Elamur Research⌕
☰
home / papers / 2008 / A Semantic Analyzer for the Comprehension of the Spontaneous Arabic Speech

A Semantic Analyzer for the Comprehension of the Spontaneous Arabic Speech

Paper · 2008 · Mourad Mars, Mounir Zrigui, Mohamed Belgacem et al.

Summary

The paper targets semantic decoding in a spoken-language understanding system for Arabic, developed within the "Oréodule" project on Arabic speech recognition, translation and synthesis. Each meaningful word of a recognized utterance must receive a semantic feature group (FSe). The authors observe that standard models, which condition on a fixed window of immediately preceding words, perform poorly on spontaneous Modern Standard Arabic.

Their decoder works differently. It first drops non-content words through a lexical filter. It then picks, for each target word, the earlier words with the strongest statistical affinity, measured by an average mutual information score. A probabilistic model combines the semantic labels of those words with lexical cues and with the illocutionary type of the utterance (request, refusal, etc.).

The lexicon of features was built with a semi-automatic, human-in-the-loop tool. On a railway-information task, the error rate falls from 48.6–57% for fixed-history baselines to about 20%. This suggests that choosing context by relevance, rather than by position, matters for loosely ordered spontaneous speech.

Key points

  • Each word carries an FSe triple {field, semantic class, micro-semantic feature} plus a syntactic group Fsy {gender, number, nature}. Synonyms, and words with the same semantic role, share one FSe.

  • The structure of sense representation (SRS) is built through man/machine cooperation. A human assigns FSe and Fsy, while the system enforces about ten integrity constraints to keep the SRS unambiguous.

  • Several unsupervised steps support the decoder:

    • A modified TF×IDF weighting extracts "reference words" that signal the request type.
    • k-means groups words into semantic classes.
    • Average mutual information (IMm), which also accounts for word absence, ranks candidate context words. Ties go to the nearest word.

  • The training corpus has 10,000 railway-information utterances labelled with 37 FSe. The test set has 100 timing-request utterances containing 500 expert-assigned FSe.

  • Error rates by setup:

    SetupError rate
    Previous word's meaning57%
    Two previous words' meanings48.6%
    Adding utterance type and lexical data37%
    Using only the pertinent FSe~20%

    The remaining errors come mostly from syntactically complex utterances.

Interesting ideas

  • I1. Selecting conditioning context by measured association instead of a fixed n-word window lets distant but relevant words drive disambiguation.
  • I2. The speech-act type of the whole utterance (request, refusal, booking vs. timing query) serves as a prior for word-level meaning.
  • I3. The mutual information variant counts co-absence as well as co-occurrence, so a word's absence becomes evidence too.
  • I4. In the lexicon tool, the human assigns features while the machine checks integrity constraints, for example that distinct words may share a feature set only if they are synonyms.
  • I5. Breaking the error analysis down by utterance complexity revealed that the remaining failures concentrate in syntactically hard inputs.

For Mova

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

  • I1 (frames): when assigning VerbNet roles or frame slots, rank candidate tree neighbours by learned affinity, and record which ones were used so the choice stays explainable.
  • I2 (frames): feed the detected speech act from prag back into word-sense and frame selection, so that an utterance-level decision can narrow lexical readings.
  • I3 (learn): use this richer association score when inducing repair rules or verb-class cues from silver data, so that negative evidence is not thrown away.
  • I4 (world): apply similar automatic consistency checks when compiling the WordNet/FrameNet/Wikidata ontology or verb classes, so that conflicting entries are flagged for review.
  • I5 (answer): report test accuracy grouped by a syntactic complexity measure, to decide when a fallback reading strategy is needed.

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

Cite

Cite as Mourad Mars, Mounir Zrigui, Mohamed Belgacem and Anis Zouaghi. 2008. A Semantic Analyzer for the Comprehension of the Spontaneous Arabic Speech. arXiv:1610.02493 · Journal: Research in Computing Science Vol34, 2008, pp. 129-140, ISSN: 1870-4069.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.1610.02493 · arXiv 1610.02493

@misc{mars2008semantic,
  title = {{A Semantic Analyzer for the Comprehension of the Spontaneous Arabic Speech}},
  author = {Mourad Mars, Mounir Zrigui, Mohamed Belgacem and Anis Zouaghi},
  year = {2008},
  doi = {10.48550/arxiv.1610.02493},
  eprint = {1610.02493},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1610.02493}
}

References

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

  1. Cattoni, R., Federico, M., Lavie, A.: Robust analysis of spoken input combining statistical and knowledge-based information sources. ASRU, Trento (2001) find
  2. Lefèvre, F.: Estimation de probabilité non paramétrique pour la reconnaissance markovienne de la parole. Thèse de l'Université Pierre et Marie Curie (2000) find
  3. Minker, W.: Compréhension automatique de la parole spontanée. L!Harmattan, Paris (1999) find
  4. Zouaghi, A., Zrigui, M., Ben Ahmed, M.: Une structure sémantique pour l!interprétation des énoncés. JEP-TALN, Fès (2004) find
  5. Zouaghi, A., Zrigui, M., Ben Ahmed, M.: Un étiqueteur sémantique des énoncés en langue arabe. RECITAL, Dourdan (2005) find
  6. 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
  7. Chaabene, N., Belguith, L.: L!étiquetage morpho-syntaxique: comment lever l!ambiguïté dans les textes arabes non voyellés ?. Journées scientifiques des jeunes chercheurs en génie électrique et informatique, Mahdia, Tunisie (2003) find
  8. Macherey, K., Och, F.J., Ney, H.: Natural Language Understanding Using Statistical Machine Translation. European conference on speech communication and technology, Aalborg (2001) 2205-2208 Mars M., Zrigui M., Belgacem M. and Zouaghi A. find
  9. Jain, A.N., Waibel, A., Touretzky, D.S.: Parsec: a structured connexionist parsing system for spoken language. International Conference on Acoustics, Speech and Signal Processing, San Francisco, (1992) 205-208 find
  10. Jamoussi, S., Smaïli, K., Fohr, D., Haton, J.P.: Un système de compréhension automatique de la parole pour l!interrogation orale d!une base de données de bourse, JEP, Fès, Maroc (2004) 281-284 find
  11. Pepelnjak, K., Gros, J., Mihelic, F., Pave"ic, N.: Linguistic Analysis in a Slovenian information retrieval system for flight services. Workshop on Spoken Dialogue Systems, Vigso, Danemark (1995) 65-68 find
  12. Knight, S., Gorell, G., Rayner, M., Milward, D., Koeling, R., Lewin, I.: Comparing grammar-based and robust approaches to speech understanding: a case study. European conference on speech communication and technology, Aalborg, Danemark, (2001) 1779-1782 find
  13. Villaneau, J., Antoine J.Y., Ridoux, O.: Combining Syntax and Pragmatic Knowledge for the Understanding of Spontaneous Spoken Sentences. Logical Aspects of Computational Linguistic, Croisic, France (2001) 279-295 find
  14. 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, Workshop on Spoken Dialogue Systems, Vigso, Danemark, (1995) 145-148 find
  15. Beesely, K.R.: Arabic morphological analysis on the Internet. International Conference on Multi-Lingual Computing, Cambridge (1998) find
  16. Ouersighni, R.: A major offshoot of the Dinar-MBC project: AraParse, a morphosyntactic analyzer for unvowelled Arabic texts. ACL/EACL01: Conference of the European Chapter, Workshop: Arabic Language Processing: Status and Prospects (2001) find
  17. Haddad, B., Yaseen, M.: A Compositional Approach Towards Semantic Representation and Construction of ARABIC. LACL, Bordeaux, France (2005) find
  18. Meftouh, K., Laskri, M.T. http://search2.computer.org/advanced/Author_Result.jsp?qtype=3&select=50&qOpt1=DC_CREATOR&sortOrder=d&queryName=Mohamed%20Tayeb%20Laskri: Generation of the Sense of a Sentence in Arabic Language with a Connectionist Approach, AICCSA'01, Beïrut, Liban (2001). ↗
  19. AL-Johar, B., McGregor, J.: A Logical Meaning Representation for Arabic (LMRA). Proceedings of the 15th National Computer Conference, Riyadh, Saudi Arabia (1997) 31-40 find
  20. Mourad Mars, Mohamed Belgacem, Mounir Zrigui, Georges Antoniadis.: analyseur morphologique pour l'arabe. CITALA 07. Rabat Maroc (2007) find
  21. Mankaï Naanaa, C.: Compréhension automatique de la langue arabe. Application: Le système Al Biruni. Thèse de l!université de Tunis ,, (1996) find
  22. Pollard, C., Sag, I.: Head-Driven Phrase Structure Grammar. University of Chicago Press, Chicago (1994) find
  23. Zouaghi, A., Zrigui, M., Ben Ahmed, M.: A statistical model for semantic decoding of Arabic language statements. Proceedings of NODALIDA, Joensuu, Finland (2005) find
  24. Rosenfeld, R.: Adaptive statistical language modelling: A maximum entropy approach. Thesis at Carnegie Mellon University, Pittsburgh, 1994. find
  25. Seneff, S.: Robust parsing for spoken language systems, Proceedings of ICASSP (1992) 189-192 find
  26. Antoine, J-Y., Goulian, J., Villaneau, J.: Quand le TAL robuste s!attaque au langage parlé: analyze incrémentale pour la compréhension de la parole spontanée. Proceedings of TALN, Batz-sur-Mer, France (2003) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mars-2016-semantic-analyzer-comprehension-spontaneous-arabic/