Body-Part Reflexives and Body-Zone Classification

A Typological Survey with Substrate Implications

Preprint — Version 2 — July 2026

Jan Ritch-Frel and Ed Phillips

PREPRINT — Not peer reviewed

Abstract

This paper documents two typological phenomena and explores their possible connections. Track A: the grammaticalization of "head" nouns into reflexive pronouns (HEAD→SELF) clusters in the Caucasus at rates above the cross-linguistic background. A Bayesian generalized linear mixed model (brms; Bürkner 2017; formula: bpdr ~ caucasus + lat_c + lon_c + (1|Family); weakly informative priors; 4 chains × 2,000 draws) yields a Caucasus effect of OR = 10.30 (95% credible interval [1.82, 55.86]; 99.4% of posterior mass positive), based on 2,463 languages across 216 families from Grambank. The wide credible interval reflects genuine uncertainty in the effect size, though the model consistently favors a positive Caucasus effect. The same grammaticalization pathway is widespread across Afro-Asiatic, attested with independent lexical roots in five of six branches. Track B: Great Andamanese and Kusunda — both isolates in geographic refugia — share a structurally parallel body-zone noun classification architecture not attested in any other documented language family. A peer-reviewed reconstruction of Proto-Kusunda (Spendley 2024) independently identifies the parallel, describing the Kusunda system as "strongly reminiscent" of the GA morphology. The paper proposes an elaboration model and explores a substrate/dispersal interpretation as a hypothesis for future testing. The R analysis pipeline (brms/rstan) is provided for independent replication, cross-validated against an independent Python/bambi implementation.

Downloads

Contact the authors for the full preprint PDF and supplementary data (12 CSV tables + R analysis pipeline with testthat test suite).

Data Sources

• Grambank v1.0.3 — Skirgård et al. (2023). "Grambank reveals the importance of genealogical constraints on linguistic diversity." Science Advances 9(16): eadg6175.
• CLICS4 — Tjuka et al. (2025). Database of Cross-Linguistic Colexifications. Zenodo.
• Lexibank v2.1 — Zenodo.
• Evseeva & Salaberri (2018/2019). "Grammaticalization of nouns meaning 'head' into reflexive markers." Linguistic Typology 22(3): 385–435; Corrigendum 23(1): 255–262.
• Spendley (2024). "Possessive prefixes in Proto-Kusunda." Himalayan Linguistics 23(1): 57–71.
• Bürkner (2017). "brms: An R Package for Bayesian Multilevel Models Using Stan." Journal of Statistical Software 80(1): 1–28.

Version History

• Version 2 (July 2026): Reframed as hypothesis-generating typological study. R/brms model (OR = 10.30). Co-authored with Ed Phillips. Removed deep-time inheritance headline claim. Added alternative explanations, falsifiability conditions.
• Version 1 (June 2026): Initial preprint.