Broken Letters, Broken Narratives: A Case Study on Arabic Script in DALL-E 3

2025-02-27

AI Summary

This study identifies systematic failures in text-to-image models (e.g., DALL·E 3) when generating Arabic calligraphy: all outputs exhibit character distortion, spurious ligatures, and cultural appropriation—revealing a structural misrepresentation of non-Western visual cultures. Methodologically, the work innovatively integrates Said’s theory of Orientalism with the historical discourse on pseudo-Arabic script into generative AI evaluation, combining prompt engineering, iterative generation experiments, visual semiotic analysis, and cross-historical comparisons (e.g., medieval pseudo-Arabic ornamentation). It proposes a critical framework asserting that “technical representation is cultural politics.” Empirically, the study demonstrates that state-of-the-art models cannot produce legitimate Arabic calligraphy and instead perpetuate algorithmic Orientalism. This work establishes a novel theoretical pathway and methodological benchmark for studying cultural representation in AI systems.

Technology Category

• Cognitive Modeling & Cognitive Systems: Computational Creativity
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Web Mining and Content Analysis: Web data quality in the era of algorithmically-generated content

Abstract

Text-to-image generative AI systems exhibit significant limitations when engaging with under-represented domains, including non-Western art forms, often perpetuating biases and misrepresentations. We present a focused case study on the generative AI system DALL-E 3, examining its inability to properly represent calligraphic Arabic script, a culturally significant art form. Through a critical analysis of the generated outputs, we explore these limitations, emerging biases, and the broader implications in light of Edward Said's concept of Orientalism as well as historical examples of pseudo-Arabic. We discuss how misrepresentations persist in new technological contexts and what consequences they may have.

Problem

• Limitations of DALL-E 3 in representing Arabic calligraphy
• Biases in text-to-image AI for non-Western art forms
• Misrepresentations and Orientalism in generative AI outputs

Innovation

• Analyzes DALL-E 3's Arabic script limitations
• Explores biases in text-to-image AI systems
• Links misrepresentations to Orientalism concepts