Estimating Related Words Computationally Using Language Model from the Mahabharata an Indian Epic

Abstract

Mahabharata' is the most popular among many Indian pieces of literature referred to in many domains for completely different purposes. This text itself is having various dimension and aspects which is useful for the human being in their personal life and professional life. This Indian Epic is originally written in the Sanskrit Language. Now in the era of Natural Language Processing, Artificial Intelligence, Machine Learning, and Human-Computer interaction this text can be processed according to the domain requirement. It is interesting to process this text and get useful insights from Mahabharata. The limitation of the humans while analyzing Mahabharata is that they always have a sentiment aspect towards the story narrated by the author. Apart from that, the human cannot memorize statistical or computational details, like which two words are frequently coming in one sentence? What is the average length of the sentences across the whole literature? Which word is the most popular word across the text, what are the lemmas of the words used across the sentences? Thus, in this paper, we propose an NLP pipeline to get some statistical and computational insights along with the most relevant word searching method from the largest epic 'Mahabharata'. We stacked the different text-processing approaches to articulate the best results which can be further used in the various domain where Mahabharata needs to be referred.

A Quantitative Social Network Analysis of the Character Relationships in the Mahabharata

“…None of the recent work in the Hindu literature has analyzed character networks as in-depth as our study. There have been recent studies analyzing the Mahabharata, but unfortunately they only look at word similarities derived from the Word2Vector word vectors and did not specifically analyze character networks [67,68]. Chandra and Ranjan [69] conducted a similar analysis on the Upanishads and Bhagavad Gita using BERT contextual embeddings for topic modelling, but again there was no analysis of the character network, as it is a difficult task to verify the quality of the character networks without annotating a ground truth network.…”