Machine Learning Based System for Identification of Light and Heavy Syllables in Sanskrit Verse

Abstract Prosody provides information regarding a rhythmic structure present in poetry or a verse for any language. In particular, for the Sanskrit language, recognizing the prosody (called “Chandas” in Sanskrit literature) of the verse is entirely dependent on the identification of the light (“Laghu”) and heavy (“Guru”) syllables. In this regard, we propose an efficient and low-complexity two-step machine learning based system for identifying the Laghus and Gurus in Sanskrit verses, by developing an ingenious representation for the syllables. Our novel idea of preparing the first data-set and categorizing the syllables using the first machine learning model enables us to solve the sequence-to-sequence conversion problem as a simple classification problem in the second machine learning model, with two classes namely, the Laghu (denoted by a ‘0’) and the Guru (denoted by a ‘1’). The proposed data representation method and the two-step technique have been found to give an accuracy of more than 99%.