Elamur Research⌕
☰
home / papers / 2008 / Arabic WordNet: Semi-automatic Extensions using Bayesian Inference

Arabic WordNet: Semi-automatic Extensions using Bayesian Inference

Paper · 2008 · Horacio Rodríguez, David Farwell, Javi Ferreres et al.

Summary

The paper addresses how to grow Arabic WordNet (AWN) without making lexicographers judge every candidate entry by hand. AWN was built manually on the EuroWordNet methodology and maps directly onto Princeton WordNet 2.0. The authors exploit Arabic root morphology: a small set of productive lexical rules derives related verbal, nominal and adjectival forms from existing verb entries. A C5.0 decision tree filters the overgenerated forms. The surviving words are translated to English with the NMSU lexicon and linked to PWN synsets, and then to synsets related to those. This produces a four-layer graph per base verb.

Earlier work scored the candidate word–synset pairs with graph heuristics. Here, the graph is turned into a Bayesian Network:

  • edge directions are added greedily to avoid cycles;
  • translation priors come from GIZA++ models trained on the UN corpus;
  • noisy-or conditional probability tables are used.

Posterior synset probabilities rank the candidates. The BN doubles the number of candidates found by the heuristics. Intersecting BN output with the heuristic output gives the best F1 (0.62) and, at strict thresholds, precision of 0.71. Both results help to cut manual effort in building wordnets.

Key points

  • AWN at the time had 11,270 synsets and 23,496 Arabic expressions, including 1,142 named-entity synsets that were extracted automatically and were still under review.
  • Every one of the 2,296 base verb forms gets its own BN, with four layers: Arabic words, English words, synsets S1 and related synsets S2. Each node may have at most 10 parents.
  • Resources used: LOGOS conjugated-verb database, NMSU Arabic-English lexicon, Arabic Gigaword, the UN (2000–2002) Arabic-English corpus and GIZA++ word alignment.
  • Results on the 11 evaluation verbs:
MethodCandidatesAcceptedPrecisionF1
Heuristics (all)2721350.500.55
BN, threshold 05542230.400.57
BN ∩ heuristics, threshold 02721540.560.62
  • Error analysis showed many rejections came from a missing shadda diacritic or a missing feminine form. Fixing these would raise accepted BN candidates at threshold 0.1 from 60 to 81 and precision from 0.60 to 0.67.

Interesting ideas

  • I1. Treat a translation pivot graph (source word → English word → synset → related synset) as a probabilistic network: observe the source words as evidence and rank target senses by posterior probability, so synsets reached through many paths gain support.
  • I2. Generate candidate words with a few productive derivation rules on purpose, then remove the noise with a learned filter, rather than hand-curating every derived form.
  • I3. Use the total number of correct items found by the most permissive method as the recall ceiling when true gold recall is unknown.
  • I4. Intersect two independent scorers (structural heuristics and a probabilistic model) to obtain a high-precision subset for human review.
  • I5. Inspect rejected candidates for systematic surface causes (here, missing diacritics or gender forms) before blaming the model, because one fix can recover many items at once.

For Mova

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

  • I1 (world): Unknown or rare English words could be attached to WordNet senses by propagating evidence from morphologically related known words, giving a scored and explainable ontology link instead of a guess.
  • I2 (tokens): English derivation (-er, -tion, -ness, un-) could propose lemmas and senses for out-of-vocabulary words, with a small classifier rejecting the bad ones.
  • I3 (learn): When evaluating silver-data extractors without full gold, the high-recall run's validated hits can serve as an explicit, documented recall denominator.
  • I4 (learn): Agreement between witness parsers or extractors can gate what enters silver data, with thresholds tuned along a precision–recall table like the paper's.
  • I5 (learn): Error buckets from rejected outputs could seed self-induced repair rules that address normalization issues in bulk.

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

🎓Universitat Politècnica de CatalunyaBarcelona · Spain31 works · 15,636 citations🎓Universitat de BarcelonaBarcelona · Spain12 works · 1,338 citations

 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 1Research subjects and languages the work is about.Semitic languages · 3,066 works

Cite

Cite as Horacio Rodríguez, David Farwell, Javi Ferreres, Manuel Bertran, Musa Alkhalifa and M. Antonia Martí. 2008. Arabic WordNet: Semi-automatic Extensions using Bayesian Inference. Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC'08).

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

@inproceedings{rodriguez2008arabic,
  title = {{Arabic WordNet: Semi-automatic Extensions using Bayesian Inference}},
  author = {Horacio Rodríguez, David Farwell, Javi Ferreres, Manuel Bertran, Musa Alkhalifa and M. Antonia Martí},
  booktitle = {Proceedings of the Sixth International Conference on Language Resources and Evaluation (LREC'08)},
  year = {2008},
  doi = {10.63317/548nxvhkxqip},
  url = {https://doi.org/10.63317/548nxvhkxqip}
}

References

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

  1. Agirre, E., Ansa, O., Arregi, X., Arriola, J., Diáz de Ilarraza, A., Pociello, E., Uria, L., (2002). Methodological issues in the building of the Basque WordNet: Quantitative and qualitative analysis. In Proceedings of the First International WordNet Conference, 21-25 January 2002. Mysore, India. find
  2. Barbu, E., and Barbu-Mititelu, V., (2005). A case study in automatic building of wordnets. In Proceedings of OntoLex 2005 - Ontologies and Lexical Resources. find
  3. Benítez, L., Cervell, S., Escudero, G., López, M., Rigau, G., Taulé, M., (1998). Methods and tools for building the Catalan WordNet. In Proceedings of LREC Workshop on Language Resources for European Minority Languages. find
  4. Black, W., Elkateb, S., Rodriguez, H, Alkhalifa, M., Vossen, P., Pease, A., and Fellbaum, C., (2006). Introducing the Arabic WordNet Project. In Proceedings of the Third International WordNet Conference, Fellbaum and Vossen (eds). find
  5. Chen, H., Lin, C., and Lin, W., (2002). Building a Chinese-English WordNet for translingual applications. ACM Transactions on Asian Language Information Processing 1 (2), 103–122. ↗
  6. Elkateb, S., Black, W., Rodriguez, H., Alkhalifa, M., Vossen, P., Pease, A., and Fellbaum, C., (2006). Building a WordNet for Arabic. In Proceedings of the Fifth International Conference on Language Resources and Evaluation, Genoa, Italy.
  7. Fellbaum, C. (ed.) (1998) WordNet: An Electronic Lexical Database. Cambridge, MA: MIT Press. find
  8. Farreres, J. (2005) Creation of wide-coverage domain-independent ontologies. PhD Thesis, Departament de Llenguatges i Sistemes Informàtics, Univertitat Politècnica de Catalunya, Barcelona, Spain. find
  9. Miháltz, M., and Prószéky, G., (2003). Results and evaluation of Hungarian nominal WordNet. In Proceedings of the Second International WordNet Conference. Masaryk University, Brno, pp. 175–180. find
  10. Niles, I., and Pease, A., (2001) Towards a Standard Upper Ontology. In Proceedings of FOIS 2001, Ogunquit, Maine, pp. 2-9. find
  11. Pease, A. (2003) The Sigma Ontology Development Environment. In Working Notes of the IJCAI-2003 Workshop on Ontology and Distributed Systems, Vol. 71 of the CEUR Workshop Proceeding series. find
  12. Rodríguez, R., Farwell, D., Farreres, J., Bertran, M., Alkhalifa, M., Martí, M.A., Black, W., Elkateb, S., Kirk, J., Pease, A., Vossen, P., and Fellbaum, C., (2008). Arabic WordNet: Current State and Future Extensions. Proceedings of The Fourth Global WordNet Conference, Szeged, Hungary. January 22-25, 2008. find
  13. Vossen P. (2004) EuroWordNet: a multilingual database of autonomous and language specific wordnets connected via an Inter-Lingual-Index. International Journal of Lexicography, Vol.17 No. 2, OUP, 161-173. find
  14. Witten, I.H. and Frank, E. (2005) Data mining: Practical machine learning tools and techniques (second edition). San Francisco, CA: Morgan Kaufmann. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/rodriguez-2008-arabic-wordnet-semi-automatic-extensions/