Elamur Research⌕
☰
home / papers / 2024 / Towards a Corpus of Spoken Maltese: Korpus tal-Malti Mitkellem, KMM

Towards a Corpus of Spoken Maltese: Korpus tal-Malti Mitkellem, KMM

Paper · 2024 · Alexandra (Sandra) Vella, Sarah Agius, Aiden Williams et al.

Summary

Maltese is a low-resource language whose existing speech material does not reflect everyday talk. Most of it consists of edited parliamentary transcripts or read-aloud prompts. The authors argue for a purpose-built spoken corpus, Korpus tal-Malti Mitkellem (KMM). Instead of collecting whatever data happens to be available, KMM is planned around representativeness and balance. Its centre is a carefully annotated, gold-standard Core collection. Around it sit two kinds of Satellite material: Donated corpora from earlier projects and Harvested media such as podcasts, radio and TV.

The Core is recorded from volunteers across a planned range of text-types. These run from reading tasks, where the text is known in advance, to prompted, task-oriented and free speech, in monologue (Mono) or dialogue (Dia). About 20 hours have been compiled and annotated by humans so far.

The authors also ran a small qualitative comparison between a Maltese ASR system's output and the human transcripts. The ASR proved good enough to give annotators a first draft, which could cut transcription effort and cost. This matters because high cost is the main reason spoken corpora remain scarce.

Key points

  • KMM structure:
    • a gold-standard Core collection that can be extended in principled ways;
    • Satellite material, split into Donated (from other projects) and Harvested (online programmes, podcasts, broadcasts).
  • The current Core holds roughly 20 hours of speech with human annotation.
  • Core speaking tasks vary in how scripted they are (read, prompted, task-oriented, free) and in how many speakers take part (Mono vs Dia).
  • Existing Maltese speech resources reviewed:
    • Korpus Malti v3.0 parliamentary transcripts (c. 43.4M + 50M tokens) and an 18,000-token speeches set;
    • Common Voice (18h, half validated);
    • MASRI-Headset (8h);
    • CommonLanguage (1h);
    • smaller corpora such as MalToBI and MAMCO.
  • The authors define a spoken corpus in four layers: audio, transcript, time-aligned transcription, and speaker/recording metadata. As context on scarcity, CLARIN lists 133 spoken corpora, but they cover only 15 languages.

Interesting ideas

  • I1. Gold core plus satellites. A small, deliberately balanced gold collection sits at the centre, with larger donated or harvested material of lower certainty kept separate from it, so each tier's quality stays known.
  • I2. Graded text-types. Data runs from easy cases, where the reference text exists in advance, to fully unscripted speech, giving a built-in difficulty gradient within one resource.
  • I3. Machine draft, human correction. Automatic output serves as a first pass that humans edit, and the draft is judged directly against human output to decide whether it actually saves effort.
  • I4. Edited records are not natural language. Cleaned-up records such as parliamentary transcripts differ stylistically from real conversation, with rhetorical repetition and different pause and syntax patterns, so they are a poor stand-in for natural usage.
  • I5. Layered resource design for several user groups. A minimal layer (audio plus text) serves technology users, while extra layers (alignment, metadata) serve researchers, which spreads the high cost of building the resource across more uses.

For Mova

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

  • I1 (learn): Keep a small hand-checked gold set strictly apart from silver data agreed by witness parsers, and label each sentence's tier so that training and evaluation never mix the two by accident.
  • I2 (answer): Build reader test suites in graded bands, from clean textbook prose to messy dialogue-like text, so drops in accuracy can be traced to a specific difficulty level.
  • I3 (learn): Let parsers or the LLM teacher pre-annotate UD trees and frames for human correction, and log the edit distance as a measure of whether the draft tier is worth using.
  • I4 (tokens): Do not assume that edited newswire-style English covers conversational input, and add explicit tests for disfluencies, repetitions and fillers in the tokenizer and tagger.
  • I5 (world): Store character-span alignment and source metadata alongside every tree node and world entity, so one annotated corpus can support both training and explainable tracing back to the source.

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 · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 2,908 works

Cite

Cite as Alexandra (Sandra) Vella, Sarah Agius, Aiden Williams and Claudia Borg. 2024. Towards a Corpus of Spoken Maltese: Korpus tal-Malti Mitkellem, KMM. Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024).

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

@inproceedings{vella2024corpus,
  title = {{Towards a Corpus of Spoken Maltese: Korpus tal-Malti Mitkellem, KMM}},
  author = {Alexandra (Sandra) Vella, Sarah Agius, Aiden Williams and Claudia Borg},
  booktitle = {Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)},
  year = {2024},
  doi = {10.63317/5k7ktu5qumgk},
  url = {https://doi.org/10.63317/5k7ktu5qumgk}
}

References

29 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Anne H. Anderson, Miles Bader, Ellen Gurman Bard, Elizabeth Boyle, Gwyneth Doherty, Simon Garrod, Stephen Isard, Jacqueline Kowtko, Jan McAllister, Jim Miller, Catherine Sotillo, Henry S. Thompson, and Regina Weinert. 1991. The HCRC map task corpus. Language and Speech, 34(4):351–366. ↗
  2. Joseph Aquilina. 1987. Maltese-English Dictionary Vol. I, A-L. Midsea Books, Valletta, Malta. find
  3. Joseph Aquilina. 1990. Maltese-English Dictionary Vol. II, M-Z. Midsea Books, Valletta, Malta. find
  4. Alexei Baevski, Yuhao Zhou, Abdelrahman Mohamed, and Michael Auli. 2020. wav2vec 2.0: A framework for self-supervised learning of speech representations. Advances in Neural Information Processing Systems, 33:12449–12460.
  5. Paul Boersma and David Weenink. 2001. PRAAT, a system for doing phonetics by computer. Glot International, 5(9/10):341–345. find
  6. Gabriela Cavaglià. 2002. Measuring corpus homogeneity using a range of measures for inter-document distance. In Proceedings of the Third International Conference on Language Resources and Evaluation (LREC 2002), Las Palmas, Canary Islands - Spain. European Language Resources Association. find
  7. Keith Cortis, Judie Attard, and Donatienne Spiteri. 2021. Malta National Language Technology Platform: A vision for enhancing Malta’s official languages using Machine Translation. In Proceedings of the First Workshop on Multimodal Machine Translation for Low Resource Languages (MMTLRL 2021), pages 12–19, Online (Virtual Mode). INCOMA Ltd. find
  8. Per Fallgren, Zofia Malisz, and Jens Edlund. 2019. How to annotate 100 hours in 45 minutes. In Proceedings of the Annual Conference of the International Speech Communication Association, INTERSPEECH, pages 341–345. ISCA. QC 20200310. find
  9. Albert Gatt and Slavomír Čéplö. 2013. Digital corpora and other electronic resources for Maltese. In Proceedings of the International Conference on Corpus Linguistics, pages 96–97. UCREL, Lancaster, UK. find
  10. Dawn Knight and Svenja Adolphs. 2022. Building a spoken corpus: what are the basics? In Anne O’Keeffe and Michael J. McCarthy, editors, The Routledge Handbook of Corpus Linguistics, pages 21–34. Routledge. ↗
  11. Robbie Love, Claire Dembry, Andrew Hardie, Vaclav Brezina, and Tony McEnery. 2017. The Spoken BNC2014: Designing and building a spoken corpus of everyday conversations. International Journal of Corpus Linguistics, 22(3):319–344. find
  12. Carlos Mena, Albert Gatt, Andrea DeMarco, Claudia Borg, Lonneke van der Plas, Amanda Muscat, and Ian Padovani. 2020. MASRI-HEADSET: A Maltese Corpus for Speech Recognition. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 6381–6388, Marseille, France. European Language Resources Association. ↗
  13. Kurt Micallef, Albert Gatt, Marc Tanti, Lonneke van der Plas, and Claudia Borg. 2022. Pre-training data quality and quantity for a low-resource language: New corpus and BERT models for Maltese. In Proceedings of the Third Workshop on Deep Learning for Low-Resource Natural Language Processing, pages 90–101, Hybrid. Association for Computational Linguistics.
  14. Oliver Niebuhr and Alexis Michaud. 2015. Speech data acquisition: the underestimated challenge. KALIPHO - Kieler Arbeiten zur Linguistik und Phonetik, 3:1–42. find
  15. Patrizia Paggio and Alexandra Vella. 2014. Overlaps in Maltese conversational and task-oriented dialogues. In Proceedings of the First European Symposium on Multimodal Communication. Linköping Electronic Conference Proceedings. find
  16. Michael Rosner and Claudia Borg. 2023. Language Report Maltese. In: Georg Rehm and Andy Way, editors, European Language Equality: A Strategic Agenda for Digital Language Equality. Springer International Publishing, Cham. ↗
  17. Michael Rosner and Jan Joachimsen. 2012. Il-Lingwa Maltija Fl-Era Diġitali – The Maltese Language in the Digital Age. META-NET White Paper Series. Georg Rehm and Hans Uszkoreit (Series Editors). Springer. Available online at http://www.meta-net.eu/whitepapers. ↗
  18. John Sinclair. 2005. Corpus and Text: Basic prinici-ples, AHDS guides to good practice. Oxbow Books. find
  19. Ganesh Sinisetty, Pavlo Ruban, Oleksandr Dymov, and Mirco Ravanelli. 2021. Commonlanguage. find
  20. Marcin Skowron, Gerhard Backfried, Eva Navas, Aivars Bērzin , š, Joachim Van den Bogaert, Franciska de Jong, Andrea DeMarco, Inma Hernáez, Marek Kováč, Peter Polák, Johan Rohdin, Michael Rosner, Jon Sanchez, Ibon Saratxaga, and Petr Schwarz. 2023. Deep Dive Speech Technology, pages 289–312. Springer International Publishing, Cham. find
  21. Artūrs Vasil , evskis, Jānis Ziedin , š, Marko Tadić, Željka Motika, Mark Fishel, Hrafn Loftsson, Jón Gu, Claudia Borg, Keith Cortis, Judie Attard, and Donatienne Spiteri. 2022. National language technology platform (NLTP): overall view. In Proceedings of the 23rd Annual Conference of the European Association for Machine Translation, pages 345–346, Ghent, Belgium. European Association for Machine Translation. find
  22. Alexandra Vella. 2013. Languages and language varieties in Malta. International Journal of Bilingual Education and Bilingualism, 16(5):532–552. ↗
  23. Alexandra Vella, Flavia Chetcuti, Sarah Grech, and Michael Spagnol. 2010. Integrating annotated spoken Maltese data into corpora of written Maltese. In Proceedings of the Workshop on Language Resources and Human Language Technologies for Semitic Languages, pages 83–90. find
  24. Seventh Conference on International Language Resources and Evaluation (LREC 2010). find
  25. Alexandra Vella and Paulseph-John Farrugia. 2006. MalToBI - Building an annotated corpus of spoken Maltese. In Speech Prosody, Dresden. find
  26. Alexandra Vella and Sarah Grech. 2022. What can a corpus tell us about phonetic and phonological variation? In Anne O’Keeffe and Michael J. McCarthy, editors, The Routledge Handbook of Corpus Linguistics, pages 281–295. Routledge. ↗
  27. Alexandra Vella, Sarah Grech, Ian Padovani, and Maria-Christina Micallef. 2024. Resources and tools for pre-processing speech data in a lesser-known variety of English. In Proceedings of the Twentieth International Congress of Phonetic Sciences. Guarant International. find
  28. Alexandra Vella, Elgar Paul Magro, and Flavia Chetcuti. 2015. Cohesion phenomena in Maltese parliamentary debates. In Proceedings of the Fifth International Conference on Maltese Linguistics. find
  29. Aiden Williams, Andrea DeMarco, and Claudia Borg. 2023. The applicability of Wav2Vec2 and Whisper for low-resource Maltese ASR. In Second Annual Meeting of the Special Interest Group on Under-resourced Languages, a satellite Workshop of Interspeech 2023. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/vella-2024-corpus-spoken-maltese-korpus-tal/