Elamur Research⌕
☰
home / papers / 2012 / Étude comparative entre trois approches de résumé automatique de documents arabes (Comparative Study of Three Approaches to Automatic Summarization of Arabic Documents) [in French]

Étude comparative entre trois approches de résumé automatique de documents arabes (Comparative Study of Three Approaches to Automatic Summarization of Arabic Documents) [in French]

Paper · 2012 · Iskandar Keskes, Mohamed Mahdi Boudabous, Mohamed Hédi Maaloul et al.

Summary

The paper compares three ways of building extractive summaries of Arabic documents, each implemented as its own system. ARSTResume is symbolic and rests on Rhetorical Structure Theory (RST). Hand-built rule "frames" made of discourse markers detect rhetorical relations between minimal text units and label each unit as nucleus or satellite. The system then builds and prunes an RST tree and keeps the nucleus sentences that fit the requested summary type. R.I.A is numerical: it scores sentences and keeps the top-ranked ones. HybridResume combines the two.

All three systems were run on one corpus of one hundred Arabic texts, where three human experts had marked the relevant sentences. The hybrid system scored best, with 73% precision, against 67% for R.I.A and 60% for ARSTResume. The authors report that the symbolic method does relatively better on long texts and the numerical one on short texts, which explains why mixing them helps on a corpus with both. The study matters as an early side-by-side look at how rule-based discourse analysis and score-based extraction trade off for Arabic.

Key points

  • Three systems were compared: ARSTResume (symbolic, RST-based), R.I.A (numerical) and HybridResume (hybrid).
  • The evaluation corpus has 100 Arabic HTML texts, each with relevant sentences chosen by three linguists. The same corpus was also used to derive twenty rhetorical relations, including condition, concession, negation, evidence and confirmation.
  • Reported precision is 60% for ARSTResume, 67% for R.I.A and 73% for HybridResume.
  • The symbolic system lost accuracy when rhetorical relations were absent, when words were ambiguous, and because it had no morphological information. The authors plan to add morphological analysis.
  • Segmentation uses punctuation plus HTML tags and outputs XML with titles, sections, paragraphs and sentences. Sentences are then split into minimal units at rule trigger markers, which matters because Arabic often separates sentences with particles such as waw/fā rather than punctuation.

Interesting ideas

  • I1. Each relation-detection rule pairs a trigger marker, which proposes a relation, with complementary clues searched near the trigger, which confirm or veto it; this two-stage propose-then-verify design keeps rules precise and easy to explain.
  • I2. Content is selected by keeping nucleus units and dropping satellites from a discourse tree, so the summary follows the text's rhetorical structure rather than word statistics.
  • I3. Each frame also states which unit is kept (the retained unit) and where the trigger sits in the segment (start, middle or end), so one declarative rule both detects the relation and fixes the nucleus side.
  • I4. The comparison is broken down by text length, which shows that rule coverage rises with length while score-based extraction falls off, and that a hybrid wins on a mixed corpus.
  • I5. Sentence segmentation treats connective particles as possible sentence boundaries when punctuation alone is unreliable.

For Mova

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

  • I1 (frames): discourse-relation and speech-act detection in prag could use trigger cues plus local confirming or cancelling evidence, and log which clue decided each relation.
  • I2 (back): the text → graph → text compressor could prune satellite-like branches of the meaning graph first when shortening output.
  • I3 (frames): relation templates in prag could carry slots for the cue's position and the dominant segment, making rules readable and testable.
  • I4 (learn): evaluations could be split by document length or other properties, so the evaluation shows where rule components and learned components each fail before they are combined.
  • I5 (tokens): the en sentence splitter could weigh connectives and markup structure alongside punctuation for run-on or poorly punctuated English input.

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 4Research subjects and languages the work is about.Semitic languages · 3,066 worksDiscourse analysis · 471 worksSummarization · 243 worksFrench · 83 works

Cite

Cite as Iskandar Keskes, Mohamed Mahdi Boudabous, Mohamed Hédi Maaloul and Lamia Hadrich Belguith. 2012. Étude comparative entre trois approches de résumé automatique de documents arabes (Comparative Study of Three Approaches to Automatic Summarization of Arabic Documents) [in French]. Proceedings of the Joint Conference JEP-TALN-RECITAL 2012, volume 2: TALN.

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

@inproceedings{keskes2012etude,
  title = {{Étude comparative entre trois approches de résumé automatique de documents arabes (Comparative Study of Three Approaches to Automatic Summarization of Arabic Documents) [in French]}},
  author = {Iskandar Keskes, Mohamed Mahdi Boudabous, Mohamed Hédi Maaloul and Lamia Hadrich Belguith},
  booktitle = {Proceedings of the Joint Conference JEP-TALN-RECITAL 2012, volume 2: TALN},
  year = {2012},
  url = {https://aclanthology.org/F12-2017/}
}

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 MINI M.R.(2001). Apprentissage Automatique et Recherche d’information: Application à l’extraction d’information de surface et au résumé de texte. Thèse de doctorat, université Paris-6 France. find
  2. A SHER N.(1993). Reference to Abstract Objects in Discourse. Kluwer Academic Publishers, Netherlands. find
  3. Azmi A.M. et Al-Thanyyan S.(2012). A Text Summarizer for Arabic. Computer Speech & Language. ISSN :0885-2308. find
  4. B ELGUITH H.L., B ACCOUR L. et M OURAD G.(2005). Segmentation de textes arabes basée sur l'analyse contextuelle des signes de ponctuations et de certaines particules. 12ème conférence sur le Traitement Automatique des Langues Naturelles (TALN’2005), Dourdan, France, 6-10 juin 2005, pp 451–456. find
  5. B OUDABOUS M.M., M AALOUL , M.H. et B ELGUITH H. L.(2010). Digital Learning for Summarizing ARABIC Documents . IceTAL, Islande. find
  6. I RAKY K., Z AKAREYA A. et F ARAWILA A.(2011). Arabic Discourse Segmentation Based on Rhetorical Methods. International Journal of Electric & Computer Sciences IJECS-IJENS Vol: 11 No: 01. find
  7. I RIA C., S ILVIA F., P ATRICIA v., V IVALDI J., S AN J UAN E. et T ORRES -M ORENO J. M.(2007). A new hybrid summarizer based on Vector Space Model, Statistical Physics and Linguistics. Lecture Notes in Computer Science 4827. 872-882. ISSN 0302-9743. find
  8. K AMP H. et R EYEL U.(1993), From Discourse To Logic , Dordrecht Kluwer. find
  9. K AMP H.(1981). Evénements, représentations discursives et référence temporelle. Langages, p 34-64. find
  10. K ESKES I.(2011). Résumé automatique de textes arabes basé sur une approche symbolique. Editeur : EUE. ISBN-13 : 978-3841780232 find
  11. K ESKES I. et M AALOUL M. H.(2010). Résumé automatique de documents arabes basé sur la technique RST . Conférence international de Rencontre des Étudiants Chercheurs en Informatique pour le Traitement Automatique des Langues (TALN /RECITAL 2010), 12ème edition, Montréal – Canada. find
  12. K ESKES I., M AALOUL M. H. et B ELGUITH L. H.(2010) ,(a). 7 اد / ا . 9!ا ص ! !4ا 5 6 !ا . :;<!ا . <!ا .= > . International Computing Conference in Arabic, 6ème édition, Hammamet – Tunisie, prix du Best Paper. find
  13. K ESKES I., M AÂLOUL M. H., B ELGUITH L. H. et B LACHE P.(2010), (b). Automatic summarization of Arabic texts based on RST technique. International Conference on Enterprise Information Systems, 12ème edition, Madeira – Portugal. find
  14. L ASCARIDES A. et A SHER N.(1993), Temporal Interpretation, Discourse Relations, and Commonsense Entailment , Linguistics and Philosophy, 16(5). find
  15. M AÂLOUL M. H.(2007). Al Lakas El’eli / ص 6ّ !ا !4ا : Un système de résumé automatique de documents arabes . IBIMA. find
  16. M ANN W. C. et T HOMPSON S. A.(1988). Rhetorical structure theory: Toward a functional theory of text organization . Text, 8(3), p 243 – 281.
  17. M ARCU D.(1999). Discourse trees are good indicator of importance in text, Advances in Automatic Text Summarization. p123 – 136. find
  18. M INEL J.L.( 2002). Filtrage sémantique : du résumé automatique à la fouille de textes. Hermès Science Publications, Paris. find
  19. M OURAD G.(1999). La segmentation de textes par l’étude de la ponctuation. CIDE’99, Document Electronique Dynamique, p 155 – 171, Damas, Syrie. find
  20. N ICOLAS U., A MINI M.R. et G ALLINARI P.(2005). Résumé automatique de texte avec un algorithme d'ordonnancement . CORIA. find
  21. T OFILOSKI M., B ROOKE J. et T ABOADA M.(2009). A Syntactic and Lexical-Based Discourse Segmenter. In Proceedings of the 47th Annual Meeting of the Association for Computational Linguistics. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/keskes-2012-etude-comparative-entre-trois-approches/