Automated Detection of Similes and Oxymorons in Koshur and Awadhi Poetry: A Hybrid Algorithm Approach

Poetry, renowned for its magical and rhythmic qualities, incorporates various elements that contribute to its artistic allure. Among these elements, Figures of Speech (FsoS) play a significant role in enhancing the beauty of poetic compositions. This research paper focuses on two specific types of FsoS: Simile, known as 'उपमा', ‘Upma’ in Hindi and Oxymoron, referred to as 'ͪवरोधाभास', ‘Virodhabhas’ in the same language. The main objective of this study is to create automated algorithms capable of identifying similes and oxymorons within Koshur and Awadhi poetry. To achieve this, the researchers designed four distinct algorithms tailored to each language. These algorithms were then applied to a dataset containing 3200 poems in both Koshur and Awadhi. Given the absence of significant NLP resources for these languages, the authors encountered challenges in their work. They had to generate their own resources, including lists of antonyms for oxymorons and comparative words for similes. Impressively, after using Cohen’s Kappa the study achieved a strong level of agreement among human annotators, with an Inter-Annotator Agreement (IAA) score of 84.4% for Koshur and 91% for Awadhi. The automated algorithms also demonstrated notable accuracy, achieving 95% in both oxymoron and Simile for Koshur and for Awadhi it is 94% and 96% respectively. The authors also implemented the concept of hybridization, where mixing of different algorithms is done to achieve higher level of accuracy and authors successfully achieved 100% accuracy after hybridization. These results highlight the efficacy of the developed algorithms in detecting these poetic devices in the respective languages, despite the limited existing NLP resources available.