Ge’ez Syntax Error Detection Using Deep Learning Approaches

The Ge’ez language, an ancient Ethiopian Semitic language, is still used in liturgical contexts and taught at university and college levels but lacks tools for part of speech tagging, morphological analyzers, and syntax error detection in written texts. This hinders the identification of syntax errors and poses a significant challenge for learners, and researchers. This study addresses this problem by developing a part of speech tagging and syntax error detection models for Ge’ez using deep learning approaches. To develop the model, a dataset of 4,981 sentences that have 30326 words and 11,747 unique words was collected for part of speech tagging. Additionally, a dataset of 1,170 sentences was collected for syntax error detection. LSTM and BiLSTM algorithms were used to develop the models. The LSTM model achieved an accuracy of 94% and 92.31% in the Gz_POS and Gz_SED tasks, respectively, and the BiLSTM model achieved an accuracy of 95.01% and 94.02% in the Gz_POS and Gz_SED. The results demonstrate the effectiveness of these deep learning algorithms for syntax error detection in the Ge’ez language. The developed model provides a feasible solution to the challenges of digitizing Ge’ez books and provides help for second-language learners. The findings contribute to the improvement of language education, research, and development in under-resourced languages. Future researchers can use the developed model and methodology as a framework for further advancements in Ge’ez language processing.