Elamur Research⌕
☰
home / papers / 2017 / Hierarchical Classification for Spoken Arabic Dialect Identification using Prosody: Case of Algerian Dialects

Hierarchical Classification for Spoken Arabic Dialect Identification using Prosody: Case of Algerian Dialects

Paper · 2017 · Soumia Bougrine, Hadda Cherroun and Djelloul Ziadi

Summary

The paper addresses automatic identification of spoken Arabic dialects that come from a single country, here Algeria. This is hard because the varieties are closely related and few resources exist for them. The authors first test statistically whether prosodic cues separate Algerian dialects. These cues are rhythm, intonation and duration-type measurements, taken per utterance after a rough split of speech into consonant and vowel segments. Based on that analysis they build HADID, a top-down hierarchical classifier. Its tree of dialects comes from historical and linguistic knowledge, and each internal node gets its own local DNN classifier (Local Classifier per Parent Node, LCPN). On a corpus of 1892 Algerian utterances averaging about 6 seconds, HADID reaches 83.8% precision for the coarse split between Pre-Hilālī and Bedouin dialect groups and 62.8% across five dialects. A flat DNN over the same features gets only 38.4%. The work matters because it shows that cheap, noise-robust prosodic features combined with a linguistically motivated hierarchy can handle close, under-resourced varieties from short speech samples.

Key points

  • Data: an Algerian Arabic dialect corpus of 1892 utterances (about 6 s each). Evaluation uses 5-fold cross-validation and hierarchical micro-precision (hP).
  • Level-1 of the hierarchy separates Pre-Hilālī from Bedouin dialects. Precision there averages 83.8% (the conclusion states 83.6%), and the whole five-dialect system averages 62.8%. No fold deviates from the mean by more than 5.2%.
  • HADID with DNNs beats HADID-SVM (62.8% vs 58.6% overall; 83.8% vs 83.0% at Level-1). The SVM variant's per-dialect results are far more uneven, with roughly a 76-point gap between its best and worst dialect.
  • A flat DNN baseline using the same preprocessing and the same ANOVA-ranked prosodic features scores 38.4%. HADID's gain is a relative improvement of about 63.5%.
  • By dialect, Ma'qilian (87%) and Sulaymite (72%) score best, and UCB scores worst (44%). UCB is misassigned to Bedouin dialects, and the flat system never recognises it at all.

Interesting ideas

  • I1. Instead of one big classifier, route each item down an expert-defined tree and train a separate small classifier at every branching point, so each decision only has to separate a few close options.
  • I2. Before building any model, the authors run a statistical analysis (with ANOVA-based ranking) to confirm the chosen features actually discriminate the classes and to keep only the strongest ones.
  • I3. Systematic confusions of one class (UCB going to Bedouin) are read as evidence that the hierarchy itself is wrong, not just the classifier.
  • I4. Results are reported per hierarchy level, per class and per fold, with spread figures, so a decent average cannot hide a collapsed class.
  • I5. The flat baseline uses exactly the same preprocessing and features, so the measured gain isolates the effect of the hierarchical design.

For Mova

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

  • I1 (world): assigning a noun or verb to a WordNet/VerbNet class could proceed coarse-to-fine along the ontology, with a separate explainable decision logged at each level.
  • I2 (learn): new parser or tagger features could first pass an ANOVA-style discriminativeness check on silver data, so features that are statistically inert never enter training.
  • I3 (tree): recurring attachment or label confusions in UD dialect conversion could flag places where our label mapping or treebank grouping needs restructuring rather than more data.
  • I4 (answer): test scores on fables or word problems could be broken down by reasoning step and question type, with fold variance, to expose categories the readers silently fail.
  • I5 (learn): every architectural change (e.g., gparse vs. perceptron routing) should be compared against a control that differs only in that single design choice.

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 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSignal Processing · 452 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksSpeech & spoken language · 480 works

Cite

Cite as Soumia Bougrine, Hadda Cherroun and Djelloul Ziadi. 2017. Hierarchical Classification for Spoken Arabic Dialect Identification using Prosody: Case of Algerian Dialects. arXiv:1703.10065.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.1703.10065 · arXiv 1703.10065

@misc{bougrine2017hierarchical,
  title = {{Hierarchical Classification for Spoken Arabic Dialect Identification using Prosody: Case of Algerian Dialects}},
  author = {Soumia Bougrine, Hadda Cherroun and Djelloul Ziadi},
  year = {2017},
  doi = {10.48550/arxiv.1703.10065},
  eprint = {1703.10065},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.1703.10065}
}

References

58 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. H. Li, B. Ma, K. Lee, Spoken language recognition: From fundamentals to practice, Proceedings of the IEEE 101 (5) (2013) 1136–1159. find
  2. K. S. Rao, V. R. Reddy, S. Maity, Language Identification Using Spectral and Prosodic Features, Springer, 2015. find
  3. T. Rong, M. Bin, Z. Donglai, L. Haizhou, C. E. Siong, Integrating acoustic, prosodic and phonotactic features for spoken language identification, in: Int. Conf. Acoust. Speech and Signal Process., 2006, pp. 205–208. find
  4. M. Djellab, A. Amrouche, A. Bouridane, N. Mehallegue, Algerian Modern Colloquial Arabic Speech Corpus (AMCASC): regional accents recognition within complex socio-linguistic environments, Language Resources and Evaluation (2016) 1–29. find
  5. A. Hanani, H. Basha, Y. Sharaf, S. Taylor, Palestinian Arabic Regional Accent Recognition, in: Speech Technology and Human-Computer Dialogue (SpeD), 2015, pp. 1–6. find
  6. M. P. Lewis, F. S. Gary, D. F. Charles, Ethnologue: Languages of the World, Eighteenth edition, Web (2015). URL http://www.ethnologue.com/statistics/size ↗
  7. M. Embarki, Les dialectes arabes modernes: état et nouvelles perspectives pour la classification géo-sociologique, Arabica 55 (2008) 583–604. find
  8. K. Versteegh, The Arabic Language, Edinburgh University Press, Cambridge, 1997. find
  9. J. Silla, N. Carlos, A. A. Freitas, A survey of hierarchical classification across different application domains, Data Mining and Knowledge Discovery 22 (1-2) (2011) 31–72. find
  10. J. C. Wells, Accents of English, Vol. 2, Cambridge University Press, 1982. find
  11. R. Collobert, J. Weston, A Unified Architecture for Natural Language Processing: Deep Neural Networks with Multitask Learning, in: Proceedings of the 25th international conference on Machine learning, ACM, 2008, pp. 160–167. find
  12. R. Sahraeian, D. Van Compernolle, F. de Wet, Under-resourced speech recognition based on the speech manifold, in: Sixteenth Annual Conference of the International Speech Communication Association, 2015. find
  13. F. Biadsy, J. Hirschberg, N. Habash, Spoken Arabic Dialect Identification Using Phonotactic Modeling, in: EACL, Athens, Greece, 2009, pp. 53–61.
  14. M. Akbacak, D. Vergyri, A. Stolcke, N. Scheffer, A. Mandal, Effective Arabic Dialect Classification Using Diverse Phonotactic Models, in: Interspeech, Florence, Italy, 2011, pp. 737–740. find
  15. J. L. Rouas, M. Barkat-Defradas, F. Pellegrino, R. Hamdi, Identification automatique des parlers arabes par la prosodie, in: Journées d’Etude sur la Parole, Dinard, France, 2006, pp. 193–196. find
  16. F. Biadsy, J. Hirschberg, Using Prosody and Phonotactics in Arabic Dialect Identification, in: Interspeech, Brighton, UK, 2009, pp. 208–211. find
  17. F. Richardson, W. Campbell, P. Torres-Carrasquillo, Discriminative N-Gram Selection for Dialect Recognition, in: Interspeech, Brighton, UK, 2009, pp. 297– 307. find
  18. C. Greenberg, A. Martin, M. Przybocki, The 2011 NIST Language Recognition Evaluation, in: Interspeech, Portland, Oregon, USA, 2012. find
  19. A. Ali, D. Najim, C. Patrick, K. Sameer, Y. Sree Harsha, J. Glass, P. Bell, S. Renals, Automatic Dialect Detection in Arabic Broadcast Speech, arXiv preprint arXiv:1509.06928v2, 2016. ↗
  20. F. S. Alorfi, Automatic Identification of Arabic Dialects Using Hidden Markov Models, Ph.D. thesis, University of Pittsburgh, USA (2008). find
  21. F. Biadsy, J. Hirschberg, D. P. W. Ellis, Dialect and Accent Recognition Using Phonetic-Segmentation Supervectors, in: Interspeech, Florence, Italy, 2011, pp. 745–748. find
  22. M. Al-Ayyoub, M. K. Rihani, N. I. Dalgamoni, N. A. Abdulla, Spoken Arabic dialects identification: The case of Egyptian and Jordanian dialects, in: ICICS, Irbid, Jordan, 2014, pp. 1–6. find
  23. N. Lachachi, A. Adla, GMM-Based Maghreb Dialect Identification System, Journal of Information Processing Systems 11 (1) (March 2015) 22–38. find
  24. M. A. Zissman, T. P. Gleason, D. M. Rekart, B. L. Losiewicz, Automatic Dialect Identification of Extemporaneous Conversational, Latin American Spanish Speech, in: ICASSP, Atlanta, Georgia, 1996, pp. 777–780. find
  25. S. Bougrine, H. Cherroun, D. Ziadi, A. Lakhdari, A. Chorana, Toward a Rich Arabic Speech Parallel Corpus for Algerian sub-Dialects, in: LREC’16 Workshop on Free/Open-Source Arabic Corpora and Corpora Processing Tools (OSACT), 2016, pp. 2–10. find
  26. W. Zaghouani, Critical Survey of the Freely Available Arabic Corpora, in: LREC’14 Workshop on Free/Open-Source Arabic Corpora and Corpora Processing Tools (OSACT). Reykjavik, Iceland, 2014, pp. 1–8. find
  27. S. Ghazali, R. Hamdi, M. Barkat, Speech Rhythm Variation in Arabic Dialects, in: Proceedings of Prosody, Aix-en-Provence, France, 2002, pp. 331–334. find
  28. S. Jothilakshmi, V. Ramalingam, S. Palanivel, A hierarchical language identification system for Indian languages, Digital Signal Processing 22 (3) (2012) 544– 553. find
  29. B. Yin, E. Ambikairajah, F. Chen, Hierarchical Language Identification based on Automatic Language Clustering, in: Interspeech, 2007, pp. 178–181. find
  30. H. Wang, X. Xiao, X. Zhang, J. Zhang, Y. Yan, A hierarchical system design for language identification, in: ISISE, 2009, pp. 443–446. find
  31. A. F. B. Laguna, R. C. L. Guevara, Development, Implementation and Testing of Language Identification System for Seven Philippine Languages, Philippine Journal of Science 144 (1) (2015) 81–89. find
  32. I. Lopez-Moreno, J. Gonzalez-Dominguez, D. Martinez, O. Plchot, J. Gonzalez-Rodriguez, P. J. Moreno, On the use of deep feedforward neural networks for automatic language identification, Computer Speech and Language 40 (2016) 46 – 59. find
  33. L. Ferrer, Y. Lei, M. McLaren, N. Scheffer, Study of Senone-Based Deep Neural Network Approaches for Spoken Language Recognition, IEEE/ACM Transactions on Audio, Speech, and Language Processing 24 (1) (2016) 105–116. find
  34. J. Leclerc, Algérie dans l’aménagement linguistique dans le monde, Web (30 avril 2012). URL http://www.tlfq.ulaval.ca/axl/europe/danemark.htm ↗
  35. P. Marçais, Algeria, Leiden: E.J. Brill, 1986, pp. 374–379. find
  36. D. Caubet, Questionnaire de dialectologie du Maghreb (d’après les travaux de W. Marçais, M. Cohen, GS Colin, J. Cantineau, D. Cohen, Ph. Marçais, S. Lévy, etc.), Estudios de dialectologı́a norteafricana y andalusı́, EDNA (5) (2000) 73–90. find
  37. P. Marçais, Esquisse grammaticale de l’arabe maghrébin, Librairie d’Amérique et d’Orient, 1977. find
  38. H. Palva, Dialects: classification, Encyclopedia of Arabic Language and Linguistics 1 (2006) 604–613. find
  39. C. Pereira, Arabic in the North African Region, in: S. Weniger, G. Khan, M. P. Streck, J. C. E. Watson (Eds.), Semitic Languages. An International Handbook, Berlin, 2011, pp. 944–959. find
  40. H. Palva, Arabic dialects: Classification, Leiden Boston: Brill, 2006, pp. 604– 613. find
  41. M. Barkat, J. Ohala, F. Pellegrino, Prosody as a Distinctive Feature for the Discrimination of Arabic Dialects, in: Eurospeech, Budapest, Hungary, 1999, pp. 395–398. find
  42. R. Hamdi, La variation rythmique dans les dialectes arabes, Ph.D. thesis, University Lumière Lyon 2 and University 7 Novembre, Tunisie (2007). find
  43. R. Bouziri, H. Nejmi, M. Taki, L’accent de l’arabe parlé à Casablanca et à Tunis: étude phonétique et phonologique, in: ICPhS, Aix-en-Provence, 1991, pp. 134– 137. find
  44. S. Ghazali, R. Hamdi, K. Knis, Intonational and Rhythmic patterns across the Arabic dialect continuum, in: 19th Annual Symposium on Arabic Linguistics, Illinois, USA, 2005. find
  45. M. Yeou, M. Embarki, S. Al-Maqtari, Contrastive focus and F0 patterns in three Arabic dialects, Nouveaux Cahiers de Linguistique Française 28 (2007) 317–326. find
  46. I. Benali, Le rôle de la prosodie dans l’identification de deux parlers algériens: l’algérois et l’oranais, in: MIDL, Paris, France, 2004, pp. 128–132. find
  47. F. Ramus, M. Nespor, J. Mehler, Correlates of Linguistic Rhythm in the Speech Signal, Cognition 75 (1) (1999) 265–292. find
  48. V. Dellwo, Rhythm and Speech Rate: a Variation Coefficient for deltaC, in: Language and language-processing, 2006, pp. 231–241. find
  49. E. Grabe, E. L. Low, Durational Variability in Speech and the Rhythm Class Hypothesis, Papers in laboratory phonology 7 (2002) 515–546. find
  50. D. Crystal (Ed.), A Dictionary of Linguistics and Phonetics, Vol. 6 of Lecture Notes in Statistics, Blackwell, Malden, 2008. find
  51. M. Alsulaiman, G. Muhammad, M. A. Bencherif, A. Mahmood, Z. Ali, KSU Rich Arabic Speech Database, Journal of Information 16 (6) (2013) 4231–4253. find
  52. R. Hamdi, M. Barkat-defradas, E. Ferragne, F. Pellegrino, Speech Timing and Rhythmic Structure in Arabic Dialects: A Comparison of Two Approaches, in: Interspeech, 2004. find
  53. R. Hamdi, S. Ghazali, M. Barkat, Syllable Structure in Spoken Arabic : a comparative investigation, in: Interspeech, Lisbonne, Portugal, 2005, pp. 2245–2248. find
  54. G. Droua-Hamdani, S.-A. Selouani, M. Boudraa, W. Cichocki, Algerian arabic rhythm classification., in: ExLing, 2010, pp. 37–40. find
  55. I. Salselas, P. Herrera, Music and speech in early development: automatic analysis and classification of prosodic features from two portuguese variants, J. Portuguese Linguistics 9 (10) (2011) 11–36. find
  56. I. Braga, L. P. Carmo, C. C. Benatti, M. C. Monard, Advances in Soft Computing and Its Applications., 2013, Ch. A Note on Parameter Selection for Support Vector Machines, pp. 233–244. find
  57. T. Wu, Feature selection in speech and speaker recognition, Ph.D. thesis, Katholieke Universiteit Leuven, Leuven (2009). find
  58. S. Kiritchenko, S. Matwin, R. Nock, A. F. Famili, Learning and Evaluation in the Presence of Class Hierarchies: Application to Text Categorization, in: Proceedings of the 19th Canadian Conference on Artificial Intelligence, Qubec, Canada., 2006, pp. 395–406. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bougrine-2017-hierarchical-classification-spoken-arabic-dialect/