Urdu & Hindi Poetry Generation using Neural Networks

TL;DR

This paper presents a neural network-based system to generate Urdu and Hindi poetry, aiming to assist poets in overcoming writer's block by providing creative prompts while ensuring originality and creating a publicly available poetry corpus.

Contribution

It introduces a novel deep learning approach for Urdu and Hindi poetry generation and provides a curated, cleaned corpus for future research in this underexplored area.

Findings

Generated poetry maintains rhyme and meter constraints.

The system helps poets overcome writer's block with creative prompts.

The curated corpus is publicly available for research.

Abstract

One of the major problems writers and poets face is the writer's block. It is a condition in which an author loses the ability to produce new work or experiences a creative slowdown. The problem is more difficult in the context of poetry than prose, as in the latter case authors need not be very concise while expressing their ideas, also the various aspects such as rhyme, poetic meters are not relevant for prose. One of the most effective ways to overcome this writing block for poets can be, to have a prompt system, which would help their imagination and open their minds for new ideas. A prompt system can possibly generate one liner, two liner or full ghazals. The purpose of this work is to give an ode to the Urdu, Hindi poets, and helping them start their next line of poetry, a couplet or a complete ghazal considering various factors like rhymes, refrain, and meters. The result will help aspiring poets to get new ideas and help them overcome writer's block by auto-generating pieces of poetry using Deep Learning techniques. A concern with creative works like this, especially in the literary context, is to ensure that the output is not plagiarized. This work also addresses the concern and makes sure that the resulting odes are not exact match with input data using parameters like temperature and manual plagiarism check against input corpus. To the best of our knowledge, although the automatic text generation problem has been studied quite extensively in the literature, the specific problem of Urdu, Hindi poetry generation has not been explored much. Apart from developing system to auto-generate Urdu, Hindi poetry, another key contribution of our work is to create a cleaned and preprocessed corpus of Urdu, Hindi poetry (derived from authentic resources) and making it freely available for researchers in the area.