The CHOCLO Latin American Culture Benchmark Dataset is a benchmark dataset designed for the task of evaluating Latin American cultural knowledge in language models. It aims to assess the accuracy of language models in representing Latin American culture, and is specifically designed to address real-world issues such as the underestimation, omissions, and biases of Latin American culture in language models. This dataset contains structured instances across multiple categories, covering seven core Latin American cultural categories: tradition, cuisine, public figures, geography, animals, plants, and cultural heritage. It also includes data from 18 Latin American countries, such as Chile, Mexico, and Argentina, providing comprehensive coverage of the diverse cultural landscape of Latin America. The dataset is built upon Wikipedia text descriptions and employs multi-stage filtering and manual verification strategies to ensure data quality and cultural appropriateness. Data Fields:

• EntityEntity Name
• CountryCountry of origin of the entity
• CategoryPhysical categories (such as food, tradition, public figures, etc.)
• DifficultyQuestion difficulty (easy, medium, hard)
• QuestionRegarding the issue of entity generation
• AnswerExpected answer