BanglaDial: A merged and imbalanced text dataset for Bengali regional dialect analysis

BanglaDial is a unified corpus of Bengali dialectal text, consisting of 60,729 sentence-level entries representing eleven regional dialects and Standard Bangla. It reflects phonological, lexical, and syntactic divergences among dialects, offering a comprehensive linguistic spectrum beyond Standard Bangla. The dataset was compiled from online public repositories, representing regions such as Chittagong, Sylhet, Barisal, Rangpur, and others. All samples are written in Bengali script and annotated with their respective dialect labels, capturing key phonological, lexical, and syntactic variations across dialects. Bengali is the world's seventh most spoken language, with over 265 million speakers, and is known for its vast literary heritage and linguistic complexity. However, existing NLP (Natural Language Processing) technologies are predominantly trained on Standard Bangla, leaving dialectal varieties significantly underrepresented and poorly supported in digital tools. BanglaDial addresses this gap by providing a curated and annotated resource to support automatic dialect identification and the development of dialect-aware natural language processing applications. This dataset also contributes to broader goals of linguistic preservation and digital inclusion for underrepresented Bengali-speaking communities.