Prompting the Past: Linguistic Transformations and Cultural Accuracy in AI-Generated Image Reconstructions for Multivocal Cultural Heritage

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cultural Heritage Applications · Depth: Expert, quick

Summary

Research explored Generative AI's role in historical image reconstructions for multivocal cultural heritage interpretation. Three prominent AI image generation models were evaluated across three heritage test cases, generating 39 images from 13 user prompts. The study involved linguistic analysis of intermediate prompt transformations and visual assessment by heritage experts for historical accuracy and cultural sensitivity. Findings indicated models produced visually compelling and sometimes distinct depictions, but also showed representation imbalances, neutralisation, amplification, human portrayal inconsistencies, and linguistic misinterpretations. Initial guidelines for structured prompt construction were proposed based on identified failure patterns. The research suggests Gen-AI can be a supplementary tool, not a definitive source, for exploring multivocal heritage, particularly in museum and visitor engagement, when used critically with expert input.

Key takeaway

For heritage professionals developing interpretive materials or museum exhibits, Generative AI offers a powerful, yet nuanced, tool for visualizing diverse historical perspectives. You should critically evaluate AI-generated images for representation imbalances and linguistic misinterpretations, always integrating expert review. Implement structured prompt construction guidelines to mitigate identified failure patterns, ensuring AI serves as a supplementary exploration tool rather than a definitive historical source.

Key insights

Generative AI offers potential for multivocal heritage visualization but requires critical use due to inherent biases and linguistic transformation challenges.

Principles

• Multivocality enriches heritage interpretation.
• AI outputs need expert validation.
• Prompt construction impacts AI accuracy.

Method

Evaluated three AI image models using 13 prompts across three heritage cases, assessing 39 images via linguistic analysis and expert visual review for accuracy and sensitivity.

In practice

• Develop structured prompts to mitigate AI biases.
• Integrate AI visuals in museum engagement.
• Use AI as a supplementary interpretive tool.

Topics

• Generative AI
• Cultural Heritage
• Image Reconstruction
• Multivocality
• Prompt Engineering
• Historical Accuracy
• Cultural Sensitivity

Best for: AI Scientist, Research Scientist, AI Ethicist

Related on AIssential

• CULTURESCORE: Evaluating Cultural Faithfulness in Video Generation Models — Video generation models lack cultural faithfulness, especially in depicting behaviors, despite high visual quality.
• Using design to interpret the past and envision the future — Design and AI can reconstruct lost heritage and innovate future products across diverse scales.
• Assessing the Geographic Diversity of AI's Platial Representations in Image Generation — AI image generation models exhibit significant geographic homogeneity, risking stereotypical platial representations despite advanced capabilities.
• Why Most AI Art Prompts Fail. And How to Write Better Ones — AI art generation is a linguistic process where precise prompts are crucial for consistent, intentional image outcomes.
• I used GAN and a 330 BC sculpture to find out what Alexander the Great looked like — GANs can synthesize historical artifacts and artistic interpretations into novel visual reconstructions.
• ‘We can stitch together our past’: the AI-generated time-travellers vlogging from history — AI-generated historical vlogs engage young audiences by making history visually immersive and relatable through a proven content format.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.