Elamur Research⌕
☰
home / papers / 2009 / Automatic Treebank-Based Acquisition of Arabic LFG Dependency Structures

Automatic Treebank-Based Acquisition of Arabic LFG Dependency Structures

Paper · 2009 · Lamia Tounsi, Mohammed Attia and Josef van Genabith

Summary

The paper ports an established treebank-annotation approach from English to Arabic. That approach automatically enriches phrase-structure trees with LFG f-structure equations, which encode deep predicate-argument dependencies. Arabic is hard for this method for three reasons: word order is flexible (VSO, SVO, VOS), morphology is rich, and grammatical function cannot be read off tree position as it largely can in English. The authors therefore redesigned the annotation algorithm for the Penn Arabic Treebank (ATB) so that it relies mainly on ATB functional tags and fine-grained morphological tags, with positional left-right matrices playing only a secondary role. The pipeline has several modules:

  • tag conflation
  • head-finding
  • lexical macros
  • functional-tag mapping
  • context matrices
  • coordination handling
  • catch-all clean-up
  • trace resolution for long-distance dependencies

About 98% of usable ATB trees yield a complete, connected f-structure. Against a 250-sentence hand-built gold standard, the output reaches roughly 95% f-score, comparable to the English, French and Chinese versions. The annotated treebank is meant to be the basis for wide-coverage deep parsing and generation resources for Arabic.

Key points

  • Data: Penn Arabic Treebank (ATB) with 23,611 sentences. After trees containing FRAG and X constituents were removed, 19,273 sentences were annotated. Of these, 98% produced a full connected f-structure, and the remaining 2% failed because of feature-value clashes.
  • The fine-grained preterminals (over 460 types) give 45,785 phrase-structure rules. Conflating tags to basic POS reduces this to 9731 rules. Keeping, for each category, the most frequent rules up to 85% coverage leaves 339 rules, which were inspected by hand to build the annotation matrices.
  • ATB functional tags map directly to LFG functions: -SBJ becomes SUBJ, -OBJ becomes OBJ, -DTV/-BNF become OBJ2, plus TOPIC, OBL and ADJUNCT. Morphological features such as gender, number, case and person become f-structure attributes through lexical macros.
  • Verbless copula clauses (tag -PRD) are analysed with a "null be" predicate that takes an XCOMP, and the subject of the XCOMP is controlled by the main subject.
  • Quality was measured against the DCU 250 gold standard (Al-Raheb et al., 2006) using triple-based precision/recall: precision 95.49, recall 94.43, f-score 95%.

Interesting ideas

  • I1. Collapse fine-grained tags to coarse POS, then hand-inspect only the frequent head of the Zipfian rule distribution (85% coverage per category); this keeps manual rule-writing small while still covering most real data.
  • I2. Assign functions first from explicit signals (functional tags, morphology), use positional context only where those are silent, and finish with a catch-all module that fixes overgeneralisations and fills defaults; this layering keeps each component simple and debuggable.
  • I3. Model a missing copula as an explicit empty predicate with an open complement and a control equation, rather than leaving the predicative relation implicit.
  • I4. Build a partial proto-structure first, and resolve long-distance dependencies afterwards as reentrancies (shared values) using trace indices, so that local and non-local analysis are cleanly separated.
  • I5. Score deep output as relation(arg1, arg2) triples that include morphological features as well as dependencies, and treat unification clashes as automatic failure signals.

For Mova

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

  • I1 (tree): Converters between UD dialects or treebanks could rank their construction patterns this way and spend human effort only on the few hundred patterns that dominate.
  • I2 (frames): Role assignment in prag could follow the same ordering (explicit UD labels and morphology, then position, then a logged clean-up pass), so each decision stays traceable to the rule that produced it.
  • I3 (frames): For "X is Y" and small clauses ("consider Kris a fool"), prag could emit an explicit "be" frame whose subject is shared with the predicate, giving the world model a uniform state assertion.
  • I4 (frames): Relative clauses and wh-questions could be handled in a dedicated final pass that links the relative pronoun's antecedent as the verb's argument, instead of being mixed into local role labelling.
  • I5 (learn): Silver-data agreement among witness parsers could be measured on such triples, and inputs whose constraints clash could be routed to repair rules instead of into training.

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

🎓Dublin City UniversityDublin · Ireland70 works · 9,447 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 4Research subjects and languages the work is about.Semitic languages · 2,908 worksMorphology · 405 worksSyntax & grammar formalisms · 214 worksTreebanks · 126 works

Cite

Cite as Lamia Tounsi, Mohammed Attia and Josef van Genabith. 2009. Automatic Treebank-Based Acquisition of Arabic LFG Dependency Structures. Proceedings of the EACL 2009 Workshop on Computational Approaches to Semitic Languages.

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

@inproceedings{tounsi2009automatic,
  title = {{Automatic Treebank-Based Acquisition of Arabic LFG Dependency Structures}},
  author = {Lamia Tounsi, Mohammed Attia and Josef {van Genabith}},
  booktitle = {Proceedings of the EACL 2009 Workshop on Computational Approaches to Semitic Languages},
  year = {2009},
  doi = {10.3115/1621774.1621783},
  url = {https://doi.org/10.3115/1621774.1621783}
}

References

28 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Y. Al-Raheb, A. Akrout, J. van Genabith, J. Dichy. 2006. DCU 250 Arabic Dependency Bank: An LFG Gold Standard Resource for the Arabic Penn Treebank The Challenge of Arabic for NLP/MT at the British Computer Society, UK, pp. 105–116. find
  2. M. Attia. 2008. Handling Arabic Morphological and Syntactic Ambiguity within the LFG Framework with a View to Machine Translation. Ph.D. Thesis. The University of Manchester, Manchester, UK. find
  3. M. Baptista. 1995. On the Nature of Pro-drop in Capeverdean Creole. Harvard Working Papers in Linguistics, 5:3-17. find
  4. A. Bies and M. Maamouri. 2003. Penn Arabic Treebank Guidelines URL: http://www.ircs.upenn.edu/arabic/Jan03release/guidelines-TB-1-28-03.pdf. ↗
  5. J. Bresnan. 2001. Lexical-Functional Syntax. Blackwell Publishers, Oxford, UK. find
  6. M. Burke, O. Lam, R. Chan, A. Cahill, R. ODonovan, A. Bodomo, J. van Genabith, and A. Way. 2004. Treebank-Based Acquisition of a Chinese Lexical-Functional Grammar. The 18th Pacific Asia Conference on Language, Information and Computation, Tokyo, Japan, pp. 161–172. find
  7. M. Burke. 2006. Automatic Treebank Annotation for the Acquisition of LFG Resources. Ph.D. thesis, School of Computing, Dublin City University, Ireland. find
  8. A. Cahill, M. McCarthy, J. van Genabith, A. Way. 2002. Automatic Annotation of the Penn Treebank with LFG F-Structure Information. LREC Workshop on Linguistic Knowledge Acquisition and Representation: Bootstrapping Annotated Language Data, Spain, pp. 8-15. find
  9. A. Cahill, M. Forst, M. McCarthy, R. ODonovan, C. Rohrer, J. van Genabith and A. Way. 2003. Treebank-Based Multilingual Unification Grammar Development. The 15th Workshop on Ideas and Strategies for Multilingual Grammar Development, at the 15th European Summer School in Logic, Language and Information, Vienna, Austria, pp. 17–24. find
  10. A. Cahill, M. Burke, R. ODonovan, J. van Genabith, A. Way. 2004. Long-Distance Dependency Resolution in Automatically Acquired Wide-Coverage PCFG-Based LFG Approximations. 42nd Meeting of the Association for Computational Linguistics, Barcelona, Spain pp. 319-326. find
  11. A. Cahill, J. van Genabith. 2006. Robust PCFG-Based Generation using Automatically Acquired LFG Approximations. ACL 2006, Sydney, Australia, pages 1033-1040. find
  12. A. Cahill, M. Burke, R. O’Donovan, S. Riezler, J. van Genabith, A. Way. 2008. Wide-Coverage Deep Statistical Parsing using Automatic Dependency Structure Annotation. Computational Linguistics, Vol. 34, No. 1, pages 81-124. ↗
  13. N. Chomsky. 1981. Lectures on Government and Binding. Dordrecht: Foris. find
  14. G. Chrupala and J. van Genabith. 2006. Improving Treebank-Based Automatic LFG Induction for Spanish. In Proceedings of the LFG06 Conference. find
  15. R. Crouch, R. M. Kaplan, T. H. King, S. Riezler. 2002. Comparison of Evaluation Metrics for a Broad Coverage Parser LREC Workshop: Beyond PARSE-VAL Towards Improved Evaluation Measures for Parsing Systems, Spain, pp. 67-74. find
  16. M. Dalrymple. 2002. Lexical Functional Grammar. Syntax and Semantics, Volume 34, Academic Press, San Diego, CA/London, U.K. ↗
  17. Y. Falk. 2001. Lexical-Functional Grammar: An Introduction to Parallel Constraint-Based Syntax. Stanford, Calif.: CSLI Publications. find
  18. A. Frank, L. Sadler, J. van Genabith, A. Way. 2003. From Treebank Resources to LFG F-Structurese. A. Abeille editor Treebanks: Building and Using Syntactically Annotated Corpora, Kluwer Academic Publishers, Dordrecht/Boston/London,The Nether-landsm pp. 367-389. find
  19. Y. Guo, J. van Genabith, H. Wang. 2007. Acquisition of Wide-Coverage, Robust, Probabilistic Lexical-Functional Grammar Resources for Chinese. Proceedings of the 12th International Lexical Functional Grammar Conference, USA, pp. 214–232. find
  20. R. Kaplan and J. Bresnan. 1982. Lexical Functional Grammar: a Formal System for Grammatical Representation, in J. Bresnan (ed.). The Mental Representation of Grammatical Relations, MIT Press, Cambridge, MA, pp. 173-281. find
  21. R. M. Kaplan, S. Riezler, T. H. King, J. T. Maxwell, A. Vasserman, and R. Crouch. 2004. Speed and Accuracy in Shallow and Deep Stochastic Parsing. In The Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics (HLT-NAACL 2004), Boston, MA, pp. 97–104. find
  22. M. Maamouri and A. Bies. 2004. Developing an Arabic Treebank: Methods, Guidelines, Procedures, and Tools Workshop on Computational Approaches to Arabic Script-based Languages, COLING 2004, Geneva, 2004.
  23. M. Marcus, G. Kim, M. Marcinkiewicz, R. MacIntyre, A. Bies, M. Ferguson, K. Katz and B. Schasberger 1994. The Penn Treebank: Annotating Predicate Argument Structure. In Proceedings of the Human Language Technology Workshop. San Francisco, CA.
  24. R. O’Donovan, M. Burke, A. Cahill, J. van Genabith, and A. Way. 2004. Large-Scale Induction and Evaluation of Lexical Resources from the Penn-II Treebank. The 42nd Annual Meeting of the Association for Computational Linguistics, Barcelona, Spain, pp. 368–375. find
  25. R. O’Donovan, A. Cahill, J. van Genabith, and A. Way. 2005. Automatic Acquisition of Spanish LFG Resources from the CAST3LB Treebank. The Tenth International Conference on LFG, Bergen, Norway, pp. 334–352. find
  26. S. Riezler, King, T., Kaplan, R., Crouch, R., Maxwell, J. T., and Johnson, M. 2002. Parsing theWall Street Journal using a Lexical-Functional Grammar and Discriminative Estimation Techniques. The 40th Annual Conference of the Association for Computational Linguistics (ACL-02), Philadelphia, PA, pp. 271–278. find
  27. P. Sells 1985. Lectures on Contemporary Syntactic Theories. CSLI Lecture Notes. Stanford, CA: CSLI. find
  28. N. Schluter and J. van Genabith 2008. Treebank-Based Acquisition of LFG Parsing Resources for French. Proceedings of the Sixth International Language Resources and Evaluation (LREC’08). find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tounsi-2009-automatic-treebank-based-acquisition-arabic/