A new arXiv paper presents BanglaBBQ, a bias benchmark designed to measure stereotypes in question-answering systems within the Bangladeshi context. The benchmark is offered in both Bangla and English, allowing for cross-lingual evaluation of model bias.
The work explicitly follows the pattern set by KoBBQ and PakBBQ, which adapted the original Bias Benchmark for QA (BBQ) to Korean and Pakistani contexts. BanglaBBQ thus continues a broader effort to move bias evaluation beyond Western-centric datasets.
Because the abstract is truncated, details on the benchmark's construction and evaluation results are not yet available from this source. This is a single paper announcement, so there are no conflicting sources to compare.