GENERATING INDIAN STYLE POETRY USING CHARACTER AND WORD LEVEL SEQUENCE MODELS

Objectives: To generate text using character level sequence models by training Indian Poetry and testing the applicability of sequence models for Indian poetry generation Methods: Recurrent Neural Network based models Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) models trained both at the character level and word level. Both models developed with 150 units with single layers. Findings: Character level trained models are performing well in generation of text, but failed at generating long sentences, they can generate at max 4- or 5-word sentences correctly. When compared LSTM and GRU the quality of text generation is more of GRU model. Further, we tested with word level trained models they are capable to generate 500-word paragraphs, but this paragraph doesn’t contain meaningful sentences. Generating a complete poem with a specific poet’s style is a challenging task. Novelty: working with Indian poets Rabindranath Tagore and Sarojini Naidu to generate India style poetry