Elamur Research⌕
☰
home / papers / 2016 / Compilation of an Arabic Children's Corpus

Compilation of an Arabic Children's Corpus

Summary

The authors address the lack of any corpus of Arabic texts aimed at children. They model their work on the Oxford Children's Corpus and define children's material as texts either written for young readers or chosen for them by adults. Over three months, two native speakers located about 70 websites, including children's magazines, forums and blogs, and collected stories by hand.

They first tried WebBootCat, but rejected it. Its boilerplate stripping cut off the start or end of stories, and harvesting by URL pulled in unrelated forum content. Each text was cleaned of formatting and given a metadata header with URL, title, author, genre and dialect. The texts were then loaded into SketchEngine and deduplicated.

The result, ACC version 1.0, has 2950 documents and roughly 1.88 million tokens. Genres are organised in a shallow hierarchy under Fiction and Non-Fiction. It is the first such resource for Arabic. It supports genre classification, comparison of child and adult vocabulary, ideology studies, and the design of reading materials.

Key points

  • The Arabic Children's Corpus (ACC) v1.0 contains 2950 documents and 1,877,615 word tokens, all drawn from the web during a 3-month project.
  • Scanned PDFs and audio/video sites were excluded because Arabic OCR on low-resolution scans needed too much correction. Scarce texts for beginner readers were typed in manually to improve balance.
  • WebBootCat was tested in both modes (seed-term queries and URL upload) and found unsuitable. Collection was therefore manual copy-paste, with provenance recorded in a spreadsheet so the two collectors would not duplicate each other.
  • Each story has a header with URL, title, author, genre and dialect. Genres form a tree of 2–3 levels, for example Fiction → Folktale → Arabian Nights. The corpus is hosted in SketchEngine, with duplicates removed using the Onion tool plus manual checks.
  • The authors report several open issues:
    • missing or uncertain authorship;
    • texts whose original or translated status is unclear;
    • spelling and grammar errors;
    • inconsistent vowelisation;
    • genre boundaries that blur, such as a moral fable or a biography containing anecdotes.

Interesting ideas

  • I1. When an automatic cleaner silently drops the beginning or end of documents, the authors chose slower manual collection that guarantees complete texts; completeness of each document can matter more than corpus size.
  • I2. Every text carries a provenance and genre header, and genre is a multi-level path rather than a single flat label, so you can query at any level of granularity.
  • I3. The authors deliberately filled a gap (very little web text for beginner readers) by hand-entering material, rather than accepting the skew of what the web happens to offer.
  • I4. Comparing children's and adult corpora by vocabulary and formulaic sequences shows which words and fixed phrases are typical for young readers.
  • I5. They note that unproofread errors (merged or split words, missing prepositions) and inconsistent diacritics complicate later analysis, so teaching-oriented corpora may need correction even though real-usage corpora usually are not edited.

For Mova

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

  • I1 (world): Stories and fables need intact openings and endings to track entities and event order, so ingestion should verify full-text integrity before building a text's world.
  • I2 (answer): Tagging fable/tale QA sets with hierarchical genre paths would let evaluation report scores per text type (folktale vs. biography vs. instructions).
  • I3 (learn): Silver data from parsers will over-represent common registers; deliberately adding simple child-level texts would balance training and evaluation toward the school-course readers Mova targets.
  • I4 (back): Lists of child-register vocabulary and formulas extracted this way could constrain generation from the meaning graph when producing simple, age-appropriate English.
  • I5 (tokens): Keeping a flagged record of tokenisation-level errors (fused or split words) in input texts would help the tokenizer and tagger handle noisy children's web text without corrupting the UD tree downstream.

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 Semitic languages

Cite

Cite as Latifa Al-Sulaiti, Noorhan Abbas, Claire Brierley, Eric Atwell and Ayman Alghamdi. 2016. Compilation of an Arabic Children's Corpus. Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16).

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

@inproceedings{alsulaiti2016compilation,
  title = {{Compilation of an Arabic Children's Corpus}},
  author = {Latifa Al-Sulaiti, Noorhan Abbas, Claire Brierley, Eric Atwell and Ayman Alghamdi},
  booktitle = {Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)},
  year = {2016},
  url = {https://aclanthology.org/L16-1285/}
}

References

22 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abbas, N., Atwell, E. and Al-Sulaiti, L. 2016. ‘An Empirical Evaluation of 10 Arabic Optical Character Recognition Tools’. International Journal of Computational Linguistics (IJCL), 7. 1-14. find
  2. Alfaifi, A., Atwell E, Abuhakema G. 2013. ‘Error Annotation of the Arabic Learner Corpus: A New Error Tagset’ in: Language Processing and Knowledge in the Web, LNCS Lecture Notes in Computer Science vol. 8105, pp.14-22. Springer. find
  3. Alghamdi, A., Atwell, E. and Brierley, C. 2016. ‘An empirical study of Arabic formulaic sequence extraction methods’ in Proceedings of LREC Language Resources and Evaluation Conference. our card
  4. Al-Hajji, F.A. 1990. ‘al-Dalil al-Biblioghrafi Likitab at-Tifl al-‘Arabi. Bibliographical Guide to Arab Children’s Books’. Sharjah. Dairatu al-Thakafa wal-‘alam. find
  5. Al-Hazza, T.C. 2006. ‘Arab Children’s Literature: An Update’. In Book Links. 15.3. 11-12. find
  6. Al-Hazza, T.C. and Lucking, B. 2012. ‘Celebrating Diversity through Explorations of Arab Children’s Literature’. In Childhood Education. 83.3. 132-135. find
  7. Atwell, E. 1987. ‘How to detect grammatical errors in a text without parsing it’ in: Maegaard, B., (editor) Proceedings of EACL '87 Third conference on European chapter of the Association for Computational Linguistics, 38-45. find
  8. Baroni, M., Kilgarriff, A., Pomikálek, J., and Rychlý, P. 2006. ‘WebBootCaT: instant domain-specific corpora to support human translators’. In Proceedings of EAMT European Association for Machine Translation. 247-252. find
  9. Dimitrova, L. and Koseska-Toszewa, V. 2009. ‘Bulgarian-Polish Corpus’. In Cognitive Studies. 9. 133-141. find
  10. Heide, F.P. and Gilliland, J. 1990. The day of Ahmed’s secret. New York. Lothrop, Lee and Shepard Books. find
  11. Heide, F.P. and Gilliland, J. 1992. Sami and the time of the troubles. New York. Clairon Books. find
  12. Kilgarriff, A., Baisa, V., Bušta, J., Jakubíček, M., Vojtěch, K., Michelfeit, J., Rychlý, P. and Suchomel, V. 2014. ‘The Sketch Engine: ten years on’. In Lexicography. 1.1. 7-36. find
  13. Mauranen, A. 2000. ‘Strange strings in translated language. A study on corpora’. In Olohan, M. (ed.), Intercultural Faultlines: Research Models in Translation Studies. I. Textual and Cognitive Aspects. Manchester: St. Jerome. 119-141. find
  14. Mdallel, S. 2003. ‘Translating Children’s Literature in the Arab World’. In Meta: Translators’ Journal. 48.1-2. 298-306. find
  15. Peterson, M.A. 2005. ‘The Jinn and the Computer: Consumption and identity in Arabic children’s magazines’. In Childhood. 12.2. 177-200. find
  16. Puurtinen, T. 2003. ‘Genre-specific Features of Translationese? Linguistic Differences between Translated and Non-translated Finnish Children’s Literature’. In Literary and Linguistic Computing. 18.4. 389-406. find
  17. Puurtinen, T. 1998. ‘Syntax, Readability and Ideology in Children’s Literature’. In Meta: Translators’ Journal. 43.4. 524-533. find
  18. Thompson, P. and Sealey, A. 2007. ‘Through children’s eyes?’ In International Journal of Corpus Linguistics. 12.1. 1-23. find
  19. Shefelman, J. 1992. A peddler’s dream. Austin, Texas. Houghton Mifflin. find
  20. Stanley, D. 2002. Saladin: Noble prince of Islam. New York. HarperCollins. find
  21. Wild, K., Kilgarriff, A. and Tugwell, D. 2012. ‘The Oxford Children’s Corpus: Using a Children’s Corpus in Lexicography’. In International Journal of Lexicography. doi: 10.1093/ijl/ecs017. OUP. 1-29. ↗
  22. Wodak, R. and Meyer, M. (eds). 2001. Methods of Critical Discourse Analysis. London. SAGE. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sulaiti-2016-compilation-arabic-children-s-corpus/