Elamur Research⌕
☰
home / papers / 2018 / Putting Figures on Influences on Moroccan Darija from Arabic, French and Spanish using the WordNet

Putting Figures on Influences on Moroccan Darija from Arabic, French and Spanish using the WordNet

Paper · 2018 · Khalil Mrini and Francis Bond

Summary

The paper asks how much of Moroccan Darija's vocabulary resembles Standard Arabic, French and Spanish, and it tries to answer with numbers rather than impressions. The authors use the Moroccan Darija Wordnet (MDW), which is linked into the Open Multilingual WordNet (OMW). Within each shared synset, they compare Darija lemmas with the Arabic, French and Spanish lemmas for the same sense.

The MDW uses a Latin alphabet in which each letter stands for one sound. Each Darija letter is given a set of possible spellings in each target language, every combination is generated, and the closest match is kept. Distance is the Levenshtein edit distance divided by the length of the longer word. Multiword expressions are left out.

Over the automatically linked synsets, about 42% have an Arabic match that is at least 60% similar. For French the figure is about 2.8%, and for Spanish about 2.2%. A manually validated subset gives similar numbers. The Darija–Arabic distance is close to the Portuguese–Galician distance, which suggests that Darija could be treated as a separate language. The work also shows a cheap way to measure lexical closeness using only aligned wordnets.

Key points

  • Data:
    • 12,224 MDW synsets were automatically linked to Princeton WordNet, and 2,319 of these were manually verified.
    • The validation set contains 617 verified synsets that link to all three other wordnets.
    • The comparison wordnets were the Arabic WordNet, WOLF for French, and the Spanish wordnet from the MCR.
  • Average normalised Levenshtein distance (automatic links): 0.4619 to Arabic, 0.7337 to French and 0.7521 to Spanish. On the validated subset the figures were 0.4393, 0.7544 and 0.7721.
  • Share of synsets with at least one pair ≥60% similar: On the automatic links, the shares were 42.02% (Arabic), 2.79% (French) and 2.18% (Spanish). On the validated subset the shares were 47.00%, 3.08% and 2.92%.
  • French–Spanish overlap: 278 French and 188 Spanish synsets passed the threshold, and 95 of them were shared. Part of the European "influence" therefore reflects how similar French and Spanish are to each other.
  • Reference pair and leftovers:
    • The Portuguese–Galician distance was 0.4760.
    • Roughly 2.7k Darija synsets matched none of the three languages; the paper gives both 2,736 and 2,763.
    • Many of these unmatched words look like Tamazight (for example, "ta-…-t" forms). Others reveal linking errors.

Interesting ideas

  • I1. Compare words only within a shared sense (a synset), not across the whole vocabulary, so that cross-language similarity is measured between true translation equivalents rather than random lookalikes.
  • I2. Give each source letter a set of possible renderings, enumerate the combinations, and keep the smallest distance; the authors treat this as an openly optimistic upper bound on similarity.
  • I3. Drop multiword expressions from edit-distance comparison, because spaces and word-order changes corrupt the metric; keep a measure only where its assumptions hold.
  • I4. Rerun the whole measurement on a small hand-verified subset to check whether noise in automatic links changes the conclusions.
  • I5. Read a raw distance against a known reference pair (Portuguese–Galician) and apply a similarity threshold to filter chance matches.

For Mova

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

  • I1 (world): When aligning WordNet-derived ontology entries with Wikidata or FrameNet labels, restrict string matching to candidates already tied to the same concept, which cuts spurious merges.
  • I2 (tokens): Lemma lookup for misspellings, British/American variants or transliterated names could generate a small set of spelling variants and keep the closest dictionary hit, while recording which variant matched for explainability.
  • I3 (learn): Evaluation scripts that score lemma or token agreement between witness parsers should exclude or separately score MWEs, so their noise does not distort the main metric.
  • I4 (learn): Recompute every silver-data statistic on a gold slice and report both, so that a large gap flags a noisy silver pipeline.
  • I5 (learn): Calibrate parser-agreement or similarity scores against a deliberately chosen baseline pair, which serves as a negative or reference control, rather than reading absolute numbers in isolation.

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

🎓Nanyang Technological UniversitySingapore · Singapore156 works · 25,644 citations🎓École Polytechnique Fédérale de LausanneLausanne · Switzerland44 works · 7,050 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksFrench · 70 works

Cite

Cite as Khalil Mrini and Francis Bond. 2018. Putting Figures on Influences on Moroccan Darija from Arabic, French and Spanish using the WordNet. Proceedings of the 9th Global Wordnet Conference.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2018.gwc-1.46

@inproceedings{mrini2018putting,
  title = {{Putting Figures on Influences on Moroccan Darija from Arabic, French and Spanish using the WordNet}},
  author = {Khalil Mrini and Francis Bond},
  booktitle = {Proceedings of the 9th Global Wordnet Conference},
  year = {2018},
  doi = {10.18653/v1/2018.gwc-1.46},
  url = {https://doi.org/10.18653/v1/2018.gwc-1.46}
}

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. Lahsen Abouenour, Karim Bouzoubaa, and Paolo Rosso. 2013. On the evaluation and improvement of Arabic wordnet coverage and usability. Language Resources and Evaluation, 47(3):891–917. ↗
  2. Diana Inkpen, Oana Frunza, and Grzegorz Kondrak. 2005. Automatic identification of cognates and false friends in french and english. In Proceedings of the International Conference Recent Advances in Natural Language Processing, volume 9, pages 251–257. find
  3. Joseph Aquilina. 1958. Maltese as a mixed language. Journal of Semitic Studies, 3(1):58. ↗
  4. Vladimir Levenshtein, editor. 1996. Binary codes capable of correcting deletions, insertions, and reversals. find
  5. Joseph Aquilina. 1972. Maltese christian words of arabic origin. find
  6. Shawn Magin. 2010. Illiteracy in the arab region: A meta study. GIA Lens, 2. find
  7. W. Black, S. Elkateb, H. Rodriguez, M. Alkhalifa, P. Vossen, A. Pease, M. Bertran, and C. Fellbaum. 2006. The arabic wordnet project. In Proceedings of LREC 2006. find
  8. Karla Mallette. 2011. European Modernity and the Arab Mediterranean: Toward a New Philology and a Counter-Orientalism. University of Pennsylvania Press. find
  9. Francis Bond and Ryan Foster. 2013. Linking and extending an open multilingual wordnet. In 51st Annual Meeting of the Association for Computational Linguistics: ACL-2013. Sofia, page 1352–1362.
  10. Khalil Mrini and Francis Bond. 2017. Building the moroccan darija wordnet (mdw) using bilingual resources. In Proceedings of the International Conference on Natural Language, Signal and Speech Processing (ICNLSSP), Casablanca, Morocco. find
  11. Joseph M Brincat. 2005. Maltese–an unusual formula. MED Magazine, 27. find
  12. Alina Maria Ciobanu and Liviu P. Dinu. 2014. An etymological approach to cross-language orthographic similarity. application on romanian. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), page 1047–1058. find
  13. Valéria de Paiva, Alexandre Rademaker, and Gerard de Melo. 2012. OpenWordNet-PT: an open Brazilian Wordnet for reasoning. EMAp technical report, Escola de Matemática Aplicada, FGV, Brazil. find
  14. Moha Ennaji. 2005. Multilingualism, cultural identity, and education in Morocco. Springer Science & Business Media. ↗
  15. Christiane Fellbaum. 1998. Wordnet: An electronic lexical database. MIT Press. find
  16. Aitor Gonzalez-Agirre, Egoitz Laparra, and German Rigau. 2012. Multilingual central repository version 3.0: upgrading a very large lexical knowledge base. In Proceedings of the 6th Global WordNet Conference (GWC 2012), Matsue. find
  17. Robert Lane Greene. 2013. Arabic: A language with too many armies and navies? The Economist. find
  18. Richard S. Harrell. 1963. A dictionary of moroccan arabic: Moroccan-english. Georgetown University Press. find
  19. Haut Commissariat au Plan du Maroc HCP. 2014. Recensement de la population. find
  20. Wilbert Heeringa, Peter Kleiweg, Charlotte Gooskens, and John Nerbonne. 2006. Evaluation of string distance algorithms for dialectology. In Proceedings of the workshop on linguistic distances, pages 51–62. Association for Computational Linguistics. find
  21. Benoît Sagot and Daria Fišer. 2008. Building a free french wordnet from multilingual resources. In Proceedings of the Sixth International Language Resources and Evaluation (LREC’08), Marrakech, Morocco. find
  22. Yves Scherrer. 2012. Recovering dialect geography from an unaligned comparable corpus. In Proceedings of the EACL 2012 Joint Workshop of LINGVIS and UNCLH, Avignon, France, pages 63–71. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mrini-2018-putting-figures-influences-moroccan-darija/