Elamur Research⌕
☰
home / papers / 2021 / A Dependency Treebank for Classical Arabic Poetry

A Dependency Treebank for Classical Arabic Poetry

Summary

Classical Arabic had only one syntactic treebank, which covers the Quran. Its poetry, which often bends grammar to keep rhyme and meter, had no annotated resource at all. The authors built ArPoT v1.0, the first dependency treebank of Classical Arabic poetry. It contains 2,685 verses from poets of roughly the 6th–13th centuries, collected from literary websites and then cleaned: duplicates, broken verses, tatweel and punctuation were removed.

The annotation follows the CATiB scheme. The authors chose it because it stays near traditional Arabic grammar and can later be converted to UD. The MSA-trained CamelParser produced the first draft, which expert annotators corrected by hand in PALMYRA. Verses that continue a sentence across lines were then joined into single trees.

Two neural parsers trained on the treebank gave baseline scores, the best reaching about 75% LAS. The resource lets researchers study and parse a historically important, morphologically rich text genre. It could also support tools such as sentence boundary detection for poetry.

Key points

  • Corpus: 2,685 verses (35,459 tokens) from more than 775 poems by 34 Classical-era poets, sourced from ADAB and ALDIWAN.
  • Pre-annotation by CamelParser (trained on CATiB) tokenized about 52% of words incorrectly. On correctly tokenized verses, it reached only 55% exact match against gold for POS, head and label together.
  • Three native-speaker linguists corrected the data over four months.
    • Two annotators covering 83% of the corpus reached kappa 0.97 on an overlap sample.
    • A third annotator started at 0.85 agreement, which rose to 0.96 after a second revision round.
  • The output is in CoNLL-U, and the annotation uses CATiB POS tags and dependency labels.
  • Baselines, averaged over three runs with default settings:
    • The transition-based pointer network model reached 81.52 UAS and 75.25 LAS.
    • The parsing-as-tagging model reached 78.43 UAS and 70.95 LAS.

Interesting ideas

  • I1. Annotators work on each verse as its own small tree. When a verse's root depends on a word in another verse, they record the label and the external parent index, and the verses are stitched into one tree only after correction. This keeps editing views small while still producing full cross-line sentences. — For Mova (tree): Fable and tale texts with fragmentary sentences could be parsed in short units, with "dangling root → external head" links stored explicitly and merged later. Each merge would then be explainable.
  • I2. Tokenization errors are fixed first, and only then are agreement and parser accuracy measured. This keeps segmentation mistakes from contaminating syntax scores. — For Mova (tokens): Report tagger and parser quality only on items where token boundaries match gold. Segmentation failures would get their own separate metric.
  • I3. The source parser's quality is judged only on the subset where its tokenization was right. This tells you how useful machine pre-annotation is for the syntax layer specifically. — For Mova (learn): When ranking witness parsers for silver data, separate tokenization disagreement from attachment disagreement. A witness should not be penalized twice for one root cause.
  • I4. The annotation scheme was picked because it stays faithful to the traditional grammar while keeping a conversion route to UD open. — For Mova (tree): Treebanks in non-UD or domain-specific schemes can be kept native and converted to UD by an explicit, auditable mapping step. They do not need to be re-annotated.
  • I5. Agreement is checked on an overlap slice, and the third annotator does a second revision round, which raised kappa from 0.85 to 0.96. This is an inexpensive way to measure and then lift consistency. — For Mova (learn): For LLM-teacher or silver labels, keep a fixed overlap sample labeled by several sources and track agreement across repair rounds as a quality gate.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics: Treebanks

Cited by

8 papers of our library cite it (of the 140884 whose bibliographies we read); 5 of them have a page here, 3 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{alghamdi2021dependency,
  title = {{A Dependency Treebank for Classical Arabic Poetry}},
  author = {Al-Ghamdi, Sharefah and Al-Khalifa, Hend and Al-Salman, Abdulmalik},
  booktitle = {Proceedings of the Sixth International Conference on Dependency Linguistics (Depling, SyntaxFest 2021)},
  year = {2021},
  url = {https://aclanthology.org/2021.depling-1.1/}
}

References

21 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahmed, Munef Abdullah, and Stefan Trausan-Matu. 2017. Using Natural Language Processing for Analyzing Arabic Poetry Rhythm. In 16th Networking in Education and Research RoEduNet International Conference, 1–5, Targu-Mures, Romania. IEEE. find
  2. Albogamy, Fahad, Allan Ramsay, and Hanady Ahmed. 2017. Arabic tweets treebanking and parsing: A bootstrapping approach. In Proceedings of the Third Arabic Natural Language Processing Workshop, pp. 94-99. find
  3. Cohen, Jacob. 1960. A coefficient of agreement for nominal scales. Educational and psychological measurement, 20 (1):37-46. find
  4. Dukes, Kais, and Nizar Habash. 2011. One-step statistical parsing of hybrid dependency-constituency syntactic representations. In Proceedings of the 12th International Conference on Parsing Technologies, pp. 92-103. find
  5. Dukes, Kais, and Tim Buckwalter. 2010. A dependency treebank of the Quran using traditional Arabic grammar. In 2010 the 7th International Conference on Informatics and Systems (INFOS), pp. 1-7. IEEE. find
  6. Fernández-González, Daniel, and Carlos Gómez-Rodríguez. 2019. Left-to-Right Dependency Parsing with Pointer Networks. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pp. 710-716. find
  7. Habash, Nizar, Abdelhadi Soudi, and Timothy Buckwalter. 2007. On Arabic transliteration. In Arabic computational morphology, pp. 15-22. Springer, Dordrecht. find
  8. Habash, Nizar, Reem Faraj, and Ryan Roth. 2009. Syntactic Annotation in the Columbia Arabic Treebank. In 2nd International Conference on Arabic Language Resources and Tools MEDAR, Cairo, Egypt. find
  9. Hajic, Jan, Otakar Smrz, Petr Zemánek, Jan Šnaidauf, and Emanuel Beška. 2004. Prague Arabic Dependency Treebank: Development in Data and Tools. In The NEMLAR International Conference on Arabic Language Resources and Tools, 110–117. find
  10. Javed, Talha, Nizar Habash, and Dima Taji. 2018. Palmyra: A platform independent dependency annotation tool for morphologically rich languages. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018). find
  11. Maamouri, Mohamed, Ann Bies, Seth Kulick, Michael Ciul, Nizar Habash, and Ramy Eskander. 2014. Developing an Egyptian Arabic Treebank: Impact of Dialectal Morphology on Annotation and Tool Development. In LREC, pp. 2348-2354. find
  12. Maamouri, Mohamed, Ann Bies, Tim Buckwalter, Mona Diab, Nizar Habash, Owen Rambow, and Dalila Tabessi. 2006. Developing and using a pilot dialectal Arabic treebank. In Proceedings of the Fifth International Conference on Language Resources and Evaluation (LREC’06). find
  13. Maamouri, Mohamed, Ann Bies, Tim Buckwalter, and Wigdan Mekki. 2004. The Penn Arabic Treebank: Building a Large-Scale Annotated Arabic Corpus. In NEMLAR conference on Arabic language resources and tools, Vol. 27, 466-467. find
  14. Ma, Xuezhe, Zecong Hu, Jingzhou Liu, Nanyun Peng, Graham Neubig, and Eduard Hovy. 2018. Stack-Pointer Networks for Dependency Parsing. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 1403-1414. find
  15. Najjar, Manal. 2012. Poetic Necessity Between the Syntax of a Sentence and the Syntax of a Text. GSTF Journal of Law and Social Sciences (JLSS), 2(1):322. find
  16. Shahrour, Anas, Salam Khalifa, Dima Taji, and Nizar Habash. 2016. Camelparser: A system for Arabic syntactic analysis and morphological disambiguation. In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: System Demonstrations, pp. 228-232. find our card
  17. Suleiman, Saleh M 1990. The semantic functions of object deletion in classical Arabic. Language Sciences, 12(2-3): 255-266 find
  18. Taji, Dima, and Nizar Habash. 2020. PALMYRA 2.0: A Configurable Multilingual Platform Independent Tool for Morphology and Syntax Annotation. In Proceedings of the Fourth Workshop on Universal Dependencies (UDW 2020), pp. 168-177. find
  19. Taji, Dima, Nizar Habash, and Daniel Zeman. 2017. Universal dependencies for Arabic. In Proceedings of the Third Arabic Natural Language Processing Workshop, pp. 166-176. find our card
  20. Vacareanu, Robert, George Caique Gouveia Barbosa, Marco A. Valenzuela-Escarcega, and Mihai Surdeanu. 2020. Parsing as tagging. In Proceedings of the 12th Language Resources and Evaluation Conference, pp. 5225-5231. find
  21. Zwettler, Michael. 1978. Oral tradition of classical Arabic poetry: its character and implications. The Ohio State University Press. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ghamdi-2021-dependency-treebank-classical-arabic-poetry/