Arabic EXAMS-Redux: Corrected High School Exam Benchmark

A corrected version of the Arabic subset from the EXAMS multilingual high-school benchmark addresses widespread text corruption. The dataset repairs issues like split diacritics, fragmented words, and non-Arabic glyphs found in the original PDF extraction. It was created by inceptlabs and last updated on 2026-05-04.

Use Cases

• Benchmarking Arabic NLP models based on high-school exam questions.
• Training models for educational question-answering using Modern Standard Arabic text.
• Evaluating text repair and normalization techniques on corrupted Arabic text.
• Studying the performance of multilingual models on Arabic-specific academic content.

Strengths

• Specifically addresses and repairs corrupted Arabic text from PDF extraction.
• Focuses on Modern Standard Arabic within a multilingual high-school exam benchmark.
• Last updated on 2026-05-04, suggesting recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect geographic or source bias inherent to the original EXAMS benchmark.

Provenance

Source

inceptlabs

Collection Method

Corrected and text-repaired version of the OALL/Arabic_EXAMS dataset.

Freshness

Last updated 2026-05-04 12:54:23