Elamur Research⌕
☰
home / papers / 2016 / Impact de l'agglutination dans l'extraction de termes en arabe standard moderne (Adaptation of a term extractor to the Modern Standard Arabic language)

Impact de l'agglutination dans l'extraction de termes en arabe standard moderne (Adaptation of a term extractor to the Modern Standard Arabic language)

Paper · 2016 · Wafa Neifar, Thierry Hamon, Pierre Zweigenbaum et al.

Summary

The paper tackles term extraction for Modern Standard Arabic. Earlier Arabic systems were poorly documented, not publicly available, and evaluated only on top-ranked terms. The authors port the open-source extractor YATEA, originally built for French and English, to MSA. They rely on MADA+TOKAN for morphological analysis, vowelled lemmas, clitic segmentation, and case and state features.

The adaptation covers YATEA's first two stages: splitting text at syntactic boundaries into maximal noun phrases, then parsing those phrases into head/modifier structures with recursive patterns. Arabic-specific additions include extra boundary items (pseudo-verbs, adverbs, fixed expressions, proclitic-based breaks), and patterns that use gender, number, case and the construct state (al-iḍāfah). Enclitics are left for later.

On a medical corpus of 15,532 words, handling proclitics raises the precision of maximal complex candidates from 65.7% to 72.1% and reduces the number of candidates. The work matters because it gives a reproducible, linguistically grounded baseline for Arabic terminology extraction.

Key points

  • The base tool is YATEA, a freely available Perl extractor with three stages: boundary segmentation, pattern-based head/modifier parsing, and statistical ranking (e.g. C-Value). Only the first two stages were adapted.
  • Preprocessing uses MADA+TOKAN, which supplies vowelled lemmas, clitic decomposition, and case/state marks. These features reduce the damage caused by ambiguous unvowelled forms.
  • Of 400 maximal candidate terms judged manually on medical texts, 288 were domain-correct (72.1%). Without proclitic handling, precision was 65.7%.
  • Remaining errors come from POS tagging mistakes, missing vowels, and the difficulty of handling agglutination. Enclitics are not yet treated.
  • Prepositions that MADA+TOKAN wrongly tags as nouns (e.g. "from", "to", "between") are explicitly forbidden at the start or end of a term.

Interesting ideas

  • I1. Instead of requiring the tagger to be correct, the authors add a list of tags and words that may never open or close a term, which absorbs a known systematic tagger error.
  • I2. Morphological case and construct-state marks serve as direct evidence for which noun is the head and which is the modifier in a nominal group.
  • I3. The paper finds candidate phrases by cutting text at "negative" boundaries (conjugated verbs, pronouns, punctuation, some adverbs and fixed expressions), so maximal noun groups emerge without a full parse.
  • I4. Clitic segmentation is turned on only where it is reliable (proclitics) and deferred where it is error-prone (enclitics), and the gain is measured against the unsegmented baseline.
  • I5. The authors criticise evaluating only top-ranked candidates and instead check a larger sample of maximal candidates, because any method looks good at the top of its ranking.

For Mova

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

  • I1 (tokens): Known recurring tagger confusions could be listed as explicit, explainable edge constraints on chunks, rather than retraining the tagger every time.
  • I2 (tree): English genitive "'s" and "of"-phrases could feed the parser and its repair rules as explicit, traceable head/modifier cues inside noun compounds.
  • I3 (world): A cheap boundary-based chunker could propose domain entity and term candidates for the knowledge layer, and each candidate stays explainable through the boundaries that delimit it.
  • I4 (learn): When adding a preprocessing step, roll it out piecemeal and keep a with/without ablation as a control, so the effect of each step on precision is visible.
  • I5 (learn): Mova's evaluations should sample beyond the most confident outputs, so that strong head-of-list results do not hide weak tails.

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, 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 Wafa Neifar, Thierry Hamon, Pierre Zweigenbaum, Mariem Ellouze and Lamia Hadrich Belguith. 2016. Impact de l'agglutination dans l'extraction de termes en arabe standard moderne (Adaptation of a term extractor to the Modern Standard Arabic language). Actes de la conférence conjointe JEP-TALN-RECITAL 2016. volume 2 : TALN (Posters).

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

@inproceedings{neifar2016impact,
  title = {{Impact de l'agglutination dans l'extraction de termes en arabe standard moderne (Adaptation of a term extractor to the Modern Standard Arabic language)}},
  author = {Wafa Neifar, Thierry Hamon, Pierre Zweigenbaum, Mariem Ellouze and Lamia Hadrich Belguith},
  booktitle = {Actes de la conférence conjointe JEP-TALN-RECITAL 2016. volume 2 : TALN (Posters)},
  year = {2016},
  url = {https://aclanthology.org/2016.jeptalnrecital-poster.20/}
}

References

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

  1. A A. M., T S. & A M. (2013). Arabic term extraction using combined approach on islamic document. Journal of Theoretical & Applied Information Technology, 58(3). find
  2. AK K. & B A. (2010). Automatic extraction of arabic multi-word terms. In IMCSIT, p. 411–418. find
  3. A S. & H T. (2006). Improving term extraction with terminological resources. In T. S, F. G, S. P & T. P , Eds., 5th International Conference on NLP (FinTAL 2006), number 4139 in LNAI, p. 380–387 : Springer. find
  4. B S., D B. & A D. (2008). A multi-word term extraction program for arabic language. In N. C. C. C), K. C, B. M, J. M, J. O, S. P & D. T , Eds., Proceedings of the LREC’08.
  5. B I., E B., E F. & S Y. (2011). Organizing contextual knowledge for arabic text disambiguation and terminology extraction. Knowledge Organization Journal, 38(6), 473–490. find
  6. B I., L W. & E B. (2014). Arabic domain terminology extraction : A literature review - (short paper). In OTM 2014 Conferences - Confederated International Conferences : CoopIS, and ODBASE 2014, p. 792–799, Amantea, Italy. find
  7. B I. & S Y. (2009). A hybrid approach for arabic multi-word term extraction. find
  8. In Natural Language Processing and Knowledge Engineering, NLP-KE 2009. International Conference on, p. 1–8 : IEEE IEEE. find
  9. B D. (1993). An endogeneous corpus-based method for structural noun phrase disambiguation. In Proceedings of the EACL’93, p. 81–86, Utrecht, The Netherlands. find
  10. C M. T., E R. & V J. (2001). Automatic term detection : a review of current systems. In Recent Advances in Computational Terminology. John Benjamins. find
  11. C K. B. & D-F D. (2013). Biomedical Natural Language Processing. John Benjamins publishing company. find
  12. D B. (2003). Conceptual structuring through term variations. In F. B, A. K, D. M. C & A. V , Eds., Proceedings of the ACL’2003 Workshop on Multiword Expressions : Analysis, Acquisition, and Treatment, p. 9–16. find
  13. D P. (2002). Acquisition automatique des termes : l’utilisation des pivots lexicaux spécialisés. find
  14. PhD thesis, Université de Montréal. H N. (2010). Introduction to Arabic Natural Language Processing. Synthesis Lectures on Human Language Technologies. Morgan & Claypool Publishers. find
  15. H N., R O. & R R. (2010). MADA+TOKAN Manual. CCLS-10-01. find
  16. H T., E C. & S S. (2014). Term ranking adaptation to the domain : genetic algorithm based optimisation of the C-Value. In S , Ed., Proceedings of PolTAL 2014 – Advances in Natural Language Processing, volume 8686 of LNAI, p. 71–83. find
  17. K I., K I. P. & M S. (2008). Reviewing and evaluating automatic term recognition techniques. In B. N & A. R , Eds., 6th International Conference on NLP (GoTAL 2008), number 5221 in LNAI, p. 248–259 : Springer. find
  18. M E., G J. L. & H C. (2012). Helping language professionals relate to terms : Terminological relations and termbases. Journal of Specialised Translation, 18. find
  19. M D. & A S. (2000). Identifying terms by their family and friends. In Proceedings of COLING 2000, p. 530–536, Saarbrucken, Germany. find
  20. P M. T., P M. & Z F. (2005). Terminology extraction : An analysis of linguistic and statistical approaches. In S. S , Ed., Knowledge Mining, volume 185 of Studies in Fuzziness and Soft Computing, p. 255–279. Springer Berlin Heidelberg. find
  21. Q. Z B. & H S. (2014). The acl rd-tec : A dataset for benchmarking terminology extraction and classification in computational linguistics. In Proceedings of the 4th International Workshop on Computational Terminology (Computerm), p. 52–63, Dublin, Ireland. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/neifar-2016-impact-de-l-agglutination-dans/