This paper addresses the challenge of preserving sentiment when translating sacred texts, with a specific focus on the Quran. The proposed approach combines advanced Artificial Intelligence (AI) techniques, particularly deep learning-based Transformer Language Models (TLMs), with a novel human validation approach. We present a comprehensive study involving a newly created parallel dataset encompassing the Arabic Quran and seven English translations, analyzing the preservation of sentiment. Our findings reveal compelling insights, with neutral sentiment ranging from 59% to 74% in English translations compared to 66% in the original Arabic Quran. Negative sentiment in some translations reached 25%, while others ranged from 14% to 17%, closely paralleling the 24% in the Arabic version. Additionally, the agreement analysis among English translations indicates varying degrees of alignment, reaching a Good level (κ = 0.62) or a Moderate level (κ from 0.47 to 0.6). However, compared to the original Arabic Quran, none of the translations achieved high levels of agreement, with only four translations reaching a Fair score (approximately 0.21). These findings underscore the complexities of translating the Quran, particularly its classical Arabic, and emphasize the need for improved sentiment analysis models, potentially incorporating mixed sentiment categories to capture sentiment more effectively.
Muslims rely primarily on the Quran and the Hadiths in all their spiritual life and consider them as sacred sources. If the Quran is God’s word, then the Hadiths are God’s instructions in the words of the Prophet Muhammad. Since Hadiths are transmitted through multiple narrators, they have been extensively studied to ensure their authenticity. The purpose of this study is to detect fabricated Hadiths, or Mawdu, which is the type of Hadith most rejected by Muslim scholars. The study utilises the central text and content of Hadith, Matn, rather than solely focusing on Hadith chain of narrators, Sanad. In order to accomplish this, we create and release the first dataset dedicated to Mawdu Hadiths, called MAHADDAT. Furthermore, we set up a Mawdu Hadith (MH) detection system based on a transformer language model, BERT, achieving a 92.47% $\text{F}1_{MH}$ score.
The performance of existing transformer-based language models in providing state-of-the-art results on many downstream tasks is well established. However, these models tend to be limited to high-resource languages or are multilingual in nature. The availability of models dedicated to Arabic dialects is limited, and even those that exist primarily support dialects written in Arabic script. This study presents the first BERT models for Moroccan Arabic dialect, also known as Darija, called DarijaBERT, DarijaBERT-arabizi, and DarijaBERT-mix. These models are trained on the largest Arabic monodialectal corpus, supporting both Arabic and Latin character representations of the Moroccan dialect. The models' performance is evaluated and compared to existing multidialectal and multilingual models on four distinct downstream tasks, demonstrating state-of-the-art results. The data collection methodology and pre-training process are described, and the Moroccan Topic Classification Dataset (MTCD) is introduced as the first dataset for topic classification in the Moroccan Arabic dialect. The pre-trained models and MTCD dataset are available to the scientific community.
6 November 2024 in International Journal of Data Science and Analytics