Further on beyond cognacy
Model adequacy and meaningful branch lengths in alignment-based phylogenetics
A companion to the talk by Gerhard Jäger, given at QC-LESTE, Paris, on 11 June 2026.
Phylogenetic linguistics usually needs a cognate-classification step before a tree can be built. The Lexibank World Tree showed that one can skip it — align, don’t classify — and still recover an excellent topology for 3,397 languages. That work left two questions open:
- What are the statistical properties of alignment characters?
- Its own caveat: “branch lengths are not meaningful.” Can we get meaningful ones?
This talk goes further beyond cognacy on both, in four steps: model adequacy, the dimensionality of the sound-class signal, a diagnosis of the branch-length pathology, and an Ornstein–Uhlenbeck fix that restores a sensible, interpretable geometry.
Resources
- 📑 Slides (PDF):
slides_paris2026.pdf - 💻 Code:
further_on_beyond_cognacyon GitLab - 🌳 Upstream world tree: Jäger (2026), doi.org/10.57754/FDAT.ccfpp-z0113
Cite
Gerhard Jäger. Further on beyond cognacy: model adequacy and meaningful branch lengths in alignment-based phylogenetics. Talk, QC-LESTE, Paris, 11 June 2026.
Building on:
Gerhard Jäger. A Phylogenetic Tree of 3,397 World Languages via Multiple Sequence Alignment. University of Tübingen, 2026. doi.org/10.57754/FDAT.ccfpp-z0113