Elamur Research⌕
☰
home / papers / 2008 / Compréhension automatique de la parole arabe spontanée –- Une modélisation numérique [Automatic Understanding of Spontaneous Arabic Speech –- A Numerical Model]

Compréhension automatique de la parole arabe spontanée –- Une modélisation numérique [Automatic Understanding of Spontaneous Arabic Speech –- A Numerical Model]

Paper · 2008 · Anis Zouaghi, Mounir Zrigui and Georges Antoniadis

Summary

The paper describes the semantic analyzer of "Oreillodule," a larger project on Arabic speech recognition, translation and synthesis. The analyzer handles spontaneous Modern Standard Arabic in task-oriented human–machine dialogue, with train timetables and reservations as the running example. Arabic is hard to process here for several reasons: agglutinated words can stand for a whole French phrase, the letter و can be either part of a word or the conjunction "and," word order is free, and speech recognition makes its own errors. The system works in four steps. It splits a transcribed message into separate requests, removes filler words and merges multiword units, decodes the meaning of each word, and then assembles an attribute/value frame. Decoding relies on a numerical (statistical) model that encodes the grammar rules, plus a semantic lexicon of semantic (TSE) and syntactic (TSY) features. The key claim concerns context. Using only the sense of the preceding word gives about 45% misunderstanding. Combining illocutionary, semantic and linguistic context lowers this to about 29%, and to 9.9% when there are enough training utterances.

Key points

  • The meaning of an utterance is output as a set of [attribute = value] pairs, where values are integers, reals or strings. This is a common choice in spoken dialogue systems and differs from the logical formulas used by systems such as LOGUS.
  • The pipeline runs in this order:
    1. segmentation of a message into its separate requests,
    2. preprocessing (stop-word and discourse-filler removal, word grouping),
    3. per-word semantic decoding,
    4. construction of the meaning frame.
  • Each entry in the semantic lexicon links a word to a set of semantic features (TSE) and a set of syntactic features (TSY): category, number and type.
  • Reported misunderstanding rates are:
    • about 45% using only the previous word's sense,
    • about 29% with combined illocutionary, semantic and linguistic context,
    • 9.9% with sufficient training data.
  • The keywords indicate that average mutual information and k-means are used to extract the relevant context automatically.

Interesting ideas

  • I1. The paper does not assume the adjacent word is the best context for disambiguation; it measures how different context types (speech act, neighbouring senses, linguistic cues) change the error rate. This kind of ablation over context type helps in any disambiguation step.
  • I2. One spoken message is first split into several independent requests, each interpreted separately and then combined. This avoids forcing a single frame onto a compound utterance such as "tell me the timetable and book a seat."
  • I3. The lexicon keeps semantic features and syntactic features as two separate sets for each word. Decoding can therefore check sense compatibility and grammatical fit independently.
  • I4. Ambiguity at the segmentation level (e.g., whether a letter is a conjunction or part of the word) is treated as a primary source of semantic error, not just a tokenizer detail.
  • I5. Error is reported as a function of training-set size (29% versus 9.9%), which separates model weakness from data shortage.

For Mova

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

  • I1 (frames): When prag picks a VerbNet role or a speech act, logging which context source drove the decision would allow a per-source ablation like the paper's 45%→29% comparison.
  • I2 (frames): Coordinated clauses with distinct speech acts could be split into separate frames before role assignment, keeping each step explainable.
  • I3 (world): Compiled knowledge entries could keep WordNet/FrameNet-derived semantic features apart from morphosyntactic ones, so each mismatch can be explained by which layer failed.
  • I4 (tokens): English analogues such as clitic 's and compound splitting could be scored by their downstream effect on frames, not only by tokenization accuracy.
  • I5 (learn): Evaluations of silver-trained components should include learning curves, so that a high error can be attributed to too little data or to a flawed method.

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 2The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 works

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

Cite

Cite as Anis Zouaghi, Mounir Zrigui and Georges Antoniadis. 2008. Compréhension automatique de la parole arabe spontanée –- Une modélisation numérique [Automatic Understanding of Spontaneous Arabic Speech –- A Numerical Model]. Traitement Automatique des Langues, Volume 49, Numéro 1 : Varia [Varia].

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

@inproceedings{zouaghi2008comprehension,
  title = {{Compréhension automatique de la parole arabe spontanée –- Une modélisation numérique [Automatic Understanding of Spontaneous Arabic Speech –- A Numerical Model]}},
  author = {Anis Zouaghi, Mounir Zrigui and Georges Antoniadis},
  booktitle = {Traitement Automatique des Langues, Volume 49, Numéro 1 : Varia [Varia]},
  year = {2008},
  url = {https://aclanthology.org/2008.tal-1.6/}
}

References

36 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Aloulou C., Belguith Hadrich L., Ben Hamadou A., « Utilisation des grammaires HPSG pour l’analyse de l’Arabe », JEI’2002, Hammamet, Tunisie, 2002. find
  2. Alrahabi M., Mourad G., Djioua B., « Filtrage sémantique de textes en arabe en vue d’un prototype de résumé automatique », JEP-TALN’04, Fès, Maroc, 2004. find
  3. Baccour L., Mourad G., Belguith Hadrich L., « Segmentation de textes arabes en phrases basée sur les signes de ponctuation et les mots connecteurs »,, troisième journées scientifiques des jeunes chercheurs en génie électrique et informatique, Mahdia, Tunisie, 2003. find
  4. Bennacef S., Bonneau-Maynard H., Gauvain J-L., Lamel L., Minker W., « A spoken language system for information retrieval », Actes de International Conference on Spoken Language Processing (ICSLP’94), 1994, p. 1271-1274. find
  5. Blachère R., Gaudefroy-Demombynes M., Grammaire de l’arabe classique, Maisonneuve & Larose, Paris, 1975. find
  6. Boualem M., Système de conversion de formalismes de langages techniques dans un environnement à syntaxe contrôlée et à contexte limité, Thèse de doctorat, Université de Sophia Antipolis, 1993. find
  7. Bousquet-Vernhettes C., Compréhension robuste de la parole spontanée dans le dialogue oral homme-machine – Décodage conceptuel stochastique, Thèse de l’université de Toulouse III, 2002. find
  8. Chaabene N., Belguith L., « L’étiquetage morpho-syntaxique : comment lever l’ambiguïté dans les textes arabes non voyellés ? », Proceedings des 3 ème journées scientifique des jeunes chercheurs en génie électrique et informatique, Mahdia, Tunisie, 2003. find
  9. Debili F., Traitement automatique de l’arabe voyellé ou non, Correspondances n°46, Correspondances de l’IRMC, Tunis, 2001. find
  10. Dichy J., Braham A., Ghazali S., Hassoun M., « La base de connaissances linguistiques DIINAR.1, DIctionnaire INformatisé de l’Arabe », Actes du colloque International Sympotium, université de Manouba, Tunisie, 2002. find
  11. Gaubert C., Stratégies et règles minimales pour un traitement automatique de l’arabe, Thèse de doctorat, Université Aix-Marseille I, 2001. find
  12. Gauvain J.L., Bennacef S., Devillers L., Lamel L., Rosset S., « Spoken language component of the MASK kiosk », Human confort & security systems, édité par Varghese K., Pfleger S., Springer Verlag, 1997, p. 93-103. find
  13. Ghazali S., Habaili H., Zrigui M., « Correspondance graphème phonème pour la synthèse de la parole arabe à partir du texte », IRSIT, Actes du Congrès dialogue homme machine, Tunis, 1990. find
  14. Haddad B., Yaseen M., « Towards understanding arabic : A logical approach for semantic representation », Arabic NLP workshop, ACL’01, 2001 find
  15. Haddad B., Yaseen M., « A Compositional Approach Towards Semantic Representation and Construction of ARABIC », Proceedings of LACL, 2005. find
  16. Compréhension automatique de la parole arabe spontanée Jaccarini A., Grammaire modulaires de l’arabe. Modélisation, mise en œuvre informatique et stratégies, Thèse de doctorat, Université de Paris IV – Sorbonne, 1997. find
  17. Knight S., Gorell G., Rayner M., Milward D., Koeling R., Lewin I., « Comparing grammar-based and robust approaches to speech understanding : a case study », Proceedings of European conference on speech communication and technology, 2001. find
  18. Kurdi M-Z., Contribution à l’analyse du langage oral spontané, Thèse de l’université Joseph-Fourier, Grenoble, 2003. find
  19. Larkey L.S., Ballesteros L., Connell M., « Improving Stemming for Arabic Information Retrieval : Light Stemming and Cooccurrence Analysis », In Proceedings of the 25th Annual International Conference on Research and Development in Information Retrieval (SIGIR 2002), Tampere, Finland, 2002, p. 275-282. find
  20. Lopez P., Roussel D., « Which rules for the robust parsing of spoken utterances with lexicalized tree adjoining grammars ? », Fourth international workshop on tree adjoining grammars (TAG+4), Philadelphie, Etats-Unis, 1998, p. 96-99. find
  21. Mankaï Naanaa C., Compréhension automatique de la langue arabe. Application : Le système Al Biruni, Thèse de doctorat de l’université de Tunis ΙΙ, 1996. find
  22. Maynard H., Lefèvre F., « Apprentissage d’un modèle stochastique de compréhension de la parole », Actes de JEP’2002, 2002. find
  23. Meftouh K., Laskri M.T., « Generation of the Sense of a Sentence in Arabic Language with a Connectionist Approach », Actes de AICCSA, 2001. find
  24. Minker W., Compréhension automatique de la parole spontanée, L’Harmattan, Paris, 1999. find
  25. Ouersighni R., » A major offshoot of the Dinar-MBC project : AraParse, a morphosyntactic analyzer for unvowelled Arabic texts », Proceedings of ACL/EACL, 2001. find
  26. Pepelnjak K., Gros J., Mihelic F., Pavešic N., « Linguistic Analysis in a Slovenian information retrieval system for flight services», ESCA Workshop on Spoken Dialogue Systems, Vigso, Danemark, 1995, p. 65-68. find
  27. Rakotomalala R., « TANAGRA : un logiciel gratuit pour l’enseignement et la recherche », Actes de EGC’2005, vol. 2, n° 3, 2005, p.697-702. find
  28. Rosenfeld R., Adaptive statistical language modelling : A maximum entropy approach, Thèse de doctorat de l’université de Carnegie Mellon, 1994. find
  29. Sadek D., Bretier P., Cadoret V., Cozannet A., Dupont P., Ferrieux A., Panaget F., « A Cooperative Spoken Dialogue System Based on a Rational Agent Model : A First Implementation on the AGS Application », ESCA Workshop on Spoken Dialogue Systems, Vigso, Danemark, 1995, p. 145-148. find
  30. Saïdane T., Zrigui M., Ben Ahmed M., « Arabic speech synthesis using a concatenation of polyphones : the results », Proceedings of Canadian Conference on AI, Montréal, Canada, 2005. find
  31. Seneff S., « Robust parsing for spoken language systems », Actes de ICASSP, 1992, p.189-192. find
  32. Servan C., Bechet F., « Décodage conceptuel et apprentissage automatique : application au TAL. Volume 49 – n° 1/2008 corpus de dialogue Homme-Machine MEDIA », TALN 2006, Leuven, Belgium, 2006. find
  33. Tuerlinckx L., « La lemmatisation de l’arabe non classique », JADT’04, Paris, 2004. find
  34. Van Noord G., Bouma G., Koeling R., Nederhof M.J., Robust grammatical analysis for spoken dialogue systems, Natural Language Engineering, 1999. find
  35. Villaneau J., Ridoux O., Antoine J-Y., « LOGUS : compréhension de l’oral spontané Présentation et évaluation des bases formelles de LOGUS », RSTI – RIA, 2004, p. 709-742. find
  36. Yan P., Zheng F., Xu M., « Robust Parsing in Spoken Dialogue Systems », European conference on speech communication and technology (EUROSPEECH’01), Aalborg, Danemark, Vol. 3, 2001, p. 2149-2152. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zouaghi-2008-comprehension-automatique-de-la-parole/