Elamur Research⌕
☰
home / papers / 2026 / Rational Communication Shapes Morphological Composition

Rational Communication Shapes Morphological Composition

Paper · 2026 · Fengyuan Yang, Yongqian Peng, Yuxi Ma et al.

Summary

The paper asks why English settled on one particular morpheme combination for a new word when other combinations, built from morphemes that existed at the time, could have named the same concept. The authors treat word formation as a cooperative game in the Rational Speech Act (RSA) framework. A pragmatic speaker scores candidate morpheme sequences by how easily a literal listener could recover the intended concept, minus how costly the form is to produce. The data are 4323 WordNet compounds and derivations whose first appearance, taken from COHA/COCA counts, falls between 1820 and 2019. A time-indexed lexicon ensures that only morphemes available at the coinage year compete. The authors rank each real word against generated alternatives. Models that combine recoverability with cost beat semantics-only and cost-only baselines on MRR and Acc@k. The gain over the semantic baseline widens as the candidate pool grows, because meaning alone stops discriminating. The work brings rational-communication accounts down to the internal structure of words and gives an interpretable model of lexical choice.

Key points

  • Dataset construction:

    • WordNet supplies the items and their glosses.
    • MorphSeg provides gold morpheme segmentation.
    • Emergence year is the first non-zero year in COHA/COCA.
    • The CMU Pronouncing Dictionary supplies phoneme and syllable counts.
    • Affixes.org extends the morpheme inventory.

  • Semantic side: each candidate is described by five statistics of morpheme-to-concept similarity, computed with Qwen3 embeddings:

    • mean, max, standard deviation and entropy of the per-morpheme similarities;
    • the similarity of the whole candidate to the target.

    A kNN gate around the target gloss discards candidates containing out-of-neighbourhood morphemes before ranking.

  • Cost side: a learned function maps each morpheme's frequency, phonological complexity and length to a cost, and these costs are summed over the morphemes. The two scores are combined in two ways:

    • Linear S1, with learnable weights α, β and b;
    • Nonlinear S1, a small MLP.

  • Results:

    • S1 models beat the single-factor baselines, and cost-only is weakest.
    • An unconstrained discriminative model over the full 13-dimensional feature vector performs best overall.
    • The model ordering holds across cumulative training windows ending 1830–1910.

  • Limitations:

    • Candidates are treated as unordered sets, ignoring morphotactics and head–modifier order.
    • Embeddings are contemporary, so historical meaning shift is only indirectly captured.
    • The evaluation covers English only.

Interesting ideas

  • I1. Judge a choice model by where the actually attested form lands among the alternatives available at that moment, using MRR and Acc@k instead of plain accuracy; this works for any task with a single observed choice.
  • I2. Describe a multi-part candidate by summary statistics of how each part relates to the target, keeping part-level evidence visible rather than pooling everything into one vector.
  • I3. Apply a hard neighbourhood gate before scoring, so the expensive ranker only sees semantically plausible candidates and the exclusion reason is explicit.
  • I4. Model a chosen form as listener recoverability traded against speaker cost, compared over an explicit set of alternatives.
  • I5. Pair the interpretable two-factor model with an opaque full-feature model used only as a ceiling, and read the gap between them as a map of missing interactions.

For Mova

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

  • I1 (back): in generation and the text → graph → text compressor, rank the original wording among candidate paraphrases produced from the same graph, to measure lexical-choice quality explainably.
  • I2 (tokens): for unseen compounds and derivations, score competing segmentations by per-morpheme similarity to a candidate gloss, so the lemma and morphology decision can cite which part carried the meaning.
  • I3 (world): when grounding an unknown word into the WordNet/Wikidata ontology, consider only concepts near its context or gloss, and log why others were rejected.
  • I4 (frames): in prag, compute implicatures by asking which cheaper or more informative alternatives the speaker passed over, and record the comparison as the explanation.
  • I5 (learn): train a black-box diagnostic beside rule-based components to locate the feature interactions that explainable repair rules still miss.

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 6The field of science this work belongs to (OpenAlex fields).Arts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksSocial Sciences · 1,772 worksPsychology · 1,552 worksDevelopmental and Educational Psychology · 499 worksCultural Studies · 20 works

🧭 Topics 1Research subjects and languages the work is about.Morphology · 405 works

Cite

Cite as Fengyuan Yang, Yongqian Peng, Yuxi Ma, Chenheng Xu and Yixin Zhu. 2026. Rational Communication Shapes Morphological Composition. arXiv:2605.03510.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2605.03510 · arXiv 2605.03510

@misc{yang2026rational,
  title = {{Rational Communication Shapes Morphological Composition}},
  author = {Fengyuan Yang, Yongqian Peng, Yuxi Ma, Chenheng Xu and Yixin Zhu},
  year = {2026},
  doi = {10.48550/arxiv.2605.03510},
  eprint = {2605.03510},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2605.03510}
}

References

34 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Affixes.org. (2008). Affixes.org: Dictionary of english affixes. (Cit. on p. 3). find
  2. Algeo, J. (1980). Where do all the new words come from? American Speech, 55(4), 264–277 (cit. on p. 1). ↗
  3. Arndt-Lappe, S. (2015). Word-formation and analogy. In Word-formation: An international handbook of the languages of europe (pp. 822–841). De Gruyter Mouton. (Cit. on p. 2). find
  4. Bauer, L. (2001). Morphological productivity (Vol. 95). Cambridge University Press. (Cit. on pp. 1, 2). find
  5. Brinton, L. J., & Traugott, E. C. (2005). Lexicalization and language change. Cambridge University Press. (Cit. on pp. 1, 3). ↗
  6. Chandra, K., Chen, T., Li, T.-M., Ragan-Kelley, J., & Tenenbaum, J. (2024). Cooperative explanation as rational communication. Annual Meeting of the Cognitive Science Society (CogSci) (cit. on p. 2). ↗
  7. Chen, G., Van Deemter, K., & Lin, C. (2018). Modelling pro-drop with the rational speech acts model. International Conference on Natural Language Generation (cit. on p. 2). find
  8. Davies, M. (2010). The Corpus of Contemporary American English as the first reliable monitor corpus of english. Literary and Linguistic Computing, 25(4), 447–464 (cit. on p. 3). ↗
  9. Davies, M. (2012). The 400 million word Corpus of Historical American English (1810–2009). In English historical linguistics 2010 (pp. 231–262). John Benjamins Publishing Company. (Cit. on p. 3). ↗
  10. Frank, M. C., & Goodman, N. D. (2012). Predicting pragmatic reasoning in language games. Science, 336(6084), 998 (cit. on pp. 1, 2). ↗
  11. Gibson, E., Futrell, R., Piantadosi, S. P., Dautriche, I., Mahowald, K., Bergen, L., & Levy, R. (2019). How efficiency shapes human language. Trends in Cognitive Sciences, 23(5), 389–407 (cit. on pp. 1, 2). ↗
  12. Goodman, N. D., & Frank, M. C. (2016). Pragmatic language interpretation as probabilistic inference. Trends in Cognitive Sciences, 20(11), 818–829 (cit. on pp. 1, 2). ↗
  13. Jaeger, T. F. (2010). Redundancy and reduction: Speakers manage syntactic information density. Cognitive Psychology, 61(1), 23–62 (cit. on pp. 1, 3). ↗
  14. Jiang, G., Hofer, M., Mao, J., Wong, L., Tenenbaum, J., & Levy, R. (2024). Finding structure in logographic writing with library learning. Annual Meeting of the Cognitive Science Society (CogSci) (cit. on p. 1). ↗
  15. Jiang, G., Hofer, M., Mao, J., Wong, L., Tenenbaum, J. B., & Levy, R. (2025). Finding structure in logographic writing with library learning ii: Grapheme, sound, and meaning systematicity. Annual Meeting of the Cognitive Science Society (CogSci) (cit. on p. 1). find
  16. Levin, B., Glass, L., & Jurafsky, D. (2019). Systematicity in the semantics of noun compounds: The role of artifacts vs. natural kinds. Linguistics, 57(3), 429–471 (cit. on p. 1). ↗
  17. Levy, R. (2008). Expectation-based syntactic comprehension. Cognition, 106(3), 1126–1177 (cit. on pp. 1, 3). ↗
  18. Ma, Y., Peng, Y., & Zhu, Y. (2025). Word embeddings track social group changes across 70 dates in china. Annual Meeting of the Cognitive Science Society (CogSci) (cit. on p. 6). find
  19. Marelli, M., & Baroni, M. (2015). Affixation in semantic space: Modeling morpheme meanings with compositional distributional semantics. Psychological Review, 122(3), 485– 515 (cit. on p. 1). ↗
  20. Mattiello, E., & Dressler, W. U. (2018). The morphosemantic transparency/opacity of novel english analogical compounds and compound families. Studia Anglica Posnaniensia, 53(1), 67–114 (cit. on p. 1). ↗
  21. Mikolov, T., Chen, K., Corrado, G., & Dean, J. (2013). Efficient estimation of word representations in vector space. arXiv preprint arXiv:1301.3781 (cit. on p. 3). ↗
  22. Miller, G. A. (1995). WordNet: A lexical database for english. Communications of the ACM, 38(11), 39–41 (cit. on pp. 2, 3).
  23. Peng, Y., Ma, Y., Wang, M., Wang, Y., Wang, Y., Zhang, C., Zhu, Y., & Zheng, Z. (2025). Probing and inducing combinational creativity in vision-language models. Annual Meeting of the Cognitive Science Society (CogSci) (cit. on p. 1). ↗
  24. Piantadosi, S. T., Tily, H., & Gibson, E. (2011). Word lengths are optimized for efficient communication. Proceedings of the National Academy of Sciences (PNAS), 108(9), 3526– 3529 (cit. on p. 1). ↗
  25. Plag, I. (1999). Morphological productivity: Structural constraints in english derivation (Vol. 28). Walter de Gruyter. (Cit. on pp. 1, 2). find
  26. Reddy, S., McCarthy, D., & Manandhar, S. (2011). An empirical study on compositionality in compound nouns. International Joint Conference on Natural Language Processing (cit. on p. 1).
  27. Salge, C., Ay, N., Polani, D., & Prokopenko, M. (2015). Zipf’s law: Balancing signal usage cost and communication efficiency. PLOS One, 10(10), e0139475 (cit. on p. 1). ↗
  28. Shannon, C. E. (1948). A mathematical theory of communication. The Bell System Technical Journal, 27(3), 379–423 (cit. on p. 1). ↗
  29. Štekauer, P., & Lieber, R. (Eds.). (2005). Handbook of word-formation (Vol. 64). Springer Science & Business Media. (Cit. on p. 1). find
  30. TheWelcomer. (2025). MorphSeg [GitHub repository]. https://github.com/TheWelcomer/MorphSeg (cit. on p. 3). ↗
  31. Xu, A., Kemp, C., Frermann, L., & Xu, Y. (2024). Word reuse and combination support efficient communication of emerging concepts. Proceedings of the National Academy of Sciences (PNAS), 121(46), e2406971121 (cit. on pp. 1, 2, 5). ↗
  32. Zaslavsky, N., Kemp, C., Regier, T., & Tishby, N. (2018). Efficient compression in color naming and its evolution. Proceedings of the National Academy of Sciences (PNAS), 115(31), 7937–7942 (cit. on pp. 2, 5). ↗
  33. Zhang, Y., Li, M., Long, D., Zhang, X., Lin, H., Yang, B., Xie, P., Yang, A., Liu, D., Lin, J., et al. (2025). Qwen3 embedding: Advancing text embedding and reranking through foundation models. arXiv preprint arXiv:2506.05176 (cit. on p. 3). ↗
  34. Zipf, G. K. (1949). Human behavior and the principle of least effort. Addison-Wesley Press. (Cit. on p. 1). find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yang-2026-rational-communication-shapes-morphological-composition/