A seminatural approach towards restoration, classification, and pattern extraction from heritage Hindi poetry: A probabilistic case study with Dohā chanda.

This article focuses on the computational restoration, metrical analysis, and statistical study of Kabir's dohā, a form of poetic couplet in old Hindi dialects. It presents a semi-natural restoration approach to correct metrical inconsistencies in a dataset of 899 dohās attributed to Kabir, followed by algorithms that classify these couplets based on metrical properties and rhyme types into nine distinct classes. The study employs deterministic finite automata for rhythm verification and statistical models—including Bernoulli trials, Chi-square goodness-of-fit tests, and hypergeometric distribution—to analyze syllabic patterns, revealing stable rhythmic structures and normal-like distributions in the dataset. The work contributes a restored dataset, automated classification tools, and statistical insights, offering a foundation for future research in style recognition, authorship attribution, and computational analysis of old Hindi poetry.