WaST: a formalisation of the Wave model with associated statistical inference and applications We propose a mathematical formalisation of the ``wave model'' originally developed in historical linguistics but with further applications in human sciences. This model assumes new traits appear in a population and spread to nearby populations depending on their closeness. It is mostly used to describe joint evolution of closely related populations, for example of several dialects. These situations of permanent contact are not accurately represented by its competitors based on tree structures. We built a fully Bayesian generative model where innovation spread along a fixed graph and disappear according to a death process. We then develop a Metropolis-Hastings within Gibbs sampler to sample from the posterior distribution on the graph. We test our method on simulated datasets as well as on several real dataset.

I was told the Wave model in linguistics had not been formalised. Here is an attempt at it. Note that many things remain to be done, check the discussion for that.
The method/model I propose, WaST, can be used on any cognacy dataset. I'll put the code on git soon.
arxiv.org/abs/2604.08220

As Laura said, our results show that quantitative phylolinguistics methods (those based on cognatized binary data) can't reconstruct deep topologies beyond 8-10 millennia.

Thanks Laura! We should have a preprint to share in a few weeks.
It's part of Emma Kopp's PhD, and co-authored with @thomaspellard.bsky.social and @rgyalrong.bsky.social

"The tool introduced in this paper seems akin to adding ketchup to a culinary masterpiece by Robuchon simply because it's available."

(The reviewer also made several very valid points. Apparently, the editor didn't re-invite that reviewer when we submitted the revised paper.)

Family Tree Model The family tree remains the main metaphor for describing the evolutionary history and relationships of languages. Languages, like biological species, evolve following processes such as descent with m...

Our chapter on the tree model (with @thomaspellard.bsky.social and @robinryder.bsky.social ) is at last published online: onlinelibrary.wiley.com/doi/10.1002/...

I only knew Chris Sims' work, and I wish I had got to meet him. But to the extent that we know something of each other through what we put out in the world, here is my personal biased favourite of his: Why Econometrics Should Always and Everywhere Be Bayesian sims.princeton.edu/yftp/EmetSoc...

Est-ce que l'échelle des ordonnées ne serait pas erronée ?
Si j'en crois le texte de l'article, l'effet après 20 ans est de 1.28%, mais d'après la figure ce serait 0.0128%

"Quand les bars-tabacs ferment : l’érosion du lien social local et la progression du vote d’extrême droite en France" www.cepremap.fr/2026/01/quan...
file:///Users/arthurcharpentier/Downloads/Figure_1-Note.png

What a pre-conference workshop on Bayesian statistical methods—a powerful toolkit for our community! 📊

Many thanks to Daniel Redhead and Ramona Roller for guiding us through STRAND and to Michael Chimento for the introduction to STbayes.

And thank you to all participants for their energy! 🙌

So I pleased to announce the conceptual spawn of FigTree: PearTree (acronym still to be finalised). If you want to dive right in it is hosted as a web app here: artic-network.github.io/peartree (click the “Example...” button for immediate candy and then click every button you can find).

Front cover of my book, titled "Comparative musicology: Evolution, universals, and the science of the world's music" (published today by Oxford University Press)

1st of my 4-page essay published in Nature today titled "Music is not a universal language - but it can bring us together when words fail" Picture caption: "Puerto Rican rapper Bad Bunny (centre) performed in Spanish at the half-time show of the 2026 American Football Super Bowl LX."

My book is now published! 🌏🎶🧪

You can download it for free at academic.oup.com/book/62353 - I’d be grateful if you do!
I also published an accessible summary with audio/video today in @nature.com: www.nature.com/articles/d41...
Try reading that first, then give the whole book a read if you like it!

So cool!
Mostly fine until 1300, understood a fair bit of 1200, and close to nothing in 1100

I guess if you want to see what it adds compared to your current version, you need to apply setdiff to the output of that and to your current list of (recursive) dependencies.

tools::package_dependencies("packagename", recursive=T)

Photo of Christian Robert, Antoine Luciano, and Robin Ryder

Antoine will be spending the next few months in Berkeley. You should definitely try to hire him as a postdoc!

The third project is ongoing, on using some neural network methods to accelerate ABC.

The second project is on using permutations for Approximate Bayesian Computation, see this thread.

bsky.app/profile/robi...

Insufficient Gibbs sampling - Statistics and Computing In some applied scenarios, the availability of complete data is restricted, often due to privacy concerns; only aggregated, robust and inefficient statistics derived from the data are made accessible....

Antoine's PhD consists of 3 separate projects.

The first, Insufficient Gibbs Sampling, relies on data augmentation to sample from the posterior based only on robust but insufficient statistics (e.g. the median, quantiles…), a common situation when privacy matters.

link.springer.com/article/10.1...

Photo of 2026 PhD thesis by Antoine Luciano, with title "Contributions à l'inférence bayésienne par simulation"

Congratulations to Dr Antoine Luciano, who defended his PhD on "Contributions to Bayesian Simulation-Based Inference"!

I had a wonderful time supervising Antoine's PhD (jointly with Christian Robert), it's always emotional to reach the end.

Something like this could be useful when reviewing though. If there are hallucinated references, I'd reject without reading any further.
But of course we don't have access to the .bib file directly for papers we are reviewing.

Thanks, that looks super relevant! I'll forward it to Jinyuan

Now en route to Paris for another PhD viva tomorrow: Antoine Luciano, supervised by Xian Robert and myself. Stay tuned!

The associated paper isn't online yet, so you'll have to be patient to read it. Code should soon be available, since that's one of the corrections we requested. ;-)

Output of an HMM, showing the Posterior probability of each word in Good Omens being written by Neil Gaiman.

Jinyuan's methodology, based on HMMs, is a significant improvement on the state of the art, and also allows for better uncertainty quantification.
Here is what the output can look like, for Good Omens - notice the very large number of switches between the two authors.

Examples of applications:
* the novel "Good Omens" was co-written by Terry Pratchett and Neil Gaiman. Who wrote which parts?
* take text written partly by a human and partly by a LLM. Which parts are due to the LLM?

Photo of a PhD thesis with title "Sequential Stylometry: A Bayesian Stylometry Framework for Mixed Authorship Texts", by Jinyuan Zhang (University of Edinburgh, 2025)

Congratulations to Dr Jinyuan Zhang, who defended his PhD in Edinburgh today! I was happy to be the external examiner.

Jinyuan worked on Hidden Markov Models for Stylometry, to attribute authorship in texts with several authors.

The thesis was supervised by Gordon Ross.

This etymology is wild!

#TIL The English word “average”, originally “custom duty” or “loss in transported goods”, comes from the French “avarie” (meaning “damage to a ship or cargo”) from the Italian “avaria” (same meaning), …

The Fréchet mean was computed only a small phylogenies (≤8 leaves). We don't know much yet about the properties of this summary. I'll be interested to see whether it can be used in practice to summarize samples of trees.

Another contribution is the definition of a Wasserstein distance over these sets of trees, including with different sets of leaves.

An important contribution of Roan's thesis is to study the properties of the Fréchet mean of trees. They showed that this can be defined, although not always uniquely. They obtain several theoretical results, such as a Law of Large Numbers and a CLT.