Elamur Research⌕
☰
home / papers / 2008 / Semantic Role Labeling Systems for Arabic using Kernel Methods

Semantic Role Labeling Systems for Arabic using Kernel Methods

Paper · 2008 · Mona Diab, Alessandro Moschitti and Daniele Pighin

Summary

This paper addresses Semantic Role Labeling (SRL) for Modern Standard Arabic. Arabic is hard for role labeling because it has rich templatic morphology, subject pro-drop, relatively free VSO/SVO order and recursive idafa possessive noun phrases. The authors hypothesize that explicit morphological cues help identify subjects, objects and argument scope. These cues are Case, Gender, Number and Definiteness agreement.

Their system is a two-stage SVM pipeline. It first detects argument boundaries over parse-tree nodes, then assigns roles with one-vs-all classifiers. It is trained and tested on the pilot Arabic Propbank from SemEval 2007 Task 18, using gold Arabic Tree Bank (ATB) trees. The system combines standard English-style attribute features, via polynomial kernels, with tree kernels over structured argument representations. These structures are AST and a morphologically enriched variant, EAST.

The best combination, AST+EAST+Poly3, reaches 82.17% global F1. This is roughly 2 points above the previous polynomial-kernel state of the art for Arabic. The result shows that exploiting morphology through structured features pays off for morphologically rich languages, even with small training data.

Key points

  • Data: the pilot Arabic Propbank (SemEval 2007 Task 18), with gold-standard ATB parse trees, because no reliable Arabic deep parser was available.
  • Language facts motivating the design:
    • About 35% of ATB sentences have a dropped subject.
    • VSO and SVO each occur around 30% of the time in the ATB.
  • Complete SRL results (P/R/F1):
    • Poly3: 81.73/78.93/80.31
    • AST: 78.09 F1
    • EAST: 79.51 F1
    • AST+Poly3: 80.86 F1
    • EAST+Poly3: 81.56 F1
    • AST+EAST+Poly3: 83.08/81.28/82.17
  • Where each component helps:
    • Polynomial degree 2–4 beats the linear kernel by over 3 points, while degrees 5–6 collapse.
    • EAST barely differs from AST for boundary detection.
    • EAST gains about 3 points over AST for role classification with gold boundaries.
  • Per-role F1: ARG0 reaches 96.70 but ARG1 only 90.57. The authors attribute the gap to word-order ambiguity between subject and object when morphology gives no clear signal.

Interesting ideas

  • I1. Feeding morphological agreement (Case, Gender, Number, Definiteness) directly into the structure used for role classification helps decide which phrase is subject versus object and how far an argument extends.
  • I2. The authors split the task into a boundary detector over all tree nodes, followed by per-role one-vs-all classifiers, so each stage can be measured separately (e.g., classification with gold boundaries).
  • I3. Summing a structural tree kernel with a kernel over hand-designed features beat either alone, suggesting that structure and attribute features capture complementary signals.
  • I4. Morphological enrichment helped role classification much more than boundary detection, which was attributed to the use of gold trees, so where a feature helps depends on the subtask.
  • I5. Per-role breakdowns exposed weak labels: rare adjuncts and ARG1 confusion caused by free word order.

For Mova

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

  • I1 (frames): When mapping UD arguments to VerbNet roles, agreement and case features from the tree could serve as explicit, explainable evidence for choosing between competing role assignments.
  • I2 (learn): Evaluating role assignment with gold spans versus predicted spans would show whether frame errors come from the parser or from the role mapper.
  • I3 (frames): Role decisions could combine path- or subtree-pattern evidence with symbolic feature rules, keeping both as separately inspectable votes.
  • I4 (learn): Ablations should report per stage and per role, with oracle inputs as a control, rather than a single global score.
  • I5 (answer): Per-role error tables on fable and test QA could reveal which question types fail because of subject–object confusion in the frames.

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

🎓Columbia UniversityNew York · United States241 works · 54,053 citations🎓University of TrentoTrento · Italy85 works · 15,524 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 3Research subjects and languages the work is about.Semitic languages · 2,908 worksMorphology · 405 worksSemantic role labeling · 145 works

Cite

Cite as Mona Diab, Alessandro Moschitti and Daniele Pighin. 2008. Semantic Role Labeling Systems for Arabic using Kernel Methods. Proceedings of ACL-08: HLT.

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

@inproceedings{diab2008semantic,
  title = {{Semantic Role Labeling Systems for Arabic using Kernel Methods}},
  author = {Mona Diab, Alessandro Moschitti and Daniele Pighin},
  booktitle = {Proceedings of ACL-08: HLT},
  year = {2008},
  url = {https://aclanthology.org/P08-1091/}
}

References

29 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Collin F. Baker, Charles J. Fillmore, and John B. Lowe. 1998. The Berkeley FrameNet Project. In COLING-ACL ’98: University of Montréal.
  2. Xavier Carreras and Lluı́s Màrquez. 2005. Introduction to the CoNLL-2005 Shared Task: Semantic Role Labeling. In Proceedings of CoNLL-2005, Ann Arbor, Michigan.
  3. John Chen and Owen Rambow. 2003. Use of Deep Linguistic Features for the Recognition and Labeling of Semantic Arguments. In Proceedings of EMNLP, Sapporo, Japan. find
  4. Michael Collins and Nigel Duffy. 2002. New Ranking Algorithms for Parsing and Tagging: Kernels over Discrete structures, and the voted perceptron. In ACL02. find
  5. Mona Diab and Alessandro Moschitti. 2007. Semantic Parsing for Modern Standard Arabic. In Proceedings of RANLP, Borovets, Bulgaria. find
  6. Mona Diab, Musa Alkhalifa, Sabry ElKateb, Christiane Fellbaum, Aous Mansouri, and Martha Palmer. 2007a. Semeval-2007 task 18: Arabic Semantic Labeling. In Proceedings of SemEval-2007, Prague, Czech Republic.
  7. Mona Diab, Mahmoud Ghoneim, and Nizar Habash. 2007b. Arabic Diacritization in the Context of Statistical Machine Translation. In Proceedings of MT-Summit, Copenhagen, Denmark.
  8. Mona Diab. 2007. Towards an Optimal Pos Tag Set for Modern Standard Arabic Processing. In Proceedings of RANLP, Borovets, Bulgaria. find
  9. Katrin Erk and Sebastian Pado. 2006. Shalmaneser – A Toolchain for Shallow Semantic Parsing. Proceedings of LREC. find
  10. Daniel Gildea and Daniel Jurafsky. 2002. Automatic Labeling of Semantic Roles. Computational Linguistics. ↗
  11. Daniel Gildea and Martha Palmer. 2002. The Necessity of Parsing for Predicate Argument Recognition. In Proceedings of ACL-02, Philadelphia, PA, USA.
  12. 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, Michigan.
  13. Aria Haghighi, Kristina Toutanova, and Christopher Manning. 2005. A Joint Model for Semantic Role Labeling. In Proceedings ofCoNLL-2005, Ann Arbor, Michigan. find
  14. Paul Kingsbury and Martha Palmer. 2003. Propbank: the Next Level of Treebank. In Proceedings of Treebanks and Lexical Theories. find
  15. Mohamed Maamouri, Ann Bies, Tim Buckwalter, and Wigdan Mekki. 2004. The Penn Arabic Treebank : Building a Large-Scale Annotated Arabic Corpus. find
  16. Mohamed Maamouri, Ann Bies, Tim Buckwalter, Mona Diab, Nizar Habash, Owen Rambow, and Dalila Tabessi. 2006. Developing and Using a Pilot Dialectal Arabic Treebank.
  17. Alessandro Moschitti, Ana-Maria Giuglea, Bonaventura Coppola, and Roberto Basili. 2005. Hierarchical Semantic Role Labeling. In Proceedings of CoNLL-2005, Ann Arbor, Michigan. find
  18. Alessandro Moschitti, Silvia Quarteroni, Roberto Basili, and Suresh Manandhar. 2007. Exploiting Syntactic and Shallow Semantic Kernels for Question Answer Classification. In Proceedings of ACL’07, Prague, Czech Republic. find
  19. Alessandro Moschitti. 2004. A Study on Convolution Kernels for Shallow Semantic Parsing. In proceedings of ACL’04, Barcelona, Spain. find
  20. Alessandro Moschitti. 2006. Making Tree Kernels Practical for Natural Language Learning. In Proceedings of EACL’06. find
  21. Alessandro Moschitti, Daniele Pighin and Roberto Basili. 2006. Semantic Role Labeling via Tree Kernel Joint Inference. In Proceedings of CoNLL-X. find
  22. Sameer Pradhan, Kadri Hacioglu, Wayne Ward, James H. Martin, and Daniel Jurafsky. 2003. Semantic Role Parsing: Adding Semantic Structure to Unstructured Text. In Proceedings ICDM’03, Melbourne, USA. find
  23. Sameer Pradhan, Kadri Hacioglu, Valerie Krugler, Wayne Ward, James H. Martin, and Daniel Jurafsky. 2005. Support Vector Learning for Semantic Argument Classification. Machine Learning. ↗
  24. 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 ACL’08, Short Papers, Columbus, Ohio, June.
  25. John Shawe-Taylor and Nello Cristianini. 2004. Kernel Methods for Pattern Analysis. Cambridge University Press. ↗
  26. Honglin Sun and Daniel Jurafsky. 2004. Shallow Semantic Parsing of Chinese. In Proceedings of NAACL-HLT. find
  27. Cynthia A. Thompson, Roger Levy, and Christopher Manning. 2003. A Generative Model for Semantic Role Labeling. In ECML’03. find
  28. Vladimir N. Vapnik. 1998. Statistical Learning Theory. John Wiley and Sons. find
  29. Nianwen Xue and Martha Palmer. 2004. Calibrating Features for Semantic Role Labeling. In Dekang Lin and Dekai Wu, editors, Proceedings of EMNLP 2004, Barcelona, Spain. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/diab-2008-semantic-role-labeling-systems-arabic/