Elamur Research⌕
☰
home / papers / 2012 / La reconnaissance automatique de la fonction des pronoms démonstratifs en langue arabe (Automatic recognition of demonstrative pronouns function in Arabic) [in French]

La reconnaissance automatique de la fonction des pronoms démonstratifs en langue arabe (Automatic recognition of demonstrative pronouns function in Arabic) [in French]

Paper · 2012 · Yacine Ben Yahia, Souha Mezghani Hammami and Lamia Hadrich Belguith

Summary

The paper tackles a preprocessing step for Arabic anaphora resolution: deciding whether a demonstrative pronoun is anaphoric (its antecedent comes earlier) or cataphoric (it points to something later, typically the noun that follows it). Knowing the direction tells a resolver which side of the pronoun to search. The authors say no such classifier existed for Arabic. Their pipeline has four stages: sentence segmentation, morphological analysis, extraction of eight contextual features, and supervised learning of classification rules with standard learners (IBk, JRip, NBTree, NaiveBayes). Most features describe the word right after the pronoun: its part of speech, particle type, definiteness, and clitics. Two lexical cues and the pronoun's identity complete the set. Evaluation used cross-validation on a corpus of 365,585 words containing 14,318 demonstratives, compared against a hand-written rule baseline. IBk scored best, with an F-measure of 94.7% using automatic MORPH2 morphology and 99.3% using manually corrected morphology. Learning clearly beats the rules when tagging is noisy. Upstream morphological quality turns out to be the main bottleneck.

Key points

  • Corpus: 365,585 words with 14,318 demonstrative pronouns. The task is binary: anaphoric vs. cataphoric.
  • Eight features are used:
    • POS+1 (category of the next word)
    • Type (kind of particle)
    • Determine (whether the next noun carries the definite article)
    • Enclitique and Proclitique+1 (clitics on the next word)
    • Bimafidhalika (the fixed phrase bmA fy before *lk)
    • MotSpec (a short list of trigger words such as OyDA, gyr, mmA, kl)
    • the pronoun itself.
  • With automatic MORPH2 tags, F-measures were: IBk 94.7%, NBTree 94.7%, JRip 93.9%, NaiveBayes 92.8%, and the rule baseline 76.72%.
  • With manually corrected tags, IBk reached 99.3% and the baseline 98.38%. The rule baseline gained about 21.66 points from better tagging, the learners about 5.
  • The main errors come from missing punctuation. In one example, a pronoun followed by a definite noun is labelled cataphoric because no comma separates them, although it is really anaphoric.

Interesting ideas

  • I1. Before running coreference, classify the direction of reference, so the resolver searches only before or only after the pronoun and avoids looking for antecedents that do not exist.
  • I2. A few local features, mostly about the very next token, nearly decide the label, which keeps the classifier small and each decision explainable.
  • I3. Running the same system on automatic tags and on gold tags separates errors caused by the upstream tagger from errors of the classifier itself.
  • I4. Hand-written rules can nearly match learned models on clean input yet collapse on noisy input, while learners degrade much more gracefully.
  • I5. Missing punctuation can make a pronoun look like a determiner of the following noun, which suggests checking boundary cues explicitly.

For Mova

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

  • I1 (frames): The coreference step could first tag each this/that/these as backward-pointing, forward-pointing, or determiner-like, then restrict the candidate search to the matching side.
  • I2 (frames): A demonstrative's role could be read off its UD context (det vs. nsubj/obj, and what follows it), giving a traceable reason for every coreference choice.
  • I3 (learn): Evaluations could routinely report scores for downstream modules on both predicted and gold trees, showing which stage deserves repair effort.
  • I4 (learn): Self-induced repair rules should be tested on predicted-tag input as well as gold, so their apparent accuracy is not inflated by clean annotations.
  • I5 (tree): The parser could flag ambiguous det/pronoun attachments of demonstratives where no comma is present, and prefer the pronoun reading when the verb's valency calls for an object.

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 3Research subjects and languages the work is about.Semitic languages · 3,066 worksCoreference resolution · 575 worksFrench · 83 works

Cite

Cite as Yacine Ben Yahia, Souha Mezghani Hammami and Lamia Hadrich Belguith. 2012. La reconnaissance automatique de la fonction des pronoms démonstratifs en langue arabe (Automatic recognition of demonstrative pronouns function in Arabic) [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{yahia2012la,
  title = {{La reconnaissance automatique de la fonction des pronoms démonstratifs en langue arabe (Automatic recognition of demonstrative pronouns function in Arabic) [in French]}},
  author = {Yacine Ben Yahia, Souha Mezghani Hammami and Lamia Hadrich Belguith},
  booktitle = {Proceedings of the Joint Conference JEP-TALN-RECITAL 2012, volume 2: TALN},
  year = {2012},
  url = {https://aclanthology.org/F12-2040/}
}

References

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

  1. C. PAICE ET G. HUSK (1987). Towards the automatic recognition of anaphoric features in English text: the impersonal pronoun it. Computer Speech and Language. ↗
  2. D. W EISSENBACHER ET A. N AZARENKO (2007). A bayesian classifier for the recognition of the impersonal occurrences of the it pronoun. In Proceedings of DAARC’07, 2007. find
  3. S. L APPIN ET H. L EASS (1994). An algorithm for pronominal anaphora resolution, Computational Linguistics, 20(4), 1994, p. 535–561.
  4. S. M EZGHANI H AMMAMI ., R.S ELLAMI , L. H ADRICH B ELGUITH (2010). A Bayesian Classifier for the Identification of Non-referential Pronouns in Arabic. The 7th INFOS 2010. find
  5. S. H AMMAMI , L. B ELGUITH , A. B EN H AMADOU (2009). A Rule-Based Method for Detecting Arabic Anaphoric Pronouns. Proceedings of the 7th DAARC’2009, Goa-India, 2009. find
  6. R. E VANS (2001). Applying machine learning toward an automatic classi cation of it. Literary and linguistic computing, 16, 2001, p. 45–57. find
  7. M. D ENBER (1998). Automatic resolution of anaphora in English. find
  8. A BDUL -M AGEED , M. 2011. Automatic detection of Arabic non-anaphoric pronouns for improving anaphora resolution. Asian Lang. Inform. Process. (March 2011). find
  9. S. B ERGSMA , D. L IN ET R. G OEBEL (2008). Distributional Identification of Non-Referential Pronouns. find
  10. ACL, Columbus Ohio, 2008, p. 10-18. find
  11. A. B OYD , W. G EGG -H ARRISON ET D. B YRON (2005). Identifying non-referential it: a machine learning approach incorporating linguistically motivated features. Workshop on Feature Engineering for Machine Learning in Natural Language Processing, 2005. find
  12. C. M ÜLLER (2007).Resolving It, This, and That in Unrestricted Multi-Party Dialog. In Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics, 2007. find
  13. D. B YRON (2002). Resolving Pronominal Reference to Abstrac Entities. In Proceedings of th 40th Annual Meeting of the Association of Computation Linguistics (ACL), Philadelphia 2002. find
  14. C. N AVARETTA (2009). Automatic Recognition of the Function of Singular Neuter Pronouns in Texts and Spoken Data. In DAARC 2009. find
  15. K. D UTTA , N. P RAKASH , S. K AUSHIK (2010). Probabilistic neural network approach to the classification of demonstrative pronouns for indirect anaphora in Hindi. find
  16. International Journal Information Technology and Intelligent Computing, 2010. find
  17. N. C HAÂBEN K AMMOUN , L. H ADRICH B ELGUITH ET A. B EN H AMADOU (2010). The MORPH2 new version: A robust morphological analyzer for Arabic texts. JADT’2010. find
  18. E. F RANK , I AN H. W ITTEN (2005). Practical Machine Learning Tools and Techniques, Second Edition. (Morgan Kaufmann series in data management systems). find
  19. B ELGUITH , L., B ACCOUR , L. AND M OURAD , G. (2005). Segmentation de textes arabes basée sur l'analyse contextuelle des signes de ponctuations et de certaines particules (TALN’2005). find
  20. M ITKOV (2002): R. Mitkov, «Anaphora resolution». find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yahia-2012-la-reconnaissance-automatique-de-la/