Construction of a Recurrent Neural Network from the Amar Kosha

Abstract

In the traditional Sanskrit education system it is considered to be an essential and integral part of learning the grammar is to understand and memorise the complete Amar Kosha. In Sanskrit literary corpus the Kosha Granthas (the written record of the Sanskrit vocabulary) there are many available scripts. Among them Amar Kosha and Halayudha Kosha hold eminent positions. This paper focuses on the construction of a recurrent neural network(RNN) taking the Amar Kosha into consideration to classify the words recorded in it with respect to four prime attributes of the words, viz. the Varga(type), Linga(gender), Antya varna(the ending phoneme of the word) and the paryayavachi (synonyms). Among various schools of grammar of Sanskrit language, the Aindra tradition is the one which dictates a language model on the basis of the morphological structure of the words. It has been observed that Panini’s artificial language machine (a model of the Shaiva school of Sanskrit grammar) includes a set of metarule that can reconstruct the lost tradition of Aindra grammar. But in order to implement the set, the words should be modelled in such a way that it can provide a well-defined mathematical model. This work provides the platform with the implementation of RNN on Amar Kosha on which Panini’s word making rules can be easily implemented in order to reconstruct the Aindra School of Sanskrit Grammar.