Hybrid Morph-Analysis Model for Marathi

Syntactic and semantics knowledge of a language is necessary for performing any text analysis task. Understanding of words (for number, gender, case, Vibhakti, tense), and their dependency on other words is important for developing applications, such as Sentiment Analysis, Question Generation, Machine Translation, Information Retrieval, etc. The information such as gender, number, case, etc. is retrieved by looking at the affixes attached to the word. Morphology is how the word is formed using small units called morphemes. Morphological Analysis (MA) is the process of finding affixes attached to the word that expresses the grammar information of the word. Marathi is a morphologically rich Indian language spoken in Maharashtra. This research paper proposes a language-independent method to find words’ morphology using Machine Learning (ML) and Deep Learning (DL) models. The proposed system has three modules, first, identify the case of the word, second identify the affix attached to the words and last fine-tune the base word using a language model. Amongst the various ML and DL techniques, Random Forest algorithm predicts the word case with an accuracy of 97.33% whereas LSTM shows a 99%.