Elamur Research⌕
☰
home / papers / 2014 / Induction of a variable granularity property grammar from the Arabic Treebank ATB (Induction d'une grammaire de propriétés à granularité variable à partir du treebank arabe ATB) [in French]

Induction of a variable granularity property grammar from the Arabic Treebank ATB (Induction d'une grammaire de propriétés à granularité variable à partir du treebank arabe ATB) [in French]

Paper · 2014 · Raja Bensalem Bahloul and Marwa Elkarwi

Summary

The paper addresses the lack of a constraint-based grammatical resource for Arabic. Building one by hand would require much time and many linguists, so the authors induce a Property Grammar (PG, in Blache's formalism) automatically from the Penn Arabic Treebank (PATB/ATB). They chose the ATB because it is phrase-structure based, linguist-validated and aimed at Modern Standard Arabic.

The method has two steps:

  1. Extract a context-free grammar (CFG) from the treebank's productions.
  2. Generate properties from the CFG rules. These are independent relations among the categories that make up each syntactic unit.

The fine-grained ATB tagset could make the grammar very large. To control this, the authors vary the granularity of categories and compare the resulting grammars. Coarser categories sharply reduce grammar size and complexity, but they lose precision. This argues for a tunable granularity mechanism.

The pipeline is fully automatic for the property types defined. It does not depend on the language or on the source treebank formalism, because it works only from the CFG. The authors present it as the first treebank-induced PG for Arabic.

Key points

  • The source resource is the ATB/PATB (Maamouri & Bies, 2004). It already includes clitic segmentation and vocalization handled with the SAMA analyzer (Kulick et al., 2010).
  • The pipeline runs treebank → CFG → PG. Properties are computed only from CFG rules, which is why the approach transfers across languages and formalisms.
  • Grammar size is controlled by moving between granularity levels of ATB categories, for example NOUN versus NOUN_NUM or NOUN_PROP.
  • No obligation properties appear at the finest granularity. The reason is that no single category form, such as NOUN, appears in every rule of a phrase type.
  • Exclusion properties, by contrast, are extremely numerous; examples include SBAR ⊗ {PP, NP, PRT, …} and PP ⊗ {ADVP, PRT, CONJP, PRON, FRAG}. Many may reflect rare constructions or gaps in the treebank rather than real restrictions.

Interesting ideas

  • I1. Instead of keeping the treebank productions as rewrite rules, the authors break them into independent, local constraints on categories, which can be checked one at a time and explained one at a time.
  • I2. Tag granularity works as an explicit dial between grammar compactness and descriptive precision, rather than as a fixed choice.
  • I3. A constraint such as obligation can vanish only because the head category is split into fine subtypes. Merging subtypes recovers the generalization.
  • I4. The authors warn that a co-occurrence never seen in a treebank is not proof that it is forbidden; sparsity or rarity may explain it.
  • I5. The authors propose feeding the induced grammar back to enrich the original treebank with a property-based layer, which could also expose annotation problems.

For Mova

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

  • I1 (tree): Induce per-head constraint inventories from UD treebanks, such as which dependents co-occur and which never do. Each parser output could then be validated with human-readable reasons.
  • I2 (tokens): Expose a coarse-to-fine tag hierarchy (UPOS → XPOS → features) so downstream checks can pick the level that keeps rules general enough without losing distinctions.
  • I3 (tree): Before mining required-dependent rules, normalize fragmented categories (e.g., proper and common nouns) so real regularities are not hidden by tagset detail.
  • I4 (learn): Treat unseen pairs as weak, count-weighted evidence. Test any induced exclusion rule against negative controls before it becomes a self-induced repair rule.
  • I5 (learn): Run induced constraints over silver parses agreed by witness parsers and flag violations for the LLM teacher tier to review.

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 6The 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 worksPsychology · 1,649 worksExperimental and Cognitive Psychology · 930 works

🧭 Topics 3Research subjects and languages the work is about.Semitic languages · 3,066 worksTreebanks · 127 worksFrench · 83 works

Cite

Cite as Raja Bensalem Bahloul and Marwa Elkarwi. 2014. Induction of a variable granularity property grammar from the Arabic Treebank ATB (Induction d'une grammaire de propriétés à granularité variable à partir du treebank arabe ATB) [in French]. Proceedings of TALN 2014 (Volume 4: RECITAL - Student Research Workshop).

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

@inproceedings{bahloul2014induction,
  title = {{Induction of a variable granularity property grammar from the Arabic Treebank ATB (Induction d'une grammaire de propriétés à granularité variable à partir du treebank arabe ATB) [in French]}},
  author = {Raja Bensalem Bahloul and Marwa Elkarwi},
  booktitle = {Proceedings of TALN 2014 (Volume 4: RECITAL - Student Research Workshop)},
  year = {2014},
  url = {https://aclanthology.org/F14-4012/}
}

References

17 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. ABEILLÉ A., CLÉMENT L., KINYON A. (2000). Building a treebank for French. Proceedings of the Second International Language Resources and Evaluation Conference. Athens, Greece. find
  2. AURAN C., BOUZON C., HIRST D.J. (2004). The Aix-MARSEC project: an evolutive database of spoken British English. Proceedings of the Second International Conference on Speech Prosody, 561-564. Nara. find
  3. BLACHE P. (2001). Les Grammaires de Propriétés : Des contraintes pour le traitement automatique des langues naturelles. Hermès Sciences Publications. find
  4. BLACHE P. (2014). A Chinese Constraint Grammar Extracted from the Chinese Treebank. Proceedings of APCLC. find
  5. BLACHE P., GUÉNOT M.-L. & VANRULLEN T. (2003). Corpus-based grammar development. Proceedings of Corpus Linguistics-03. find
  6. BLACHE P., RAUZY S. (2012). Hybridization and Treebank Enrichment with Constraint-Based Representations. Proceedings of LREC-2012. find
  7. DUCHIER D., PROST J.-P., DAO T.-B-.H. (2009). A Model-Theoretic Framework for Grammaticality Judgements. FG, volume 5591 of Lecture Notes in Computer Science, page 17-30. Springer. find
  8. KAY P., FILLMORE C. (1999).Grammatical Constructions and Linguistic Generalizations: the what’s x doing y construction. Language. find
  9. LAMMIE GLEEN M., STRASSEL S. (2005). Linguistic Resources for Meeting Speech Recognition. MLMI-05. find
  10. MAAMOURI M., BIES A. (2004). Developing an Arabic Treebank: Methods, Guidelines, Procedures, and Tools. Proceedings of COLING-04. Geneva, Switzerland.
  11. MAAMOURI M., BIES A., KROUNA S., GADDECHE F., BOUZIRI B. (2009). Penn Arabic Treebank guidelines v4.8. Technical report, Linguistic Data Consortium, University of Pennsylvania. find
  12. MAAMOURI M., ZAGHOUANI W., VIOLETTA CAVALLI-SFORZA V., GRAFF D., CIUL M. (2012). Developing ARET: An NLP-based Educational Tool Set for Arabic Reading Enhancement. NAACL-HLT 2012. Montreal.
  13. MOHRI M., ROARK B. (2006). Probabilistic Context-Free Grammar Induction Based on Structural Zeros. Proceedings of the Seventh Meeting of the Human Language Technology conference- North American Chapter of the Association for Computational Linguistics (HLT-NAACL 2006). New York. REHBEIN I., VAN GENABITH. find
  14. J. (2007) Treebank annotation schemes and parser evaluation for German. Proceedings of the Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 630–639. Prague, Czech Re public. find
  15. SONG Z., ISMAEL S., GRIMES S., DOERMANN D., STRASSEL S. (2012). Linguistic Resources for Handwriting Recognition and Translation Evaluation. LREC 2012. Istanbul. find
  16. TOUNSI L., VAN GENABITH J. (2010). Arabic Parsing Using Grammar Transforms. LREC 2010.
  17. VANRULLEN T., BLACHE P., PORTES C., RAUZY S., MAEYHIEUX J.-F., GUENOT M.-L., BALFOURIER J.-M., BELLENGIER E. (2005). Une plateforme pour l’acquisition, la maintenance et la validation de ressources lexicales. Actes de TALN, pp. 41-48. Paris, France. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bahloul-2014-induction-variable-granularity-property-grammar/