5-Dialects-BN Benchmark for Bangla Dialectal LLM Evaluation
September 10, 2026
5-Dialects-BN is a multi-annotation benchmark comprising 6,000 entries across five regional Bangla dialects. The dataset aligns Romanized transliteration with dialectal text, Standard Bangla, and English to address performance degradation in low-resource, dialectally diverse settings.
HOW THIS AFFECTS YOU
●
builderYou can use these annotations to improve model accuracy for users speaking regional Bangla dialects.
●
researcherYou can utilize this dataset to evaluate how dialectal diversity impacts multilingual model performance.