Published Paper Details:

This paper describes an SRL system custom designed for the Marathi language. It employs SVM, a powerful and well-established machine learning technique for classification tasks. The system first extracts a set of linguistic features from its training data, then trains the SVM classifier on those features. It tests the accuracy and performance of the SVM classifier in relation to the test dataset which is independent of the training dataset. Therefore, in order to increase the robustness of the system, datasets tagged manually from Marathi textbooks that have rich annotation quality have been used for testing. One of the biggest strengths in this SRL system is that its feature extraction is efficient and makes it easy for the classifier to capture linguistic patterns and relationships inside the text. The more elaborate recognition techniques make the identification of the proper semantic roles along with their correct labels possible in this system. This method has a disadvantage in that it is fixed on a set of features that might limit its adaptability to other linguistic structures or types of languages. In this implementation, the system was trained on a dataset of 38,929 tokens and tested on another set of 9,631 tokens. The findings from the evaluation suggest that the methodology is highly effective, as evidenced by the system achieving a notable average recall of 84.00%, precision of 85.00%, and an F1-score of 84.00%. Such results signify that the system exhibits strong performance in accurately categorizing the semantic roles in Marathi text, demonstrating significant potential for support vector machines in semantic role labeling tasks for Indian languages.