Elamur Research⌕
☰
home / papers / 2012 / Automatic Extraction and Evaluation of Arabic LFG Resources

Automatic Extraction and Evaluation of Arabic LFG Resources

Paper · 2012 · Mohammed Attia, Khaled Shaalan, Lamia Tounsi et al.

Summary

The paper builds deep-grammar lexical resources for Arabic automatically instead of writing them by hand. The starting point is an existing algorithm (Tounsi et al., 2009) that adds LFG f-structure equations to the Penn Arabic Treebank (ATB). From the annotated trees the authors derive two resources.

  • ArabicSubcats is a probabilistic lexicon of subcategorization frames. Each frame records preposition values for obliques and whether the frame is active or passive.
  • LDD paths are sequences of grammatical functions that connect a displaced discourse function (TOPIC, FOCUS, etc.) to its gap or resumptive pronoun.

Surface forms are lemmatized by re-running Buckwalter morphology on each word. The analysis kept is the one whose diacritization and POS agree with the ATB gold tags. Frame probabilities are relative frequencies conditioned on the lemma.

ArabicSubcats was evaluated against a hand-built lexicon from a rule-based Arabic LFG parser. Recall is high for verbs but weaker for nouns and adjectives. Because no attested LDD path is longer than 5 functions, the authors suggest bounding functional uncertainty equations instead of using the Kleene star. The resources are intended to add deep dependencies to the output of shallow PCFG parsers.

Key points

  • Source data: the ATB has 22,524 sentences and 787,235 tokens.
  • ArabicSubcats size: 240 frame types over 3,295 lemmas, giving 7,746 lemma-frame pairs (about 2.35 frames per lemma).
  • Hand-made baseline: the Attia (2008) LFG parser lexicon has 64 frame types, 2,709 lemmas and 2,901 pairs (about 1.07 frames per lemma).
  • Graded recall: the comparison was run at four levels, from the full argument list down to transitivity only. Verbs rise from 0.78 to 0.96. Nouns and adjectives are lower, reaching at most 0.77 and 0.86.
  • Noun error analysis: in about 80% of mismatched nouns, the ATB tagged a verbal noun as a common noun.
  • Path length: the longest attested LDD path contains 5 GFs. Resumptive pronouns can mark the lower end of an LDD.

Interesting ideas

  • I1. Build the valency lexicon directly from treebank annotation and attach a relative-frequency probability to each lemma–frame pair, so that dominant frames stand out from rare ones
  • I2. Evaluate in stages by removing detail step by step (preposition values, then clausal complements, then obliques), which shows whether disagreement lies in fine distinctions or in core transitivity
  • I3. Replace an unbounded path pattern with an empirical upper bound taken from the longest path attested in the corpus
  • I4. Choose the lemma by re-analysing each word morphologically and keeping the analysis that agrees with independent gold evidence (diacritics plus POS)
  • I5. Manually inspect a small sample of mismatches against a hand lexicon to find out whether the error comes from the treebank, the lexicon, or the extraction

For Mova

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

  • I1 (world): Verb-class knowledge could carry corpus-derived frame probabilities from UD treebanks alongside VerbNet, which would help rank competing frame readings.
  • I2 (learn): Frame and role evaluation could report the same graded scores, which would separate core argument errors from errors in preposition or complement detail.
  • I3 (tree): Recovering extraction and relative-clause dependencies over UD trees could limit search depth to the lengths seen in treebanks, keeping inference cheap and explainable.
  • I4 (tokens): When the English lemmatizer returns several analyses, the one consistent with the tagger's tag could be selected, and any disagreement logged as an explicit decision.
  • I5 (learn): Disagreements between silver data and a reference could be audited in small sampled batches, and the findings could drive self-induced repair rules.

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 · Ireland50 works · 8,937 citations

Also on the byline: British University in Dubai.

📚 Shelves 4 Computer Science · 9,086 works Artificial Intelligence · 8,889 works Arts and Humanities · 9,296 works Language and Linguistics · 9,206 works

🧭 Topics 3 Evaluation & benchmarks · 685 works Semitic languages · 1,484 works Syntax & grammar formalisms · 161 works

Cite

Cite as Mohammed Attia, Khaled Shaalan, Lamia Tounsi and Josef van Genabith. 2012. Automatic Extraction and Evaluation of Arabic LFG Resources. Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12).

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

@inproceedings{attia2012automatic,
  title = {{Automatic Extraction and Evaluation of Arabic LFG Resources}},
  author = {Mohammed Attia, Khaled Shaalan, Lamia Tounsi and Josef {van Genabith}},
  booktitle = {Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12)},
  year = {2012},
  doi = {10.63317/3pmha53egzrc},
  url = {https://doi.org/10.63317/3pmha53egzrc}
}

References

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

  1. Attia, Mohammed. (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
  2. Austin, P. K. (2001). Lexical functional grammar. In International Encyclopedia of the Social and Behavioral Sciences, eds. N. J. Smelser and P. Baltes, 8748-8754. Oxford, UK: Elsevier Science Ltd. ↗
  3. Bikel, Dan. (2004). A distributional analysis of a lexicalized statistical parsing model. In Proceedings of the 9th Conference on Empirical Methods in Natural Language Processing (EMNLP), Barcelona, Spain. find
  4. Buckwalter, Tim. 2004. Buckwalter Arabic Morphological Analyzer. Linguistic Data Consortium. (LDC2004L02). find
  5. Cahill, A., M. Burke, R. O'Donovan, J. van Genabith, and A. Way. (2004). Long-distance dependency resolution in automatically acquired wide-coverage PCFG-based LFG approximations. In: ACL 2004 - 42nd Annual Meeting of the Association for Computational Linguistics, 21-26 July 2004, Barcelona, Spain. find
  6. Cahill, A., M. Burke, R. O'Donovan, S. Riezler, J. van Genabith and A. Way (2008). Wide-Coverage Deep Statistical Parsing using Automatic Dependency Structure Annotation . In Computational Linguistics, Vol. 34, No. 1, pp. 81-124 find
  7. Dalrymple, Mary. (2001). Lexical Functional Grammar. New York: Academic Press. ↗
  8. Falk, Yehuda N. (2002). Resumptive Pronouns in LFG. In The LFG 02 Conference, Athens, pp. 154-173. find
  9. Guo, Yuqing and van Genabith, Josef and Wang, Haifeng. (2007). Treebank-based acquisition of LFG resources for Chinese. In: Lexical Functional Grammar 2007, 28-30, California, USA. find
  10. Kaplan, Ronald M., and Annie Zaenen. (1989). Long-distance Dependencies, Constituent Structure and Functional Uncertainty. In Alternative Conceptions of Phrase Structure, ed. by M. R. Baltin and A. S. Kroch. Chicago: University of Chicago Press. find
  11. Maamouri, M. and Bies, A. (2004). Developing an Arabic Treebank: Methods, guidelines, procedures, and tools. In Workshop on Computational Approaches to Arabic Script-based Languages, COLING. our card
  12. O'Donovan, Ruth, Michael Burke, Aoife Cahill, Josef van Genabith and Andy Way (2005) Large-Scale Induction and Evaluation of Lexical Resources from the Penn-II and Penn-III Treebanks, Computational Linguistics, pages 329 – 366. find
  13. Petrov, Slav, Leon Barrett, Romain Thibaux, and Dan Klein. (2006). Learning Accurate, Compact, and Interpretable Tree Annotation. In Proceedings of COLING-ACL. our card
  14. Rehbein, Ines and Josef van Genabith. (2009). Automatic Acquisition of LFG Resources For German - As Good As It Gets. Proceedings of the LFG09 Conference. find
  15. Cambridge, UK Schluter, Natalie and Josef van Genabith. (2008). Treebank-Based Acquisition of LFG Parsing Resources for French. In: the Sixth International Language Resources and Evaluation Conference (LREC'08), Marrakech, Morocco Tounsi, Lamia, Mohammed Attia and Josef van Genabith. (2009). Automatic Treebank-Based Acquisition of Arabic LFG Dependency Structures. EACL-Workshop on Computational Approaches to Semitic Languages, Athens, Greece. find
  16. Vadas, D. and J. R. Curran. (2007). Adding Noun Phrase Structure to the Penn Treebank. The 45th Annual Meeting of the Association of Computational Linguistics, pp. 240–247, Prague, Czech Republic. our card

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/attia-2012-automatic-extraction-evaluation-arabic-lfg/