Cahiers de Recherches en Sciences du Langage

From Cultural Texts To Ai-generated Images: A Qualitative Comparative Case Study Of Yennayer Via Dall-e 3

Auteurs : Mahdjour Nora .

Résumé

Abstract: This study examines how text-to-image generative AI (DALL-E 3) represents local cultural specificities, using Yennayer (the Amazigh New Year celebrated in Algeria) as a focused case study. By comparing culturally dense textual descriptions from the source corpus with generated images, the analysis evaluates three dimensions: material artifacts and dress, spatial and architectural context, and symbolic or emotional resonance. While the system effectively conveys broad affective cues (warmth, festivity), its reliability decreases when encountering culturally specific elements, frequently substituting generic or cross-regional visual patterns for localized markers. The study argues that visual plausibility is distinct from cultural accuracy, advocating for evaluation procedures that include explicit checks of local cultural markers and involve culturally informed reviewers. Ultimately, the article contributes to ongoing discussions surrounding algorithmic cultural bias, representational diversity, and the deployment of generative AI in digital heritage documentation.

Mots clés

text-to-image AI generation (DALL-E 3) ; cultural representation ; Yennayer ; qualitative comparative case-study