Elamur Research⌕
☰
home / papers / 2022 / A Romanization System and WebMAUS Aligner for Arabic Varieties

A Romanization System and WebMAUS Aligner for Arabic Varieties

Summary

Arabic has been missing from the BAS WebMAUS forced-alignment suite for two reasons. Phonetically annotated speech for individual varieties is scarce, and Arabic script maps poorly onto dialect phonology: short vowels are usually left unwritten, and spoken varieties have no standard spelling. The authors avoid both problems. They do not start from Classical Arabic script, and they do not rely on a pronunciation lexicon covering all dialects. Instead they design Al-Tamimi Romanization (ATR), an ASCII transcription in which each symbol stands for exactly one phoneme. It is meant to cover nearly every Arabic variety. On top of ATR they trained acoustic models and integrated a variety-independent Arabic aligner into WebMAUS. The aligner can be used through the web interface or called programmatically over REST, and the paper gives step-by-step recipes for both. Version 1 was released in December 2021 and version 2 in April 2022. The resource matters because it gives phoneticians a free, open tool for segmenting large dialectal Arabic corpora at the phoneme level. Until now that kind of work was blocked by unvowelized, script-only transcriptions.

Key points

  • ATR has 98 phonemes in two mirrored sets:

    • a "long" set: geminate consonants and long vowels;
    • a "short" set: singleton consonants and short vowels.

    Each set has its own graphemes and its own acoustic models. Every symbol is also mapped to IPA and X-SAMPA.

  • The inventory grew out of the team's research on several varieties:

    • Lebanese, Levantine, Bahraini, Jizani and Saudi (Jeddah) Arabic;
    • Algerian Arabic–French code-switching.

    It includes dialect-specific sounds such as /zˤ/ and /ɡ/.

  • The authors explain why existing schemes fall short:

    • Arabizi conflates sounds; for example, both "2" and "a" are used for hamza.
    • Buckwalter can encode short vowels, but they are rarely written in practice.
    • Most LDC Arabic datasets are in Arabic script without vowelization.

  • At the time of writing, WebMAUS offers 22 tools, supports 42 languages and varieties, and uses 1243 HTK-based phone models. It has processed about 17 million files since 2013.

  • Google Cloud Speech-to-text covers 15 Arabic varieties, but it outputs only unvowelized script.

  • Arabic WebMAUS v1 was released on 2021-12-17 and v2 on 2022-04-15.

Interesting ideas

  • I1. When the standard spelling hides the information you need, define a transparent one-symbol-per-unit transcription layer and work on that, rather than on the lossy orthography.
  • I2. The long/short sets mirror each other but get separate symbols and separate models, because the variants differ systematically and measurably. Distinctions with real acoustic consequences are kept apart instead of merged.
  • I3. One superset inventory covers many varieties, and dialect-specific units are simply members of it, so a single model serves all varieties without a per-variety lexicon.
  • I4. The service checks its own calls: each one returns an explicit success flag, and the authors list the usual failure causes, such as typographic instead of ASCII apostrophes or a missing "@" before file options.

For Mova

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

  • I1 (tokens): Normalized tokens could carry a deterministic, reversible annotation layer for spelling variants and contractions, so each downstream decision has an explicit input form.
  • I2 (tree): During UD dialect conversion, constructions that behave differently should keep distinct labels or features rather than being collapsed to one relation, so the parser can learn each pattern separately.
  • I3 (world): A single ontology could absorb regional or domain-specific senses as tagged members of one inventory, instead of separate per-domain lexicons.
  • I4 (answer): Every answering call could return an explicit status plus a checklist of known input pitfalls, so failures are explainable rather than silent.

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: Linguistics & languages

Topics Semitic languages

Cite

Cite as Jalal Al-Tamimi, Florian Schiel, Ghada Khattab, Navdeep Sokhey, Djegdjiga Amazouz, Abdulrahman Dallak and Hajar Moussa. 2022. A Romanization System and WebMAUS Aligner for Arabic Varieties. Proceedings of the Thirteenth Language Resources and Evaluation Conference.

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

@inproceedings{altamimi2022romanization,
  title = {{A Romanization System and WebMAUS Aligner for Arabic Varieties}},
  author = {Jalal Al-Tamimi, Florian Schiel, Ghada Khattab, Navdeep Sokhey, Djegdjiga Amazouz, Abdulrahman Dallak and Hajar Moussa},
  booktitle = {Proceedings of the Thirteenth Language Resources and Evaluation Conference},
  year = {2022},
  url = {https://aclanthology.org/2022.lrec-1.789/}
}

References

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

  1. Al-Tamimi, J. and Ferragne, E. (2005). Does vowel space size depend on language vowel inventories? Evidence from two Arabic dialects and French. In Interspeech, pages 2465–2468. find
  2. Al-Tamimi, J. and Khattab, G. (2011). Multiple cues for the singleton-geminate contrast in Lebanese Arabic: Acoustic investigation of stops and fricatives. Proceedings of the 17th ICPhS, Hong Kong, China. find
  3. Al-Tamimi, J. and Khattab, G. (2015). Acoustic cue weighting in the singleton vs geminate contrast in Lebanese Arabic: The case of fricative consonants. Journal of the Acoustical Society of America, 138(1):344–360. find
  4. Al-Tamimi, J. and Khattab, G. (2018). Acoustic correlates of the voicing contrast in Lebanese Arabic singleton and geminate stops. Journal of Phonetics, Invited manuscript for the special issue of Journal of Phonetics, “Marking 50 Years of Research on Voice Onset Time and the Voicing Contrast in the World’s Languages” (eds., T. Cho, G. Docherty & D. Whalen), 71:306–325. find
  5. Al-Tamimi, J. (2007a). Indices Dynamiques et Perception des Voyelles : Étude Translinguistique en Arabe Dialectal et en Français. Ph.D. thesis, Université Lyon 2, Lyon, France. find
  6. Al-Tamimi, J. (2007b). Static and Dynamic cues in vowel production: A cross dialectal study in Jordanian and Moroccan Arabic. In Proceedings of the 16th International Congress of Phonetic Sciences (ICPhS), Saarbrücken, Germany, pages 541–544. find
  7. Al-Tamimi, J. (2017). Revisiting acoustic correlates of pharyngealization in Jordanian and Moroccan Arabic: Implications for formal representations. Laboratory Phonology: Journal of the Association for Laboratory Phonology, 8(1):1– 40, November. find
  8. Al-Tamimi, J. (2021). Acoustic correlates of epilaryngeal constriction in Levantine Arabic guttural consonants. In Proceedings of the 4th Phonetics and Phonology in Europe (PaPE), Barcelona, Spain, 21-23 June 2021. find
  9. Amazouz, D., Adda-Decker, M., and Lamel, L. (2018). The French-Algerian Code-Switching Triggered audio corpus (FACST). page 8. find
  10. Amazouz, D. (2019). Linguistic and Phonetic Investigations of French-Algerian Arabic Code-Switching:. Ph.D. thesis. find
  11. Khattab, G. and Al-Tamimi, J. (2013). Influence of geminate structure on early Arabic templatic patterns. In M Vihman et al., editors, The Emergence of Phonology: Whole-word Approaches and Cross-linguistic Evidence, pages 374–414. January. find
  12. Khattab, G. and Al-Tamimi, J. (2014). Geminate timing in Lebanese Arabic: The relationship between phonetic timing and phonological structure. Journal of Laboratory Phonology, 5(2):231–269, January. find
  13. Khattab, G. and Al-Tamimi, J. (2015). The acquisition of gemination in Lebanese Arabic Children (Article: 0870). In The Scottish Consortium for ICPhS 2015 (Ed.), Proceedings of the 18th International Congress of Phonetic Sciences. Glasgow, UK: The University of Glasgow. ISBN 978-0-85261-941-4. find
  14. Kisler, T., Reichel, U., and Schiel, F. (2017). Multilingual processing of speech via web services. Computer Speech & Language, 45:326– 347, September. find
  15. Moussa, H. and Al-Tamimi, J. (2018). Focus and F0 patterns in Jeddah Arabic. In Proceedings of the BAAP (British Association of Academic Phoneticians), Kent University, UK. find
  16. Moussa, H. A. (2020). Jeddah Arabic Intonation: An Autosegmental-metrical Approach. Ph.D. thesis, Newcastle University. find
  17. Schiel, F. (1999). Automatic Phonetic Transcription of Non-Prompted Speech. In Proc. of the ICPhS, pages 607–610, San Francisco, August. find
  18. Sokhey, N. (2019). Vowel-Rounding in Bahrain: An Interactional Study on Accommodation Based on Perceived Interlocutor Identity. Ph.D. thesis. find
  19. Winkelmann, R., Harrington, J., and Jänsch, K. (2017). EMU-SDMS: Advanced speech database management and analysis in R. Computer Speech & Language, 45:392–410, September. find
  20. Young, S., Evermann, G., Gales, M., Hain, T., Kershaw, D., Liu, X., Moore, G., Odell, J., Ollason, D., Povey, D., Ragni, A., Valtchev, V., Woodland, P., and Zhang, C. (2015). The HTK Book (version 3.5a). 12. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/tamimi-2022-romanization-system-webmaus-aligner-arabic/