Quantifying Cultural Fairness in Multilingual AI: A Cross-Lingual Evaluation of Representational Bias in Digital Humanities Corpora
Journal of Intelligent Decision Making and Information Science · 27.08.2026 · 3 (9s) · 2621-2639
DOI: 10.17762/jidmis.v3.2560 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Intelligent Decision Making and Information Science, online erschienen am 27.08.2026. Verfasst von Mohammed Saleh Al-Shizawi. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Despite claims of inclusion, multilingual artificial intelligence (AI) models remain dominated by Western linguistic and cultural assumptions. This mismatch poses serious challenges to the Digital Humanities (DH), as AI plays an increasingly important role in mediating cultural perception. To determine how well multilingual natural language processing (NLP) systems portray non-western cultures, this article objectively analyzed their representational fairness. They were comparative, experimentally developed, and included three balanced corpora: English folklore, Chinese popular reviews, and Yoruba oral narratives (each containing approximately 55,000 tokens). Multilingual transformers Multilingual BERT (mBERT) and XLM-RoBERTa (XLM-R) were assessed using five metrics: semantic similarity, representational stability, lexical expressiveness, distributional parity, and a compound Cultural Fairness Index (CFI). Cosine similarity, variance testing, and bootstrap confidence interval were used to perform statistical testing (1,000 iterations). mBERT was moderate in terms of fidelity across languages (mean CFI = 0.519 +/-0.006 Yoruba; 0.501 +/- 0.009 Chinese), whereas XLM-R demonstrated a great deal of fidelity but lower culture-fidelity (mean CFI = 0.523 +/- 0.021 Yoruba; 0 English consistently outperformed non-Western languages in 15 to 25 percent of semantic similarity measures (p-value < 0.001). Embedding space visualization revealed that English and semantic compression of Yoruba representations clustered hierarchically. There is still a Western body of hierarchy that has been coded into cross-lingual AI systems, confusing linguistic coverage with cultural understanding. The suggested Cultural Fairness Index provides a repeatable methodology for quantifying cultural representation in multilingual AI, paving the door for more equitable and inclusive computational humanities research.
Zitieren
APA
Al-Shizawi, M. S. (2026). Quantifying Cultural Fairness in Multilingual AI: A Cross-Lingual Evaluation of Representational Bias in Digital Humanities Corpora. Journal of Intelligent Decision Making and Information Science, 3(9s), 2621-2639. https://doi.org/10.17762/jidmis.v3.2560
Vancouver
Al-Shizawi MS. Quantifying Cultural Fairness in Multilingual AI: A Cross-Lingual Evaluation of Representational Bias in Digital Humanities Corpora. Journal of Intelligent Decision Making and Information Science. 2026;3(9s):2621-2639. doi:10.17762/jidmis.v3.2560
BibTeX
@article{AlShizawi2026,
author = {Al-Shizawi, Mohammed Saleh},
title = {Quantifying Cultural Fairness in Multilingual AI: A Cross-Lingual Evaluation of Representational Bias in Digital Humanities Corpora},
journal = {Journal of Intelligent Decision Making and Information Science},
year = {2026},
volume = {3},
number = {9s},
pages = {2621--2639},
doi = {10.17762/jidmis.v3.2560}
}
EUVIMED