sijantanvir/BanglaSocialBench

BanglaSocialBench是一个孟加拉语（bn）数据集，专注于社会语用学和文化评估，特别涉及亲属关系和称呼语。数据集规模在1K到10K之间，包含多个配置：地址术语（分为代词性和名词性子集，涵盖学术、家庭、医院、办公室、社交互动、社交称呼、亲属关系和敬语场景）、亲属推理和社交习俗。任务类型包括多项选择和问答，旨在评估模型在孟加拉社会文化语境下的理解和推理能力。

BanglaSocialBench is a Bengali (bn) dataset dedicated to sociopragmatic and cultural evaluation, with a specific focus on kinship and address forms. Spanning a scale of 1K to 10K instances, it includes multiple configurations: address terms (classified into pronominal and nominal subsets, covering scenarios including academia, family, hospital, office, social interaction, social address, kinship, and honorific usage), kinship reasoning, and social customs. The task types encompass multiple-choice questions and question answering, designed to assess models' understanding and reasoning abilities in the context of Bengali sociocultural scenarios.