Elamur Research⌕
☰
home / papers / 2026 / To make someone do something: mining alert-style directives in Bulgarian social media for low-resource language modelling

To make someone do something: mining alert-style directives in Bulgarian social media for low-resource language modelling

Paper · 2026 · Ruslana Margova and Stanislav Penkov

Summary

Persuasive language that signals urgency, threat or hidden danger can amplify disinformation and propaganda narratives. Detecting it automatically is still poorly studied for low-resource languages. This paper addresses Bulgarian, a morphologically rich language. Drawing on Searle's notion of directive speech acts, it asks which constructions distinguish highly reactive social media posts from less reactive ones.

The corpus has 13,989 public Facebook posts collected via CrowdTangle (2014–2024), all mentioning "пропаганда". Each post gets a linguistic weight: total engagement multiplied by reaction entropy. From the most heavily weighted posts, the authors mine n-grams and skip-grams without labels or predefined lexicons. They score these for alert enrichment and cluster them with K-means into rhetorical families. Posts are then enriched with alert features. These features feed an interpretable logistic regression model, a TF–IDF + Linear SVM baseline and an LDA topic–alert analysis.

The main finding is that rhetorical salience, and not only topic, is associated with engagement. The work contributes reusable resources for building and evaluating language models for content moderation and propaganda-alert detection in Bulgarian.

Key points

  • Data: 13,989 Bulgarian Facebook posts (file propaganda_bg_media.json) from 2014–2024. Raw data are not shared, in line with Meta policy.
  • Mining used the top 20% of posts by linguistic weight. It extracted 2–5 token n-grams and skip-grams, with POS tagging to discard purely functional patterns. Each construction received an enrichment score and a z-normalised alert_z.
  • The 18,837 multiword constructions were clustered with K-means (k = 20). Nine clusters showed elevated alert_z, grouping into five families: modality (e.g., "трябва да"), negation, prediction, identity framing and accusatory tone.
  • Released resources include about 18,000 ranked constructions and a post-level feature matrix. The matrix combines alert counts, scores, cluster coverage and engagement signals.
  • Limitations:
    • Selection with a single keyword biases the corpus.
    • The proxy labels are circular, since they are built from the same constructions the model uses.
    • There is no human annotation.
    • Temporal drift across the decade is not modelled.

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 worksSocial Sciences · 1,772 worksSociology and Political Science · 577 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksLow-resource languages · 267 works

Cite

Cite as Ruslana Margova and Stanislav Penkov. 2026. To make someone do something: mining alert-style directives in Bulgarian social media for low-resource language modelling. Proceedings of the Second Workshop on Language Models for Low-Resource Languages (LoResLM 2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.loreslm-1.6

@inproceedings{margova2026make,
  title = {{To make someone do something: mining alert-style directives in Bulgarian social media for low-resource language modelling}},
  author = {Ruslana Margova and Stanislav Penkov},
  booktitle = {Proceedings of the Second Workshop on Language Models for Low-Resource Languages (LoResLM 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.loreslm-1.6},
  url = {https://doi.org/10.18653/v1/2026.loreslm-1.6}
}

References

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

  1. Stavros Assimakopoulos. 2020. Incitement to discriminatory hatred, illocution and perlocution. Pragmatics and Society, 11(2):177–195. find
  2. John Langshaw Austin. 1975. How to do things with words. Harvard university press. find
  3. Alberto Barrón-Cedeño, Tamer Elsayed, Preslav Nakov, Giovanni Da San Martino, Maram Hasanain, Reem Suwaileh, Fatima Haouari, Nikolay Babulkov, Bayan Hamdan, Alex Nikolov, Shaden Shaar, and Zien Sheikh Ali. 2020. Overview of checkthat! 2020: Automatic identification and verification of claims in so… — [arXiv:2007.07997](https://arxiv.org/abs/2007.07997) find
  4. Roni Danziger. 2025. Interpretive constructs: The case of incitement to violence and terror. Journal of Pragmatics, 245:35–49. find
  5. Claes H De Vreese, Frank Esser, Toril Aalberg, Carsten Reinemann, and James Stanyer. 2018. Populism as an expression of political communication content and style: A new perspective. The international journal of press/politics, 23(4):423– 438. find
  6. Robert M. Entman. 1993. Framing: Toward clarification of a fractured paradigm. Journal of Communication, 43(4):51–58. find
  7. Norman Fairclough. 2003. Analysing Discourse: Textual Analysis for Social Research. Routledge, London. find
  8. Norman Fairclough. 2013. Language and power. Routledge. find
  9. Vaibhav Garg, Ganning Xu, and Munindar P Singh. 2025. Understanding inciting speech as new malice. IEEE Transactions on Computational Social Systems. \#\# Acknowledgments Muhammad Shahid Khan. 2025. Analysis of Violence Incitation in Social Media from Urdu Content using NLP Techniques. Ph.D. thesis, C… find
  10. This research is funded by GATE project: Horizon 2020 WIDESPREAD-2018-2020, TEAM-ING Phase 2 Program under grant agreement no. 857155, the Program "Research, Innovation, and Digitalization for Smart Transformation" 2021-2027 (PRIDST), under grant agreement no. BG16RFPR002-1.014-0010-C01, and Johanne… find
  11. Preslav Nakov, David Corney, Maram Hasanain, Firoj Alam, Tamer Elsayed, Alberto Barrón-Cedeño, Paolo Papotti, Shaden Shaar, and Giovanni Da San Martino. 2021. Automated fact-checking for assisting human fact-checkers. In Proceedings of the 30th International Joint Conference on Artificial Intellig… — [arXiv:2103.07769](https://arxiv.org/abs/2103.07769) find
  12. Loukas Konstantinou and Evangelos Karapanos. 2025. Behavior change interventions combating online misinformation: A scoping review. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, pages 1–19. find
  13. Ramona Kreis. 2017. The “tweet politics” of president trump. Journal of Language and Politics, 16(4):607–618. find
  14. Peng Qi, Yuhao Zhang, Yuhui Zhang, Jason Bolton, and Christopher D Manning. 2020. Stanza: A python natural language processing toolkit for many human languages. arXiv preprint arXiv:2003.07082.
  15. João A Leite, Olesya Razuvayevskaya, Carolina Scarton, and Kalina Bontcheva. 2025. A cross-domain study of the use of persuasion techniques in online disinformation. In Companion Proceedings of the ACM on Web Conference 2025, pages 1100–1103. find
  16. Hannah Rashkin, Eunsol Choi, Jin Yea Jang, Svitlana Volkova, and Yejin Choi. 2017. Truth of varying shades: Analyzing language in fake news and political fact-checking. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2931–2937, Copenhagen, Den…
  17. Ekaterina Loginova. 2025. Fine-tuned transformers for detection and classification of persuasion techniques in slavic languages. In Proceedings of the 10th Workshop on Slavic Natural Language Processing (Slavic NLP 2025), pages 151–156. find
  18. John R Searle. 1975. A taxonomy of illocutionary acts. find
  19. Stuart Macdonald and Nuria Lorenzo-Dus. 2020. Intentional and performative persuasion: The linguistic basis for criminalizing the (direct and indirect) encouragement of terrorism. In Criminal Law Forum, volume 31, pages 473–512. Springer. find
  20. Kiril Simov, Petya Osenova, and 1 others. 2005. Bultreebank: Building a bulgarian treebank. In Proceedings of a Treebank-related workshop. find
  21. Kate Starbird and Thomas Wilson. 2020. Cross-platform disinformation campaigns: Lessons learned and next steps. Harvard Kennedy School (HKS) Misinformation Review. find
  22. Ruslana Margova. 2023. Linguistic Markers of Fake News. Ph.D. thesis, Sofia University St. Kliment Ohridski. Doctoral dissertation. Open access. find
  23. JoAnne Sweeny. 2019. Incitement in the era of trump and charlottesville. Cap. UL Rev., 47:585. find
  24. Giovanni Da San Martino, Alberto Barrón-Cedeño, Henning Wachsmuth, Rostislav Petrov, and Preslav Nakov. 2020. SemEval-2020 task 11: Detection of propaganda techniques in news articles. In Proceedings of the Fourteenth Workshop on Semantic Evaluation (SemEval 2020), pages 1377–1414, Barcelona (onli…
  25. Irina Temnikova, Silvia Gargova, Ruslana Margova, Veneta Kireva, Ivo Dzhumerov, Tsvetelina Stefanova, and Hristiana Krasteva. 2023. New bulgarian resources for studying deception and detecting disinformation. find
  26. Rakhmonova Amira Ulfatovna. 2025. Linguistic characteristics of political discourse. FRON-TIERS OF KNOWLEDGE AND INTERDIS-CIPLINARY DISCOVERY, 1(1):366–374. find
  27. Josiemer Mattei, Katherine L Tucker, Luis M Falcón, Carlos F Rı́os-Bedoya, Robert M Kaplan, H June O’Neill, Martha Tamez, Sigrid Mendoza, Claudia B Dı́az-Álvarez, Jonathan E Orozco, and 1 others. 2021. Design and implementation of the puerto rico observational study of psychosocial, environmental,… find
  28. Hariram Veeramani, Surendrabikram Thapa, and Usman Naseem. 2023. Lowresourcenlu at blp-2023 task 1 & 2: Enhancing sentiment classification and violence incitement detection in bangla through aggregated language models. In BLP 2023-1st Workshop on Bangla Language Processing, Proceedings of the Worksh… find
  29. Shahan Ali Memon and Kathleen M. Carley. 2020. Characterizing COVID-19 misinformation communities using a novel twitter dataset. In Proceedings of the 5th International Workshop on Mining Actionable Insights from Social Networks (MAISoN 2020), co-located with CIKM, pages 1–9, Online. ArXiv:2008.0079… — [arXiv:2008.00791](https://arxiv.org/abs/2008.00791) find
  30. Claire Wardle. 2024. A conceptual analysis of the overlaps and differences between hate speech, misinformation and disinformation. Department of Peace Operations (DPO). Office of the Special Adviser on the Prevention of Genocide (OSAPG). United Nations. find
  31. B Ruth Wodak. 2015. The Politics of Fear: What Right-Wing Populist Discourses Mean. SAGE Publications, London, UK. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/margova-2026-make-someone-do-something-mining/