BAREC: Balanced Arabic Readability Evaluation Corpus

Over 1 million words of Arabic text annotated for readability at the sentence level. The BAREC dataset was developed by CAMeL-Lab for the BAREC Shared Task 2026, featuring fine-grained annotations across 19 readability levels. It also includes mappings to coarser 7, 5, and 3-level readability schemes.

Use Cases

• Train Arabic readability classifiers based on the 19-level annotation scheme.
• Benchmark text simplification models against the sentence-level readability scores.
• Develop multi-granularity NLP tools using the provided 7, 5, and 3-level readability mappings.
• Conduct linguistic analysis of Arabic text difficulty across a large corpus.

Strengths

• Large-scale corpus containing over 1 million words.
• Fine-grained readability annotation across 19 distinct levels.
• Provides multiple annotation schemes (19, 7, 5, and 3 levels) for flexibility.

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 source bias inherent to the original collection methods.

Provenance

Source

CAMeL-Lab

Collection Method

Developed for the BAREC Shared Task 2026; annotation methodology not detailed.

Freshness

Last updated 2026-05-17 14:06:28; freshness should be verified.