henhaobio
Sign in

bioRxivGeneticsPreprint

Triangulation fails when neither linguistic, genetic, nor archaeological data support the Transeurasian narrative

Tian ZhengiD, Yuxin TaoiD, Kongyang ZhuiD, Guillaume JacquesiD, Robin RyderiD, José Andrés Alonso de la FuenteiD, Anton AntonoviD, Zi‐Yang XiaiD, Yuxuan ZhangiD, Xiaoyan JiiD, Xiaoying ReniD, Guanglin HeiD, Jianxin GuoiD, Rui WangiD, Xiaomin YangiD, Jing ZhaoiD, Dan XuiD, Russell David GrayiD, Menghan ZhangiD, Shaoqing WeniD, Chuan-Chao WangiD, Thomas PellardiD

PDFDOI
In-depth full-text digest available · Jump to digest ↓

Abstract

Robbeets et al.1 argue that the dispersal of the so-called "Transeurasian" languages, a highly disputed language superfamily comprising the Turkic, Mongolian, Tungusic, Koreanic, and Japonic language families, was driven by Neolithic farmers in the West Liao River region of China. They adduce evidence from linguistics, archaeology, and genetics to support their claim. An admirable feature of the Robbeets et al.s paper is that all their datasets can be accessed. However, a closer investigation of all three types of evidence reveals fundamental problems with each of them. Robbeets et al.s analysis of the linguistic data does not conform to the minimal standards required by traditional scholarship in historical linguistics and contradicts their own stated sound correspondence principles. A reanalysis of the genetic data finds that they do not conclusively support the farming-driven dispersal of Turkic, Mongolian, and Tungusic, nor the two-wave spread of farming to Korea. Their archaeological data contain little phylogenetic signal, and we failed to reproduce the results supporting their core hypotheses about migrations. Given the severe problems we identify in all three parts of the "triangulation" process, we conclude that there is neither conclusive evidence for a Transeurasian language family nor for associating the five different language families with the spread of Neolithic farmers from the West Liao River region.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (12 Jun 2022) · CC BY 4.0

Reanalysis of Robbeets et al.'s linguistic, genetic and archaeological datasets finds that none supports a Transeurasian language family or its spread with West Liao River Neolithic millet farmers.

Robbeets et al. used 'triangulation' of linguistic, archaeological and genetic evidence to argue that Turkic, Mongolian, Tungusic, Koreanic and Japonic form one Transeurasian family that Neolithic farmers spread from China's West Liao River region. Working from that study's openly accessible datasets, this reanalysis tests each line of evidence separately, finds that none supports the claim, and notes that the triangulation procedure was never defined.

Key results

  1. ShownOnly 17 of the 3166 cognate sets that Robbeets et al. list for a Transeurasian family meet the regular sound-correspondence criteria that those authors set out themselves.

    Of 3166 sets, 317 span more than one family, 50 span more than two and 2 span all five. Among the 50, five include words that the original authors marked as borrowings, and two assign forms to the wrong language. A systematic computer-assisted check left 17 conforming etymologies (Figure 1a, SI 1), and most of those have other problems. · Linguistics; Figure legends

  2. ShownNone of the 43 agropastoral word comparisons meant to identify Proto-Transeurasian speakers as early millet farmers meets every criterion that the original authors laid out.

    Of the 43 comparisons, 31 follow the proposed sound correspondences, 29 can be reconstructed at the Proto-Transeurasian level, 27 occur in more than three branches, 22 have comparable meanings and 9 belong to agropastoral vocabulary. None satisfies all criteria (Figure 1b, SI 2). · Linguistics; Figure legends

  3. ShownReanalysis of the original genetic data cannot separate Hongshan (Neolithic) from Upper Xiajiadian (Bronze Age) as sources for ancient Korean and Japanese populations, so the two-wave farming spread to Korea is unsupported. Robbeets et al.'s own SI 13 admits this lack of resolution.

    f4(Mbuti, X; Hongshan, Upper_Xiajiadian) gave no significant Z-scores. qpWave found Hongshan and Upper Xiajiadian homogeneous relative to all newly generated samples (P = 0.371). Two-way qpAdm models could not tell the two ancestries apart, and populations that the original study assigned Upper Xiajiadian ancestry could be modelled with Hongshan instead (Figure 2a to c). · Genetics; Figure legends

  4. ShownRe-running the phylogenetic analysis of the archaeological traits does not recover separate Neolithic migration clades to Korea and to Primorye. It also finds no close tie between Bronze Age West Liao sites and the Mumun and Yayoi cultures.

    In the majority-rule (≥ 50%) consensus tree of the corrected data (Figure 3), West Liao sites Xiaohexi Chahai I (#8) and Xiaohexi Bayinchanhan (#7) group with Chulmun, Zaisanovka and one Mumun site (#117) in a single clade. A Yayoi plus Mumun clade is supported but shows no close relation to Bronze Age West Liao sites. · Archaeology; Figure legends

  5. SuggestedThe archaeological feature classes carry little, conflicting tree-like signal and seem to have different histories, so combining them into one phylogeny is questionable.

    Using Robbeets et al.'s feature classes (SI 3), a ceramics-only run gives a highly unresolved, rake-shaped tree that supports none of their hypotheses. Other subsets show even less signal, and the buildings/houses, shell and bone artifacts and burials subset recovers no clade that matches a known archaeological culture. The main text reports no numeric measure of tree signal. · Archaeology

Methods, data and reproducibility

A point-by-point re-examination of Robbeets et al. (2021). Their cognate sets were screened by computer against their own stated sound-correspondence criteria, and their 43 agropastoral comparisons were checked against every criterion the original authors gave. The genetic data were reanalysed with f4, qpWave and qpAdm, and the archaeological traits were given new phylogenetic analyses, in full and by feature class.

  • Cognate sets that Robbeets et al. (2021) proposed for Turkic, Mongolian, Tungusic, Koreanic and Japonic · n = 3166 cognate sets
  • Agropastoral lexical comparisons that Robbeets et al. proposed to link Proto-Transeurasian speakers with early farming · n = 43 comparisons
  • Robbeets et al.'s ancient genetic data, including Hongshan, Upper Xiajiadian and ancient Korean and Japanese populations, with Mbuti as the f4 outgroup · human (ancient populations)
  • Robbeets et al. (2021) archaeological feature matrix covering Neolithic to Bronze Age sites across Japan, Korea, the Primorye and northern China · n = 171 features for 255 sites
Methods & tools
f4 statistics, qpWave, qpAdm, BEAST (used in the original analysis), majority-rule consensus tree
Reproducibility gaps
The text prints no code or data links. The detailed linguistic and archaeological reanalyses are in Supplementary Information 1 to 3, which is not part of this text. The work relies on Robbeets et al.'s publicly accessible datasets.
Limitations (3)
  • Robbeets et al. never define their 'triangulation' procedure, so the three corrected analyses could not be combined using the original method. · stated by the authors
  • The genetic reanalysis shows that Hongshan and Upper Xiajiadian cannot be told apart with these data (non-significant f4, qpWave P = 0.371). That leaves the two-wave model unresolved rather than refuted, and beyond swapping Hongshan in as a source the reanalysis proposes no positive alternative population history. · noted in this summary
  • The main text gives no quantitative tree-likeness statistic, and much of the linguistic and archaeological detail is in Supplementary Information 1 to 3, which is not in this text. The negative findings therefore cannot be fully weighed from the main text. · noted in this summary

Read this if you work on Northeast Asian prehistory or language-dispersal models, or want an example of re-testing a combined linguistic, ancient genetic and archaeological-phylogenetic claim from its published data.

Citations
48

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISRobbeets '22Yunusbayev '14Wang '20Santos '20Zeng '23Kutanan '18
88% semantic matchEvolutionary Biology
M. Robbeets, M. Hudson, C. Ning+18 more·bioRxiv·Oct 5, 2022