language:
- ar
size_categories:
- 1K<n<10K
Dataset Card for Dataset Name
GATmath & GATLc Datasets We introduce GATmath (7k questions) and GATLc (9k questions), two large-scale Arabic benchmarks for reasoning and language understanding, derived from the General Aptitude Test (GAT). This dataset card aims to be a base template for new datasets. It has been generated using this raw template.
Dataset Details
License: CC BY-NC-SA
Dataset Description
Size: 9036 questions
-GATLc Tasks: [1]Verbal Analogy: Identify relationships between word pairs (part-to-whole, cause-effect, transformation). [2]Sentence Completion: Fill blanks with the correct word to ensure grammar, logic, and coherence. [3]Contextual Error: Detect the word that is wrongly placed or contradicts the sentence meaning. [4]Semantic Association & Disparity: Find words/phrases that are semantically related or the odd one out. [5]Reading Comprehension: Understand and analyze texts from various domains, answering questions that test deep understanding and reasoning.
- Curated by: AlBallaa S, AlTwairesh N, AlSalman A, Alfarhood S
- Shared by: AlBallaa S
- Language(s) (NLP): Arabic
Dataset Sources
Source Data
The General Aptitude Test (GAT), Saudi Arabia
Citation
AlBallaa S, AlTwairesh N, AlSalman A, Alfarhood S (2025) GATmath and GATLc: Comprehensive benchmarks for evaluating Arabic large language models. PLoS One 20(9): e0329129. https://doi.org/10.1371/journal.pone.0329129