
This dataset explores the nuanced cultural biases inherent in "common sense" datasets, primarily showcasing implicit Western or U.S.-centric perspectives.

Dataset Overview
The AMAMMERε Dataset allows for the exploration of cultural commonsense reasoning across two distinct cultures: Ghana (representing a low-resource language group) and the USA. By juxtaposing these cultural contexts, our dataset offers a profound look into the variances in commonsense reasoning as it pertains to daily activities such as shopping, meal planning, and transportation.

Highlights:
525 Questions: A curated selection from existing commonsense datasets, rewritten in unspecified, Ghana Specified and US Specified cultural settings to spotlight cultural differences.
Format: Available in CSV, EXCEL and JSONL formats.
Annotations: Validated by human annotators from the two select cultures: US and Ghana to ensure reliability and cultural relevance.

Find comprehensive details in our paper.

Structure:

ID: Unique identifier for the question.
Context: Scenario or premise of the question.
Question: The question text.
Options (A-D): Multiple-choice options.
Answer Key: Correct option number, if applicable.
Answer Text: Full text of the correct answer, when available.
Country: The primary country or cultural context of the question.
Type: Classification of the question as explicit or implicit in cultural context.
Dimension: The subject area or theme of the question.
OriginalDatasetID/SetNumber/QuestionNumber: Metadata tracking the question's origin.
DatasetOrigin: The specific original dataset from which the question was derived.

AMAMMERε © 2024 is licensed under CC BY 4.0 