Elamur Research⌕
☰
home / papers / 2026 / Does Machine "know" interpersonal pragmatics? Evidence from MARBERT's learning of emoji pragmatics in Arabic digital discourse

Does Machine "know" interpersonal pragmatics? Evidence from MARBERT's learning of emoji pragmatics in Arabic digital discourse

Summary

The paper asks whether a pretrained Transformer can recognise the relational work that emojis do in Arabic social media, beyond sentiment or emotion. The author collected 8504 unique Facebook posts containing emojis using Python scripts. Each post was manually annotated for five interpersonal pragmatic functions (IPFs): Politeness, Respect, Solidarity, Empathy and Encouragement. The categories are grounded in speech act theory, Brown & Levinson's politeness theory and Spencer-Oatey's rapport management theory. MARBERT was fine-tuned as a multi-label classifier, and the results were read both statistically and through those pragmatic theories.

On held-out data the model reached 93% accuracy, with micro F1 of 0.61 and macro F1 of 0.56. Politeness and Respect were recognised better than Solidarity. The author explains this as a gradient of explicitness: conventionalised facework is easier to learn than identity-based solidarity, which depends on shared cultural memory.

The work matters because it links interpersonal pragmatics with NLP. It also shows that some social meanings are only weakly recoverable from text alone.

Key points

  • Data: a corpus of 8504 unique emoji-bearing Facebook posts in Arabic digital discourse (ADD), manually labelled for 5 IPFs in a multi-label setup.
  • Model: MARBERT, an Arabic Transformer, fine-tuned for the task. No other model was evaluated, which the author lists as a limitation.
  • Overall scores on unseen data: accuracy 93%, micro F1 0.61, macro F1 0.56.
  • Per-function results:
    • Politeness and Respect, linked to "quality face" and stable conventional patterns, scored best.
    • Solidarity, linked to "identity face", scored worst.
    • Solidarity was also the rarest label, so the label distribution was imbalanced.
  • Stated limitations:
    • The corpus covers only one platform (Facebook).
    • The input is limited to text plus emoji, with no conversation history, speaker relations or images.
    • Only a single model was tested.

Interesting ideas

  • I1. Rather than inventing ad-hoc tags, the label set is derived from established pragmatic theories (politeness, rapport management, speech acts), so every category has a theoretical definition that can be checked.
  • I2. Labels are multi-label, not one-per-item, because a single marker or post can perform several relational functions at once.
  • I3. The same symbol (e.g., folded hands, red heart) can signal gratitude, apology, respect or solidarity depending on context, so the function is inferred rather than looked up.
  • I4. Model performance is interpreted as following a gradient of how explicit or conventionalised a function is, which turns per-class error into a linguistic diagnostic.
  • I5. The large gap between 93% accuracy and roughly 0.6 F1 shows how imbalanced multi-label data can make a headline metric look strong.

For Mova

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

  • I1 (frames): prag could add a small, theory-defined layer of relational functions (polite, respectful, solidary, empathic, encouraging) on top of speech acts, each with explicit criteria so the output stays explainable.
  • I2 (frames): the speech-act and implicature output in prag should allow several co-occurring functions per utterance instead of forcing one winner, with a separate justification for each.
  • I3 (frames): politeness markers and discourse particles should be resolved from the surrounding frame and context rather than via a fixed lexicon mapping.
  • I4 (learn): evaluation sets could be grouped by explicitness (conventional cue vs. inference-heavy), so that regressions can be traced to the kind of reasoning that failed.
  • I5 (learn): pragmatic classifiers should always be reported with micro, macro and per-label F1, plus majority-class baselines as negative controls, rather than accuracy alone.

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 Discourse analysis Semitic languages

 Institutions

🎓Ibb UniversityIbb · Yemen4 works · 6 citations

Cite

Cite as Mohammed Q. Shormani. 2026. Does Machine "know" interpersonal pragmatics? Evidence from MARBERT's learning of emoji pragmatics in Arabic digital discourse. arXiv:2608.01174.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2608.01174 · arXiv 2608.01174

@misc{shormani2026does,
  title = {{Does Machine "know" interpersonal pragmatics? Evidence from MARBERT's learning of emoji pragmatics in Arabic digital discourse}},
  author = {Mohammed Q. Shormani},
  year = {2026},
  doi = {10.48550/arxiv.2608.01174},
  eprint = {2608.01174},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2608.01174}
}

References

46 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abdul-Mageed, M., Elmadany, M., Nagoudi, E. M. B. (2022). CAMeLBERT: A Collection of Pre-trained Models for Arabic NLP. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics (pp. 628–640). find
  2. Abdul-Mageed, M. & Elmadany, A. (2021). ARBERT & MARBERT: Deep bidirectional transformers for Arabic. In Proceedings of the 59th annual meeting of the association for computational linguistics and the 11th international joint conference on natural language processing (volume 1: long papers) (pp. 7088-7105). our card
  3. Abu-Nimer, M., Khoury, A., & Welty, E. (2007). Unity in diversity: Interfaith dialogue in the Middle East. US Institute of Peace Press. find
  4. Acheampong, F. A., Nunoo-Mensah, H., & Chen, W. (2021). Transformer models for text-based emotion detection: a review of BERT-based approaches. Artificial Intelligence Review, 54(8), 5789-5829. http://dx.doi.org/10.1007/s10462-021-09958-2 ↗
  5. Ahamad, R., Mishra, K. N. (2025). Exploring sentiment analysis in handwritten and E-text documents using advanced machine learning techniques: a novel approach. Journal of Big Data 12(1) 11. http://dx.doi.org/10.1186/s40537-025-01064-2 ↗
  6. Albluwi, A. (2026). A Novel emoji-aware computational framework for body-shaming detection in gulf Arabic social media discourse. Unpublished PhD Diss., The Catholic University of America. find
  7. Austin, J. L. (1962). How to Do Things with Words. Oxford University Press. find
  8. Bou-Franch, P. & Blitvich, P. G. C. (eds.). (2018). Analyzing digital discourse: New insights and future directions. Springer. find
  9. Bou-Franch, P. (2020). Pragmatics and digital discourse in Spanish research. In Dale A. K., & Felix-Brasdefer, C.J. (eds.), The Routledge Handbook of Spanish Pragmatics (pp. 533–547). Routledge. find
  10. Brown, P., & Levinson, S. C. (1987). Politeness: Some Universals in Language Usage. Cambridge University Press. our card
  11. Caspi, A., & Raz, G. (2025). Using emojis that alter the meaning of written messages to communicate interpersonal relations. Media Psychology, 28(3), 385–411. https://doi.org/10.1080/15213269.2024.2374778 ↗
  12. Cavalheiro, B. P., Prada, M., & Rodrigues, D. L. (2024). Examining the effects of reciprocal emoji use on interpersonal and communication outcomes. Journal of Social and Personal Relationships, 41(8), 2147-2168. https://doi.org/10.1177/02654075231219032 ↗
  13. Conlan, C. J. (2005). Face threatening acts, primary face threatening acts, and the management of discourse. Broadening the horizon of linguistic politeness, Pragmatics & Beyond New Series (pp. 65Á83). Amsterdam, Netherlands: John Benjamins. http://dx.doi.org/10.1075/pbns.139.13con Conneau, A., Khandelwal, K., Goyal, N., Chaudhary, V., Wenzek, G., Guzmán, F., ... & Stoyanov, V. (2020, July). Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th annual meeting of the association for computational linguistics (pp. 8440-8451). http://dx.doi.org/10.18653/v1/2020.acl-main.747 ↗
  14. Culpeper, J. (2011). Impoliteness: Using language to cause offence (Vol. 28). Cambridge University Press.http://dx.doi.org/10.1017/CBO9780511975752 Danesi, M. (2017). The semiotics of emoji: The rise of visual language in the age of the internet. Bloomsbury Publishing. ↗
  15. Devlin, J., Chang, M. W., Lee, K., & Toutanova, K. (2019). Bert: Pretraining of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers) (pp. 4171-4186). find
  16. Ge, J., & Herring, S. C. (2018). Communicative functions of emoji sequences on Sina Weibo. First Monday, 23(11). https://doi.org/10.5210/fm.v23i11.9413 ↗
  17. Gibson, W., Huang, P., & Yu, Q. (2018). Emoji and communicative action: The semiotics, sequence and gestural actions of ‘face covering hand’. Discourse, context & media, 26, 91-99. http://dx.doi.org/10.1016/j.dcm.2018.05.005 ↗
  18. Goffman, E. (1955). On face-work: An analysis of ritual elements in social interaction. Psychiatry, 18(3), 213-231. https://doi.org/10.1080/00332747.1955.11023008 ↗
  19. Hancock, P. M., Hilverman, C., Cook, S. W., & Halvorson, K. M. (2024). Emoji as gesture in digital communication: Emoji improve comprehension of indirect speech. Psychonomic bulletin & review, 31(3), 1335-1347. http://dx.doi.org/10.3758/s13423-023-02411-1 ↗
  20. Hand, C. J., Burd, K., Oliver, A., & Robus, C. M. (2022). Interactions between text content and emoji types determine perceptions of both messages and senders. Computers in human behavior reports, 8, 100242. https://doi.org/10.1016/j.chbr.2022.100242 ↗
  21. Herring, S. C., & Ge-Stadnyk, A. (2024). Emoji and Illocutionarity. In Gill, M, Malmivirta, A. & Wårvik, B. (eds), Structures in discourse: Interaction, adaptability, and pragmatic functions, (pp. 124-155). John Benjamins. http://dx.doi.org/10.1075/pbns.345.07her ↗
  22. Hurley, Z. (2019). Imagined affordances of Instagram and the fantastical authenticity of female Gulf-Arab social media influencers. Social Media+ Society, 5(1), 2056305118819241. http://dx.doi.org/10.1177/2056305118819241 ↗
  23. Kralj Novak, P., Smailović, J., Sluban, B., & Mozetič, I. (2015). Sentiment of emojis. PloS one, 10(12), e0144296. http://dx.doi.org/10.1371/journal.pone.0144296 ↗
  24. Leech, G. N. (2014). The pragmatics of politeness. Oxford University Press. find
  25. Locher, M. A., & Watts, R. J. (2005). Politeness theory and relational work. Journal of Politeness Research: Language, Behavior, Culture, 1(1). http://dx.doi.org/10.1515/jplr.2005.1.1.9 ↗
  26. Logi, L., & Zappavigna, M. (2023). A social semiotic perspective on emoji: How emoji and language interact to make meaning in digital messages. New Media & Society, 25(12), 3222-3246. http://dx.doi.org/10.1177/14614448211032965 ↗
  27. Morsi, W. K. (2026). The pragmatics of emoji: A multimodal investigation of politeness and sarcasm in student interactions on Facebook, twitter, and reddit. Social Sciences & Humanities Open, 13, 102938. http://dx.doi.org/10.1016/j.ssaho.2026.102938 ↗
  28. Moschini, I., & Sindoni, M. G. (Eds.). (2021). Mediation and multimodal meaning making in digital environments. Routledge. http://dx.doi.org/10.4324/9781003225423 ↗
  29. O’Driscoll, J. (2011). Some Issues with the Concept of Face: When, what, how and how much? In Politeness across cultures (pp. 17-41). London: Palgrave Macmillan UK.http://dx.doi.org/10.1057/9780230305939_2 O’Halloran, K., Tan, S., Smith, B., & Podlasov, A. (2010). Challenges in designing digital interfaces for the study of multimodal phenomena. Information design journal, 18(1), 2-21.http://dx.doi.org/10.1075/idj.18.1.02hal Salve, H. N., & Tubil, P. N. T. (2025). Bilingual Code-switching using XLM-RoBERTa. TechRxiv. 11. https://doi.org/10.36227/techrxiv.175756344.42614762/v1 ↗
  30. Sasamoto, R. (2023). Perceptual resemblance and the communication of emotion in digital contexts: A case of emoji and reaction GIFs. Pragmatics, 33(3), 393-417. http://dx.doi.org/10.1075/prag.21058.sas ↗
  31. Satar, M., Hauck, M., & Bilki, Z. (2023). Multimodal representation in virtual exchange: A social semiotic approach to critical digital literacy. Language Learning & Technology, 27(2), 72-96. http://dx.doi.org/10.64152/10125/73504 ↗
  32. Searle, J. R. (1969). Speech Acts: An Essay in the Philosophy of Language. Cambridge University Press. http://dx.doi.org/10.1017/CBO9781139173438 ↗ our card
  33. Shormani, M. Q. (2025). What fifty-one years of linguistics and artificial intelligence research tell us about their correlation: A scientometric analysis. Artificial Intelligence Review,58, 379 (2025). https://doi.org/10.1007/s10462-025-11332-5 ↗
  34. Shormani, M. Q., & Alenezi, M. A. N. (2026). Arabic Nominalization as Form of Pragmatic Actions in Digital Discourse: A Corpus-Based Study. Corpus Pragmatics, 10, 35. https://doi.org/10.1007/s41701-026-00237-5 ↗
  35. Shormani, M. Q., Hassan, I. A., & Alshawsh, M. Y. (2026). Machine learning and emoji prediction: How much accuracy can MARBERT achieve?. arXiv preprint arXiv:2604.21108. https://doi.org/10.48550/arXiv.2604.21108 ↗ our card
  36. Spencer-Oatey, H. (2000). Rapport Management: A Framework for Analysis. In Culturally Speaking: Managing Rapport Through Talk across Cultures (pp. 11-46). London and New York, NY: Continuum. find
  37. Spencer-Oatey, H. (2008). Culturally speaking: Culture, communication and politeness theory. (2 nd ed.) Bloomsbury Publishing. find
  38. Spencer-Oatey, H., & Wang, J. (2026). Communication strategies for building and maintaining collaborative relations: A Sino-American Case Study of Rapport and ‘Politeness’. Studies of Applied Linguistics in Asia, 1(1), 18–34. https://doi.org/10.53941/sala.2026.100003 ↗
  39. Tieu, L., Qiu, J. L., Puvipalan, V., & Pasternak, R. (2025). Experimental evidence for a semantic typology of emoji: Inferences of co-, pro-, and post-text emoji. Quarterly journal of experimental psychology (2006), 78(4), 808–826. https://doi.org/10.1177/17470218241255786 van der Vlist, F. N., Helmond, A., Burkhardt, M., & Seitz, T. )2022(. API governance: The case of Facebook’s evolution. Social Media + Society, 8(2), 20563051221086228. http://dx.doi.org/10.1177/20563051221086228 ↗
  40. Vásquez, C. (ed.). (2022). Research methods for digital discourse analysis. Bloomsbury Publishing. find
  41. Weissman, B. (2022). Emoji semantics/pragmatics: Investigating commitment and lying. In Proceedings of the fifth international workshop on emoji understanding and applications in social media (pp. 21-28). http://dx.doi.org/10.18653/v1/2022.emoji-1.3 ↗
  42. Yule, G. (2022). The study of language. Cambridge University Press. find
  43. Yus, F. (2011). Cyberpragmatics: Internet-mediated communication in context. John Benjamins Publishing Company. find
  44. Yus, F. (2014). Not all emoticons are created equal. Linguagem em (Dis)Curso, 14(3), 511–529. http://dx.doi.org/10.1590/1982-4017-140304-0414 ↗
  45. Yus, F. (2025). Emoji pragmatics. Springer Nature. http://dx.doi.org/10.1007/978-3-031-73723-7 ↗
  46. Zhao, P., Jia, J., An, Y., Liang, J., Xie, L. & Luo, J. (2018). Analyzing and predicting emoji usages in social media. In Companion Proceedings of the Web Conference 2018 (pp. 327-334). Availability of data and materials The training, validation, and evaluation datasets generated or analyzed during the current study will be available on GitHub at [https://github.com/..../ marbert_emoji_modeling...] Funding This research project did not receive internal or external funding. Conflicts of Interest Authors declare no conflicts of interest. ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shormani-2026-machine-know-interpersonal-pragmatics-evidence/