Kafi Noonoo Grammar Errors Detection Using Deep Learning Approaches

In recent years, the existence and amount of textual information in different domains and languages are increasing radically. The same is happening to the Kafi noonoo language that is spoken by the Kaffa people in the southwestern part of Ethiopia. Kafi noonoo textual information is documented daily and weekly in sentence forms. However, Kafi noonoo sentences may have different grammatical errors, such as subject-verb disagreement (SVD), adjective-noun disagreement (AND), and adverb-verb disagreement (AVD) problems in sentence structure. This study developed a Kafi noonoo grammar error detection model by using deep learning approaches, in which both bidirectional long short-term memory (BiLSTM) and bidirectional gated recurrent unit (BiGRU) neural network algorithms were applied. For the development of the proposed model, Python programming languages and integrated packages have been used. The Kafi noonoo datasets were collected from Kafi noonoo linguistics-related sources using different techniques. Different experiments have been conducted by bidirectional short-term memory (BiLSTM) and bidirectional gated recurrent unit (BiGRU) separately with different feature extraction technique combinations on the collected dataset. As an experiment result shows, a bidirectional gated recurrent unit neural network achieved a test accuracy of 92%, precision of 83%, recall of 82%, and f1_score of 82% with the keras word embedding technique, while bidirectional long short-term memory achieved 91% of testing accuracy, 83% of recall, 83% precision, and 83% of f1_score with word2vect word embedding techniques.