Abstract:

Automatic readability assessment for Hindi text has remained largely unsolved despite. Hindi being spoken by over 886 million internet users and around 14.7 lakh schools across India. Existing tools based on Flesch-Kincaid and related formulas fail on Devanagari script because they do not account for matra complexity (vowel diacritics), conjunct consonant density (virama-based fused consonants) or India's CBSE grade structure. This paper presents the Hindi Readability Score (HRS), a corpus-validated readability formula designed from scratch for the Devanagari script. HRS incorporates two novel features not found in any prior readability formula: “conjunct density detected via Unicode virama analysis (U+094D) and matra complexity based on guru/laghu syllable weight”. We validate HRS against a 49 - sentence corpus drawn from NCERT Class 1-12 textbooks, Constitution of India, legal texts and Hindi news sources. HRS achieves Pearson r = 0.81 with human-assigned difficulty ratings used and a Mean Absolute Error of 1.67 school grades. We also present the Hindi Grade Level (HGL) formula mapping HRS to CBSE school grades (Class 1 to college) calibrated via least-squares regression. The complete implementation is released as an open-source Python package (pip install hindi-readability) with zero external dependencies.

Keywords:

Hindi NLP, readability assessment, Devanagari script, text complexity, CBSE grade level, Indic language processing, conjunct consonants, matra complexity.

Aim/Objectives

Methods