Elamur Research⌕
☰
home / papers / 2026 / Do Large Language Models Understand Double Mismatches? Evidence from Farsi

Do Large Language Models Understand Double Mismatches? Evidence from Farsi

Paper · 2026 · Maryam Mohammadi

Summary

The paper asks whether an LLM can generate a rare Farsi honorific construction that combines two agreement mismatches. In colloquial Tehrani Farsi, speakers use the plural pronoun šoma for one respected person, which is the first mismatch. They may also keep the verb and possessive or reflexive forms singular, breaking the Polite Plural Generalization, to add closeness to respect. This is the second mismatch. The author first analyses the pattern. Singular marking must apply to all agreement targets together, and it works only with the free pronoun šoma, not the clitic -tun. A pilot then prompted GPT-4 in Farsi with six scenarios involving an older but intimate addressee and four involving parents speaking to small children. GPT-4 never produced the double mismatch. It used full plural agreement with older addressees and plain singular "to" with children. Four native speakers mostly chose the double mismatch in the first setting. The author calls the results preliminary, but they suggest that rare yet pragmatically meaningful forms may vanish from LLM output.

Key points

  • The phenomenon involves two layers: plural šoma for a single honorific referent (honorific), plus singular verb and possessive/reflexive agreement (intimacy). It is infelicitous without real closeness, for example toward a manager.
  • Felicity conditions: mixing a plural possessive clitic with a singular verb is ruled out. The effect requires the free pronoun šoma, and dropping šoma removes the honorific reading.
  • GPT-4 was tested on six contexts with an older addressee, such as grandchild–grandmother and niece–uncle. It always produced the single-mismatch form with uniform plural agreement.
  • In four parent–young-child scenarios, GPT-4 always used singular "to". Human informants also used double mismatches less often in this setting, which the author links to a newer practice among younger parents.
  • Stated limitations: there were few prompts, only one model was tested, and no statistical analysis was done. Only production was studied, not perception. Future work plans more contexts, more models (e.g., ParsBERT) and a larger native-speaker baseline.

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

🎓Bielefeld UniversityBielefeld · Germany35 works · 1,196 citations

 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.Language models & pretraining · 3,320 works

Cite

Cite as Maryam Mohammadi. 2026. Do Large Language Models Understand Double Mismatches? Evidence from Farsi. The Proceedings of the First Workshop on NLP and LLMs for the Iranian Language Family.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.silkroadnlp-1.3

@inproceedings{mohammadi2026do,
  title = {{Do Large Language Models Understand Double Mismatches? Evidence from Farsi}},
  author = {Maryam Mohammadi},
  booktitle = {The Proceedings of the First Workshop on NLP and LLMs for the Iranian Language Family},
  year = {2026},
  doi = {10.18653/v1/2026.silkroadnlp-1.3},
  url = {https://doi.org/10.18653/v1/2026.silkroadnlp-1.3}
}

References

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

  1. Deepak Alok. 2021. The morphosyntax of magahi addressee agreement. Syntax, 24:263–296. find
  2. Penelope Brown and Stephen C. Levinson. 1987. Politeness: Some Universals in Language Usage. Cambridge University Press, Cambridge, UK.
  3. Bernard Comrie. 1975. Polite plurals and predicate agreement. Language, 51(2):406–418. find
  4. Charles A. Ferguson. 1991. Individual and social in language change: Diachronic changes in politeness agreement in forms of address. In Robert L. Cooper and Bernard Spolsky, editors, The Influence of Language on Culture and Thought: Essays in Honor of Joshua A. Fishman’s Sixty-Fifth Birthday, pages… find
  5. Nikta Gohari Sadr, Sahar Heidariasl, Karine Megerdoomian, Laleh Seyyed-Kalantari, and Ali Emami. 2025. We politely insist: Your llm must learn the persian art of taarof. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 1819–1838. Association for Comput… find
  6. Elin McCready. 2019. The Semantics and Pragmatics of Honorification. Oxford University Press, Oxford. find
  7. Kangsan Noh, Sanghoun Song, and Eunjeong Oh. 2024. How language models understand honorific mismatches in korean. Language Research, 60(3):303– 322. find
  8. Paul Portner, Miok Pak, and Raffaella Zanuttini. 2019. The speaker–addressee relation at the syntax– semantics interface. Language, 95(1):1–36. find
  9. Christopher Potts. 2007. The expressive dimension. Theoretical Linguistics, 33(2):165–198. find
  10. Zrinka Puškar-Gallien. 2019. Resolving polite conflicts in predicate agreement. Glossa: A Journal of General Linguistics, 4(1):33. find
  11. Mohammad Rasekh Mahand. 2014. Persian clitics: Doubling and agreement. Modern Language Journal, 24:16–33. find
  12. Stephen Wechsler. 2011. Mixed agreement, the person feature, and the index/concord distinction. Natural Language & Linguistic Theory, 29(3):999–1031. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/mohammadi-2026-do-large-language-models-understand/