Answers are grounded in this paper and cite their sources.
The digital transformation of religious scholarship has altered how Islamic scriptures are accessed globally, particularly for non-Arabic-speaking communities like Indonesia. This paper proposed a novel BERT-based post-training framework with multi-task optimization specifically designed to enhance Indonesian Islamic text translation quality. The proposed approach addressed unique characteristics of religious texts, including hierarchical structures, inter-textual references, and domain-specific terminology. The framework integrated BERT-based encoders for Arabic and Indonesian languages, enhanced with domain-specific adaptations. A multi-task learning framework was introduced that simultaneously optimized four objectives: translation quality assessment, terminology consistency verification, semantic alignment detection, and theological accuracy validation. Extensive experiments demonstrated that the proposed approach achieved a Pearson correlation coefficient of 0.89 with human expert ratings, significantly outperforming existing baselines (e.g., ALBERTIR at 0.72). The framework maintained 87% of its performance with only 30% of training data, demonstrating robustness in low-resource scenarios. Additionally, the hierarchical attention mechanism and cultural context integrator preserved theological nuances, achieving 92.3% theological accuracy and 94.1% terminology consistency. This work successfully bridged the gap between technical NLP advancements and religious scholarship, offering a reproducible approach for sensitive-domain translation.
No takes yet. Share an insight, caveat, or question.
Darmawan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: