are initially recorded as oral narratives, raw memories, interviews, and informal speech (narratives rich and valuable in content, yet often lacking coherent literary structure and refined language).[2]

The central question of this article begins here:

Can artificial intelligence help recreate these raw, oral narratives in the form of distinguished, signature literary styles, without distorting the historical authenticity and lived experience of the narrator?

This article seeks to outline practical models for such an approach—an approach in which AI is employed not as a substitute for the narrator or author, but as a "stylistic assistant" and a tool for controlled literary recreation. What follows is an exploration of the conceptual foundations, technical models, literary examples, collaborative and ethical frameworks, and the risks involved in this path.

1. Conceptual and Theoretical Foundations

1.1. Why Simulate Literary Style?

The goal of simulating literary style in resistance literature is not to distort content or falsify historical narrative, but to enhance the literary grandeur of narratives that have inevitably become simplified, fragmented, or scattered during oral transmission. In this approach, historical content (time, place, actions, and persons) remains fixed; only the form of expression is elevated.

In other words, style simulation seeks to bridge the gap between "raw narrative" and "literary text" without tampering with the reality of events, using tools of stylistic transfer.

1.2. Resistance Literature at the Border of History and Literature

Resistance literature must be understood at the border between history and literature—where historical reality meets artistic language and imagination. Many oral narratives, though powerful in content, lack narrative cohesion, precise imagery, and literary rhythm. In this context, AI emerges as a "stylistic specialist assistant"—a tool that can aid in reconstructing form, but should not become the owner of the narrative. This perspective lowers AI from the status of "creator" to that of "tool," clarifying its relationship with the human author and historical narrator.

1.3. The Core Risk: Well-written Texts, Devoid of Lived Experience

The fundamental risk in this field is producing texts that are literarily beautiful and technically precise, yet emptied of real lived experience and ethical impact. In resistance literature, we are not dealing with the free literary embellishment of history, but with recording history in a literary form.

Therefore, in depicting and expressing events, artistic excess or deficiency must be avoided so that historical narrative is not compromised for the sake of well-written literary style. Any style simulation must be conducted within a framework of historical sensitivity, ethics, and responsibility toward human suffering.

2. Technical Models for Implementing Style Simulation

2.1. Specialized Model Fine-tuning (Author-Centric)

In this model, a general language model (like GPT) is fine-tuned on a limited, specialized corpus of works by a specific author with a signature style or a particular literary movement to learn their stylistic features.[3] This training encompasses not only vocabulary but also:

Sentence patterns and unique linguistic style,

Recurrent metaphors, similes, and imagery patterns,

Rhythm and prose cadence, methods of scene-setting and narration,

Even punctuation and meaningful silences in the text.[4]

Practical Example: Recreating a combatant's memories in the style of Reza Amirkhani:

The model is trained on Amirkhani's narrative and fictional works, and possibly on literary critiques of his style. The simple oral text, after accurate transcription, is fed to the model. The output is a narrative that preserves historical content but is elevated into Amirkhani's narrative, social, and dramatic prose style.

Key Point: In designing this process, explicit instructions in training data and human oversight must guarantee that historical content (events, sequence, and names) is unalterable, and only the form of expression is permitted to transform.

2.2. Multi-layered Style Transfer

Text style transfer in AI literature is defined as a method for changing linguistic and stylistic choices without manipulating content independent of style. In the multi-layered approach, text is decomposed into these components:

Core Content Layer: Facts, events, and actions.

Base Linguistic Layer: Neutral grammatical structure and sentence construction.

Target Style Layer: Stylistic features of the reference author or poet.

The practical process could be as follows:

The initial oral text is accurately transcribed, including cues for tone, pause, and emphasis. The AI model, utilizing style-transfer techniques, separates the narrative core from the linguistic shell. This core is then reconstructed in a new style (e.g., the poetry of Hamid Sabzevari, the prose style of Beyhaqi, or the social prose of Jalal Al-e Ahmad). The narrator's voice, at the level of key expressions, point of view, and event sequence, is preserved as an underlying layer.

The main challenge in this use of AI is preventing the dominance of style over historical content. The style should not become so overpowering that the audience perceives the reference author more than the historical narrator or views the narrative merely as an "independent literary work" detached from the event.

3. Literary-Artistic Models in Text Recreation

3.1. Epic-Historical Recreation Model

Many contemporary resistance narratives, if recreated in classical or grand forms, reveal their epic, symbolic, and trans-historical dimensions more clearly.

Example: A raw oral narrative: "It was morning. It was raining. I was in the trench with my friend Ahmad. Suddenly a bullet came, and Ahmad fell."

Recreation in a style resembling Beyhaqi's historiography:

"As the morning of a spring day, the rain had moistened the earth, I, with my longtime companion, named Ahmad, had entered the shelter of the trench. Suddenly, an arrow of death, from an unknown direction, arrived and struck Ahmad to the ground, as if a young branch were severed from an ancient tree."

In this recreation: The event, characters, and outcome remain unchanged. But the method of imagery, simile, and prose rhythm elevates the narrative from the level of personal report to that of a tale with tragic and epic dimensions.

3.2. Model of Blending Oral Prose and Refined Literature

In this model, the authenticity of dialect and local speech is preserved but combined with more literary syntactic and lexical structures. Oral narrative with local dialect:

"It was early in the war, we were sitting on the street, and suddenly we saw planes coming..."

Recreation preserving authenticity but with literary elevation: "In those early days of the war, as we sat upon the cobblestones of the street, suddenly the roar of planes shattered the air..." Techniques used:

Converting colloquial speech to standard prose, without removing local idioms and flavor.

Preserving the rhythm and tone of speech in sentence structure.

Adding sensory descriptions (cobblestones, roar) without altering the reality of the event.

This model builds a bridge between the lived language of the narrator and the literary language of a broader audience.

4. Collaborative and Ethical Models

4.1. The "Three-Layer Correction" Model

To ensure authenticity and ethical integrity, AI output must pass through three filters:

A. Technical-Stylistic Filter: Review to answer:

Is the target author's style correctly imitated?

Are there signs of superficial imitation or stylistic confusion?

B. Content-Historical Filter:

Have historical facts, event sequences, and identities been distorted?

Has anything been omitted or added that alters the meaning of the event?

C. Ethical-Human Filter: Review to answer:

Has the dignity of the narrator and their suffering been preserved?

Has the narrative been reduced to mere entertainment or fast consumption? [5]

Description of the practical final text evaluation process:

AI produces the initial version.

A stylistic specialist reviews it for literary and stylistic quality.

A historian confirms historical accuracy and the text's alignment with sources.

When possible, the narrator themselves or their family review the final text and express their approval or critique.

4.2. The "Writing with Multiple Styles" Model

In this model, a single event is narrated in several different styles to demonstrate the multi-faceted nature of truth and the possibility of plural interpretations.

Example: The event of a combatant's martyrdom on the front:

Version in the style of Mr. Jalal Al-e Ahmad: With a sharp socio-critical approach.

Version in the style of Mrs. Simin Daneshvar: Focusing on human, emotional aspects and family relationships.

Version in the narrator's own style: Minimally edited, in raw form.

Benefit: Demonstrates how style influences perspective and interpretation. Avoids imposing a single reading or solidifying an "official narrative" upon diverse experiences.

5. Critical Considerations and Risks

5.1. Risk of "Aestheticizing Tragedy"

When suffering, violence, and death are described with excessively poetic or romantic language, there is a risk that the reality of violence is softened and the audience, instead of ethical confrontation, becomes absorbed in the beauty of form.

Practical suggestion: In designing instructions for the AI model, emphasize consciously preserving a level of the narrator's roughness, silence, stammer, and broken language. Also, the contrast between the beauty of form and the ugliness of content should not be erased but highlighted, so the ethical and historical shock of the narrative remains.

5.2. Risk of "Unintended Distortion through Style"

The style of each writer is tied to a specific worldview and ideology. Using the style of Jalal Al-e Ahmad, Simin Daneshvar, or any other writer can, without explicitly changing objective content, impose a particular political, social, or ethical interpretation onto the text.[6]

Solutions: For highly sensitive events, use more neutral styles or styles closer to the narrator's own language. Add an introductory explanation to texts recreated in a specific style, e.g., "This narrative has been reconstructed through the lens of the narrative style of [Author's Name]." Additionally, provide the audience with the opportunity to compare the recreated version with the raw version.

5.3. Issue of "Dual Authorship"

The final text is based both on the memories and lived experience of the narrator and is the result of AI intervention and the style of the reference author. This situation challenges the classical model of single authorship and creates a form of multiple ownership.[7]

Suggestion for transparency: Clearly state in the book's introduction, platform, or database: "This narrative is based on the memories of [Narrator's Name], inspired by the narrative style of [Author's Name], and recreated with the aid of AI tools." This way, the roles of AI, the human team, and the narrator are distinguished and documented legally and ethically.

6. Sample Hypothetical Practical Project

6.1. Project Outline

Project Name: "Unspoken Words; War Memories in the Styles of the Greats Who Weren't There"

Goal: To transform 100 hours of oral interviews with combatants and their families into 100 literary pieces in the styles of 10 prominent Iranian authors, while fully preserving content authenticity and ethical sensitivity.

6.2. Implementation Stages

Selection of Model Authors: For example Jalal Al-e Ahmad (social-critical prose), Simin Daneshvar (emotional-psychological prose), Reza Amirkhani (contemporary narrative prose), Hedayatollah Behboudi (documentary historiographical prose), and others.

Model Training: For each author, a separately configured AI model is fine-tuned. Models are trained not only on their works but, if possible, also on literary critiques and stylistic analyses of them.

Production Process: Each memory is first transcribed simply and faithfully. The text is then fed to the relevant model to produce the styled version. The output is reviewed and revised by a three-person team (literary scholar, historian, and psychologist/ethicist). When possible, the final version is read to the original narrator or their family, and their feedback is recorded.

Outputs: A print book with artistic design; an interactive website where users can view a single event in different styles; a digital audio platform with professional voice acting and the ability to compare raw and recreated versions.

7. Final Questions and Theoretical Horizon

This discussion ultimately leads to several fundamental questions that extend beyond a single project:

Is "style" truly separable from "content," or does every style inevitably impose its own interpretation on the text? [8]

Where is the boundary between "recreation" and "literary forgery"?

If AI produces a text that even experts cannot distinguish from the original author's work, is this a sign of artistic success or an instance of fraud and disruption of literary trust?

Does human suffering need "literary beautification," or should it remain in its raw, unadorned state to preserve its authenticity?

The proposed answer of this article is that the best use of AI in this field is not replacing authentic voices, but building bridges between generations and cultures. Transforming a simple oral memory into a refined text can bring today's youth, who relate better to a certain language, closer to the experiences of the past generation—provided the narrator's voice is not silenced and history is not reduced to mere aesthetics.

Concluding Remark:

All these technologies should serve one primary goal: listening to silenced voices and transmitting them accurately, honorably, and majestically to the future. Any technology that compromises this goal, no matter how advanced, is unworthy of use in this sensitive domain

Sources:

Transforming Text: The Future of Style Transfer. https://scisimple.com/en/articles/2025-07-08-transforming-text-the-future-of-style-transfer--a3lmyge

Fine-Tuning LLMs: A Guide With Examples
https://www.datacamp.com/tutorial/fine-tuning-large-language-models

[Literature Review] A Survey of Text Style Transfer: Applications and Ethical Implications (Optional Secondary Review)
https://www.themoonlight.io/en/review/a-survey-of-text-style-transfer-applications-and-ethical-implications

Artworks Without an Author: Reflections on AI-Generated Art https://observatory.tec.mx/edu-news/ai-generated-art/