Elamur Research⌕
☰
home / papers / 2014 / The International Corpus of Arabic: Compilation, Analysis and Evaluation

The International Corpus of Arabic: Compilation, Analysis and Evaluation

Paper · 2014 · Sameh Alansary and Magdy Nagi

Summary

Arabic NLP has long lacked a large, balanced and publicly searchable corpus. Earlier resources were small (PATB, NEMLAR), restricted to news (Arabic Gigaword, KALIMAT), or large but unanalyzed (KACST, ArabiCorpus). Bibliotheca Alexandrina started the International Corpus of Arabic (ICA) in 2006 to fill this gap. ICA covers written Modern Standard Arabic from every Arab country and from Arabic publications abroad. Its target is 100 million words, and about 80 million had been collected at the time of the paper.

Texts were gathered and sorted into genres by hand from four source types: press, net articles, books and academic writing. Genre proportions follow how much each genre is actually used, not equal quotas. The corpus was analyzed with BAMAE, an enhancer built on the Buckwalter Arabic Morphological Analyzer. On one million words it reached precision of 92–95% and recall of 89–92%. In a feature-by-feature comparison, BAMAE outperformed MADA. The analyzed corpus is offered through a free web search interface, since copyright prevents downloading the texts.

Key points

  • ICA is planned at 100 million words of written MSA, with roughly 80 million compiled so far. Its genre scheme has 11 genres, 24 subgenres and 4 sub-sub-genres, adapted from the Dewey decimal classification (for example, history and geography are split).

  • Balance follows real-use distribution: Literature makes up about 12% and Biography about 2%. Egyptian MSA is the most widely represented variety.

  • BAMAE scores, evaluated on 1,000,000 words:

    • precision 95–92% and recall 92–89%, depending on how many qualifiers are retrieved per word;
    • root 99.45%, stem pattern 94.34% and lemma 96.54% (MADA does not provide these qualifiers).

  • Accuracy against MADA on the same ICA text:

    QualifierBAMAEMADA
    Diacritization89.61%78.78%
    Tags93.94%85.28%
    Stems96.97%91.34%
    Number96.10%64.93%
    Gender96.53%66.67%
    Definiteness96.53%60.61%

  • The ICA website supports exact-match, lemma, root and stem search. Results can be filtered by word class, stem pattern, number, gender, definiteness, country, source and genre.

Interesting ideas

  • I1. Corpus balance is defined by how often each genre and source actually occurs, not by fixed equal slices, so the sample resembles real usage.
  • I2. Before comparing two analyzers, the authors map one system's labels onto the other's (e.g. singular "s" → "SG") and then score each feature separately. This shows exactly where each system fails instead of giving one blended number.
  • I3. Several error types are traced to passing the base analyzer's output through without refining it, such as coarse adverb tags or ignoring morphosyntax for number and gender. A post-analysis enhancement layer over a rule-based analyzer fixed much of this.
  • I4. Lemma search on an ambiguous surface form first lists every possible lemma and lets the user choose, instead of silently picking one.
  • I5. Corpus design is treated as a cyclical process that is re-evaluated during compilation rather than fixed at the start.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (learn): weight silver-data and evaluation sets by the real frequency of text types (tales, school texts, word problems), so scores reflect what Mova will actually meet.
  • I2 (tree): when comparing witness parsers or converting UD dialects, normalize the label inventories first and report per-relation or per-feature agreement.
  • I3 (tokens): add an explicit, explainable correction pass over the tagger and morphology output that repairs known systematic confusions.
  • I4 (tokens): keep the full set of candidate lemmas and analyses visible with the chosen one, so later stages and explanations can show what alternatives were rejected.
  • I5 (learn): treat the curriculum and data mix as revisable checkpoints, re-auditing coverage after each training round.

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

 Institutions

🎓Alexandria UniversityAlexandria · Egypt42 works · 287 citations🔬Bibliotheca AlexandrinaAlexandria · Egypt3 works · 36 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 3,066 works

Cite

Cite as Sameh Alansary and Magdy Nagi. 2014. The International Corpus of Arabic: Compilation, Analysis and Evaluation. Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/v1/w14-3602

@inproceedings{alansary2014international,
  title = {{The International Corpus of Arabic: Compilation, Analysis and Evaluation}},
  author = {Sameh Alansary and Magdy Nagi},
  booktitle = {Proceedings of the EMNLP 2014 Workshop on Arabic Natural Language Processing (ANLP)},
  year = {2014},
  doi = {10.3115/v1/w14-3602},
  url = {https://doi.org/10.3115/v1/w14-3602}
}

References

19 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 Abdelali, James Cowie&Hamdy S. Soliman. 2005, Building a modern standard Arabic corpus. In Proceedings of workshop on computational modeling of lexical acquisition. The split meeting. Croatia, (25-28 July). CaminoR. find
  2. Rizzo. 2010, Getting on with corpus compilation: From theory to practice. In ESP World Journal, Issue 1 (27), Volume 9. find
  3. Charles F. Meyer. 2002, English Corpus Linguistics: An Introduction, Cambridge University Press. find
  4. Hajic J., Smrz O., Zemánek P., Šnaidauf J., &Beška E. (2004, September), Prague Arabic Dependency Treebank: Development in Data and Tools. InProc. of the NEMLAR Intern. Conf. on Arabic Language Resources and Tools (pp. 110-117). find
  5. Jan Hajic J., OtkarSmrz O., Petr Zemánek P., Jan Šnaidauf J., &Emanuel Beška E. 2004, (2004, September), Prague Arabic Dependency Treebank: Development in Data and Tools. In Proceedings of the NEMLAR Intern. Conf. on Arabic Language Resources and Tools (pp. 110-117). (2004, September). Jan Hajic, OtakarSmrz, Tim Buckwalter ,& Hubert Jin. find
  6. September. 2005, Feature-based tagger of approximations of functional Arabic morphology. In Proceedings of the Workshop on Treebanks and Linguistic Theories (TLT), Barcelona, Spain. find
  7. John Sinclair. 1991, Corpus, Concordance and Collocation (Describing English Language). Oxford University Press. find
  8. Mahmoud El-Haj & Rim Koulali. 2013, KALIMAT a Multipurpose Arabic Corpus. In Proceedings of the 2 nd Workshop on Arabic Corpus Linguistics (WACL-2) . find
  9. Mohamed Maamouri, Ann Bies, Tim Buckwalter &WegdanMekki. 2004, The Penn Arabic Treebank: Building a Large-Scale Annotated Arabic Corpus, In Proceedings of NEMLAR conference on Arabic language resources and tools (pp. 102-109). find
  10. Muhammad Atiyya, Khalid Choukri& Mustafa Yaseen. (2005), Specifcations of the Arabic Written Corpus. NEMLAR Project. September 29 th 2005. find
  11. Nizar Habash , Owen Rambow, & Ryan Roth. 2009, MADA+ TOKAN: A Toolkit for Arabic Tokenization, Diacritization, Morphological Disambiguation, POS Tagging, Stemming and Lemmatization. In Proceedings of the 2 nd International Conference on Arabic Language Resources and Tools (MEDAR), Cairo, Egypt.‏ Nizar Habash and Owen Rambow. 2005, Arabic Tokenization, Part-Of-Speech Tagging and Morphological Disambiguation in One Fell Swoop. In Proceedings of ACL’05, Ann Arbor, MI, USA.
  12. Petter Zemanek. 2001, CLARA (Corpus Linguae Arabica): An Overview. In Proceedings of ACL/EACL Workshop on Arabic Language. find
  13. Piotr Pęzik. 2010, New Ways to Language. Chapter 21 (pp. 433-46), WydawnictwoUniwersytetuŁódzkie-go. find
  14. Ryan Roth, Owen Rambow, Nizar Habash, Mona Diab, and Cynthia Rudin. 2008, Arabic Morphological Tagging, Diacritization, and Lemmatization Using Lexeme Models and Feature Ranking. In Proceedings of ACL-08: HLT, Short Papers (Companion Volume), pages (117–120), Columbus, Ohio, USA, June 2008.
  15. Sameh Alansary, Magdy Nagi & Noha Adly. 2008, Towards Analysing the International Corpus of Arabic (ICA): Progress of Morphological Stage. In Proceedings of 8th International Conference on Language Engineering, Egypt. find
  16. Sameh Alansary. 2012, BAMAE: Buckwalter Arabic Morphological Analyser Enhancer. in Proceedings of 4 th international conference on Arabic language processing, Mohamed Vth University Souissi, Rebate, Morocco, May 2-3 2012. find
  17. Sue Atkins S., Jeremy Clear J.& Nicholas Ostler N. (1992), Corpus Design Criteria, Literary and linguistic computing, 7(1), 1-16. find
  18. Tim Buckwalter. 2004, Buckwalter Arabic Morphological Analyzer Version 2.0. Linguistic Data Consortium (LDC) catalogue number LDC2004L02, ISBN 1-58563-324-0. ↗
  19. WajdiZaghouani. 2014, Critical Survey of the Freely Available Arabic Corpora. In Proceedings of Workshop on Free/Open-Source Arabic Corpora and Corpora Processing Tools.‏ find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/alansary-2014-international-corpus-arabic-compilation-analysis/