BAREC-LexBench

Benchmark Summary

BAREC-LexBench is a benchmark for Arabic lemma-level readability assessment, designed to evaluate readability lexicons and computational models independently of their granularity levels. It is built through a large-scale pairwise comparison task, where annotators determine which of two Arabic lemmas is easier to read and understand. It consists of 36K annotated pairs covering open-class lemmas from four parts-of-speech: adjectives, nouns, proper nouns, and verbs.

---

Languages

• Arabic (Modern Standard Arabic)

---

Dataset Structure

BAREC-LexBench compares lemmas across:

• Four same-POS categories: adjectives, nouns, verbs, and proper nouns.
• One corss-POS category: compares lemmas from different parts-of-speech.

This dataset consists of 6 subsets: Adj (7.2K pairs), Noun (7.2K pairs), Verb (7.2K pairs), Prop (7.2K pairs), Cross (7.2K pairs), and All (36K pairs). Each subsets has four splits: train (50%), tune (12.5%), dev (12.5%), and test (25%).

Data Instances

{'Lex1': 'داخِل', 'Lex1_dediac': 'داخل', 'POS1': 'adj', 'POS1_ar': 'صفة', 'POS1_en': 'adjective', 'Gloss1': 'entering', 'Lex2': 'فُوسْفُورِيّ', 'Lex2_dediac': 'فوسفوري', 'POS2': 'adj', 'POS2_ar': 'صفة', 'POS2_en': 'adjective', 'Gloss2': 'phosphorous', 'Set': 'Set1', 'Split': 'Train', 'POS': 'adj', 'label': '1', 'Answer': 'A'}

Data Fields

• Lex1: The first lemma, with diacritics.
• Lex1_dediac: The first lemma without diacritics.
• POS1: Part of speech of the first lemma.
• POS1_ar: Part of speech of the first lemma in Arabic.
• POS1_en: Part of speech of the first lemma in English.
• Gloss1: English gloss of the first lemma.
• Lex2: The second Arabic lemma, with diacritics.
• Lex2_dediac: The second lemma without diacritics.
• POS2: Part of speech of the second lemma.
• POS2_ar: Part of speech of the second lemma in Arabic.
• POS2_en: Part of speech of the second lemma in English.
• Gloss2: English gloss of the second lemma.
• Set: Dataset subset identifier (Set1, ..., Set8).
• Split: Dataset split (Train, Tune, Dev, or Test).
• POS: Part of speech category for the two lemmas being compared (adj, noun, verb, noun_prop, or cross).
• label: Label indicating which lemma is easier to read (1, 2, or 0 if equal).
• Answer: One-to-one mapping from label ({'1': 'A', '2': 'B', '0':'C'}).

---

Citation

If you use BAREC-LexBench in your work, please cite the following paper:

@inproceedings{elmadani-etal-2026-barec-lex,
    title = "A Benchmark for Arabic Lemma-Level Readability Assessment",
    author = "Elmadani, Khalid N. and Saeed, Mostafa and Taha-Thomure, Hanada and Habash, Nizar",
    booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
    month = oct,
    year = "2026",
    address = "Budapest, Hungary",
    publisher = "Association for Computational Linguistics"
}