<![CDATA[This study aims to evaluate the effectiveness of the DeepSeek artificial intelligence platform as a correction tool in the Murasalat course (Arabic correspondence) in the Arabic Language Education Study Program at Semarang State University. The background of this research is the complexity of Arabic proficiency for non-native Arabic speakers, encompassing grammar, vocabulary, and cultural sensitivity. The method used is a mixed-methods approach involving 30 fourth-semester students divided into experimental and control groups. The primary data consisted of students’ Arabic correspondence texts collected through pre-tests, post-tests, as well as questionnaire responses and semi-structured interviews. Quantitative data were analyzed using paired sample t-tests to determine differences between pre- and post-intervention scores, and effect size was calculated using Cohen’s d. Additionally, DeepSeek’s correction accuracy was measured across morphological, syntactic, semantic, and pragmatic error categories. Qualitative data were analyzed using thematic analysis to examine students’ perceptions of the tool. The results indicate that DeepSeek has high accuracy in detecting errors at the morphological, syntactic, semantic, and pragmatic levels, surpassing manual correction capabilities in several aspects. Statistical analysis revealed a significant improvement in scores among the experimental group, with a mean gain of 7,73 points and a very strong effect size (Cohen's d = 3.71). Students showed positive perceptions of this tool, as it provided instant feedback that enhanced their self-confidence. However, limitations were identified in detecting complex cultural-pragmatic nuances. This study concludes that integrating DeepSeek into the Arabic language curriculum can support more effective and accurate independent learning.]]>
Copyrights © 2026
