A Character Level Sanskrit-Malayalam Parallel Morphological Analyzer Using Deep Learning

Abstract

Morphological analysis is the process of describing grammatical variations of words such as gender, number, person, tense, aspect, modality, etc, based on different morphemes of a word. A morphological analyzer splits the word into its constituent morphemes.Sanskrit words are categorized into Subantas, Tinantas, Kridantas, Taddithas, Samasas etc. Each of these categories follows different rules for sandhi splitting. This paper proposes a parallel deep learning architecture that can perform morphological analysis of Sanskrit and Malayalam parallelly, with an overall percentage accuracy of 96.2. The model is also able do morph analysis of Subanta, Tinanta, Tadditha and Samasa words of Sanskrit with a percentage accuracy of 95.96, 96.93, 95.83 and 95.67 respectively. The individual Malayalam morphological analyzer has a percentage accuracy of 98.25 and Sanskrit morphological analyzer has a percentage accuracy of 96.52.