Deep learning based Character-level approach for Morphological Inflection Generation

In this paper, we present our work on morphological inflection generation of Sanskrit using a deep learning approach. Sanskrit is a morphologically rich language which came into use from the Vedic period. A basic understanding of the language formation is needed to study the abundant literature in it. Here a computational model for word formation in Sanskrit is proposed using deep learning based models. They are applied here to attain the morphological changes that a root word undergoes to result in the surface form. The approach is in character level so as to capture the character level transformations. The best performance was obtained from the Bidirectional Gated Recurrent Unit architecture with an accuracy of 98.42% and an F1-Score of 0.9838. This model is purely dependent on the dataset and does not require any external linguistic resources.