Elamur Research⌕
☰
home / papers / 2026 / TestiMole-Conversational: A 30-Billion-Word Italian Discussion Board Corpus (1996–2024) for Language Modeling and Sociolinguistic Research

TestiMole-Conversational: A 30-Billion-Word Italian Discussion Board Corpus (1996–2024) for Language Modeling and Sociolinguistic Research

Paper · 2026 · Matteo Rinaldi, Rossella Varvara and Viviana Patti

Summary

Compared with English, Italian has very little web-scale text. This limits the development of native Italian LLMs. There has also been no large-scale study of the informal, dialogic language used on Italian discussion boards. Researchers at the University of Turin built TestiMole-Conversational to address both gaps. It is an archive of Italian online discussions collected from Usenet newsgroups and many independently run web forums, covering 1996 to 2024. The team scraped and cleaned the material, which raised challenges such as extracting text reliably and anonymizing users. Each post is stored as its own JSONL record with its exact writing time and its position in the thread. The resulting resource contains about 30 billion word tokens and is presented as the largest Italian collection of forum and Usenet text. The authors see three uses for it. It offers natively Italian conversational text for LLM pre-training, including help-seeking exchanges and moderation behaviour. It preserves forum content that is disappearing as sites go offline. Its timestamps also support diachronic and sociolinguistic research on Italian computer-mediated communication.

Key points

  • Size: roughly 30B word tokens, from about 470 million forum messages and 90 million Usenet messages, spanning 1996–2024.
  • The dataset is released as JSONL files on Hugging Face (mrinaldi/TestiMole). Each row is one post, and thread order is kept in a progressive_number field.
  • It is far larger than the previous biggest Italian Usenet resource, the NUNC corpora (127M tokens, University of Turin).
  • The authors motivate the work with data scarcity: Italian makes up only about 2% of Common Crawl, against roughly 44% for English. The main earlier Italian social-media source is TWITA, which was used to train AlBERTo.
  • Intended research uses include language modelling, domain adaptation, conversational analysis, and studies of diachronic lexical and orthographic change, such as SMS-style abbreviations and emoticon styles. Sociological uses include research on trolling and on communities like "Forum dei brutti".

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of TurinTurin · Italy33 works · 6,637 citations

 Fields, topics and concepts

📚 Shelves 5The 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 worksHuman-Computer Interaction · 160 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,894 worksDialogue systems · 654 works

Cite

Cite as Matteo Rinaldi, Rossella Varvara and Viviana Patti. 2026. TestiMole-Conversational: A 30-Billion-Word Italian Discussion Board Corpus (1996–2024) for Language Modeling and Sociolinguistic Research. Proceedings of the 12th Workshop on Challenges in the Management of Large Corpora.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/5643evropidu

@inproceedings{rinaldi2026testimole,
  title = {{TestiMole-Conversational: A 30-Billion-Word Italian Discussion Board Corpus (1996–2024) for Language Modeling and Sociolinguistic Research}},
  author = {Matteo Rinaldi, Rossella Varvara and Viviana Patti},
  booktitle = {Proceedings of the 12th Workshop on Challenges in the Management of Large Corpora},
  year = {2026},
  doi = {10.63317/5643evropidu},
  url = {https://doi.org/10.63317/5643evropidu}
}

References

22 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Paolo Gajo, Silvia Bernardini, Adriano Ferraresi, and Alberto Barrón-Cedeño. 2023. Hate speech detection in an Italian incel forum using bilingual data for pre-training and fine-tuning. In Proceedings of the Ninth Italian Conference on Computational Linguistics (CLiC-it 2023), pages 211–218, Venice,… find
  2. Giuseppe Antonelli. 2016. L’e-taliano tra storia e leggende. In L’e-taliano. Scriventi e scritture nell’era digitale, pages 11–28. Franco Cesati Editore. find
  3. Emanuele Ferdinando Barbera. 2013. Una introduzione ai NUNC: storia della creazione di un corpus, volume Molti occhi sono meglio di uno: saggi di linguistica generale 2008-12, pages 97– 114. Qu. A. S. A. R. find
  4. Sara Gemelli and Gosse Minnema. 2024. Manosphrames: exploring an Italian incel community through the lens of NLP and frame semantics. In Proceedings of the First Workshop on Reference, Framing, and Perspective @ LREC-COLING 2024, pages 28–39, Torino, Italia. ELRA and ICCL. find
  5. Manuel Barbera and Carla Marello. 2011. Tra scritto-parlato, Umgangssprache e comunicazione in rete: i corpora NUNC. In «Studi di Grammatica Italiana» XXVII (2008, recte 2011) = Per Giovanni Nencioni. Convegno Internazionale di Studi. Pisa - Firenze, 4-5 Maggio 2009, pages 157– 185. Le Lettere. find
  6. Claire Hardaker. 2010. Trolling in asynchronous computer-mediated communication: From user discussions to academic definitions. Journal of Politeness Research, 6(2):215–242. find
  7. Sebastian Hoffmann. 2007. Processing internet-derived text—creating a corpus of usenet messages. Literary and Linguistic Computing, 22(2):151–165. find
  8. Marco Baroni, Silvia Bernardini, Adriano Ferraresi, and Eros Zanchetta. 2009. The wacky wide web: a collection of very large linguistically processed web-crawled corpora. Language resources and evaluation, 43(3):209–226. find
  9. Miloš Jakubíček, Adam Kilgarriff, Vojtěch Kovář, Pavel Rychlỳ, and Vít Suchomel. 2013. The tenten corpus family. In 7th international corpus linguistics conference CL, volume 2013, pages 125–127. Valladolid. find
  10. Valerio Basile, Mirko Lai, and Manuela Sanguinetti. 2018. Long-term social media data collection at the university of turin. In Proceedings of the Fifth Italian Conference on Computational Linguistics (CLiC-it 2018), pages 41–46, Turin, Italy. CEUR Workshop Proceedings. find
  11. Kevin Lund and Curt Burgess. 1996. Producing high-dimensional semantic spaces from lexical co-occurrence. Behavior Research Methods, Instruments, & Computers, 28(2):203–208. find
  12. Vladimír Benko. 2014. Aranea: Yet another family of (comparable) web corpora. In International Conference on Text, Speech, and Dialogue, pages 247–256. Springer. find
  13. Matt Mahoney. 2000. Usenet as a text corpus. Technical report, Florida Tech, CS Dept. Available online at: https://cs.fit.edu/. ↗
  14. Curt Burgess and Kay Livesay. 1998. The effect of corpus size in predicting reaction time in a basic word recognition task: Moving on from kučera and francis. Behavior Research Methods, Instruments, and Computers, 30:272–277. find
  15. Elena Pistolesi. 2018. L’italiano in rete: Usi, varietà e proposte di analisi. AggiornaMenti, pages 17— -26. find
  16. Marco Polignano, Valerio Basile, Pierpaolo Basile, Marco de Gemmis, and Giovanni Semeraro. 2019. AlBERTo: Modeling Italian Social Media Language with BERT. IJCoL [Online]. find
  17. Antonia Cava and Fabrizia Pasciuto. 2023. Misoginia online: un’analisi netnografica sul forum dei brutti, pages 469–482. Sette città. find
  18. Roland Schäfer, Felix Bildhauer, et al. 2012. Building large corpora from the web using a new efficient tool chain. In Lrec, pages 486–493.
  19. Patrick Conley, Curt Burgess, and Dotty Hage. 1999. Large-scale databases of proper names. Behavior research methods, instruments, & computers : a journal of the Psychonomic Society, Inc, 31:215–9. find
  20. Jasmin Schröck and Harald Lüngen. 2015. Building and annotating a corpus of german-language newsgroups. In NLP4CMC 2015. 2nd Workshop on Natural Language Processing for Computer-Mediated Communication / Social Media. Proceedings of the Workshop , September 29, 2015 University of Duisburg-Essen, Camp… find
  21. Language Resource References Barbera, Manuel and Colombo, Simona and Marello, Carla. 2011. NUNC - A Multilanguage Suite of Newsgroups Corpora. Università degli Studi di Torino. PID http://www.bmanuel.org/projects/ng-HOME.html. ↗
  22. Shaoul, Cyrus and Westbury, Chris. 2013. A reduced redundancy USENET corpus (2005-2011). Edmonton, AB: University of Alberta. PID http://www.psych.ualberta.ca/ westbury-lab/downloads/usenetcorpus.download.html. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/rinaldi-2026-testimole-conversational-30-billion-word/