Evolving Dialects.

October 10, 2026 by 17 Comments

I found this site via Facebook; it’s based on research by James Burridge (seen here in 2022) and Bert Vaux (at LH in 2017), and the intro is as follows:

[blockquote deleted]

The site is so spiffy and designed-to-the-hilt that I’m a little afraid of it, but I trust those who know more about statistics and dialects will look into it, and hopefully share what they find in the comment thread.

Update. Commenters point out that the site is apparently AI slop, and I regret having posted it.

Comments

  1. Image Yuval says
    October 13, 2026 at 3:40 am

    Everything about the design of this website screams “built with AI”, for what it’s worth.

  2. Image Greg says
    October 13, 2026 at 4:55 am

    Regrettable that the site doesn’t give any idea of who authored it. (… Or who had an AI make it, and takes responsibility for it.) It cites Burridge and Vaux, but doesn’t say either of them were actually involved in this site.

    As Yuval says, this looks very much like something made by a 2026-era LLM tool. That’s understandable for getting a tool like this built with a reasonable amount of work — but I’d have more confidence in it if there were human authors visibly taking responsibility who can be expected to know what they’re doing and have checked the results.

    The fun interactive bits don’t work for me anyway, though: there’s just a message saying it’s loading the data, indefinitely, and the maps remain gray. A shame, because Vaux’s surveys definitely make some interesting data to play with!

  3. Image Brett says
    October 13, 2026 at 7:19 am

    The descriptions have characteristic LLM disfluencies too. Mention of “statistical physics” was particularly glaring to me. In this context, it should be “statistical mechanics,”* if the program were actually using entropy methods, which it doesn’t orherwise seem like it is.

    * At MIT, the undergraduate physics courses 8.044 and 8.08 were Statistical Physics I & II. The graduate courses, 8.333 and 8.334 were Statistical Mechanics I & II.

  4. Image languagehat says
    October 13, 2026 at 8:26 am

    Everything about the design of this website screams “built with AI”, for what it’s worth.

    Argh! You must be right, and that’s why (subliminally) I found it offputting. Sorry for propagating slop, and let this be a warning to me for the future.

  5. Image David Eddyshaw says
    October 13, 2026 at 9:42 am

    There seem to be a surprising number of LLM-generated pseudolinguistics websites out there nowadays. A search for [random West African languagej course/lessons/grammar reliably turns up several, as I discovered recently when looking for a link to the not-too-bad FSI Mooré course materials. Most are evidently based on plagiarism of legit online stuff, mixed in with random “hallucinations” (LLM-pusher-speak for “grotesque errors”), but some are pretty much completely made up.

    Evidently the techbros have realised the awesome spending power of language nerds, and are trying to tap into it. Either that, or they’re trying to take down everyone interested in Language (a bunch of Commies anyway) by infecting their computers with viruses, so that all online discourse henceforward will be confined to MAGAspeak. Doubleplus ungood!

  6. Image mollymooly says
    October 13, 2026 at 10:51 am

    I regret having posted it—if it’s any consolation, I for one welcome all salutary lessons, especially the vicarious ones.

  7. Image Brett says
    October 13, 2026 at 10:55 am

    @David Eddyshaw: One point off for capitalizing “doubleplus.”

  8. Image Y says
    October 13, 2026 at 4:32 pm

    At the bottom of the page are links to two papers (on arXiv), the recent one being Statistical physics of language change inferred from time evolving maps by Burridge and Vaux. Burridge has written other papers along these line; click on his name at the arXiv page.

    (The second one, Burridge 2025, has an apparent typo. In figure 17 he offers maps showing the distribution of soda/pop words in future decades, as predicted from his model. The four maps appear identical.)

    People have been modeling language change using statistical models for quite a while. I have noticed papers on the subject but haven’t been following the literature. This study uses supposedly more complex tools, inspired by statistical physics, plus data from Vaux’s surveys. He fits his model to the data, looks at the residuals, and all that. This is not bad science, though it doesn’t set my own particular heart aflutter.

    The website could well have been designed by AI, or done using some common templates. I looked at the page source but didn’t find any obvious clues either way, but I’m not in that biz. I find it repulsive, despite and because of its trying to appear slick. Human-made or machine-made, it’s slop in my eyes.

  9. Image Brett says
    October 13, 2026 at 5:06 pm

    What they seem to be doing in the those papers is manifestly statistical modeling, arguably statistical mechanics (although I would tend to say it isn’t), and definitely not statistical physics.

  10. Image Madeline Kalvis says
    October 13, 2026 at 5:23 pm

    As overworked grad students increasingly lean on AI to meet impossible deadlines, the line between “have a human make your website” and “have Claude make your website” blurs to nonexistence.

  11. Image Y says
    October 13, 2026 at 5:47 pm

    @Brett: What’s the difference between statistical physics and statistical mechanics? And in other works by the same author on similar subjects, while the use of either is of course metaphorical, the mathematical techniques are borrowed from whatever you’d call the study of diffusion, magnetic domains, bubbles, etc.

    @MK: Burridge is a full professor. All the same, not everyone knows how to build even a basic website. If I had to do it I would very grudgingly use some off-the-shelf canned solution like WordPress, while cursing having to fill my head with useless stuff. I would not trust Fraude or any other AI to do it for me, but that’s just me. Maybe some universities have people on staff who do web design for the faculty, the way they used to have staff for typing and drafting?

  12. Image Greg says
    October 13, 2026 at 6:02 pm

    Was Burridge (or a student working with him) actually involved in making this? The site doesn’t say, as far as I can see.

  13. Image languagehat says
    October 13, 2026 at 6:27 pm

    As far as I can tell they just claim it’s based on his work; I too see no sign that he was involved.

  14. Image Brett says
    October 13, 2026 at 6:53 pm

    @Y: Statistical mechanics is a suite of mathematical and conceptual tools that can be applied to all kinds of systems. Statistical physics is simply the application of statistical mechanics to problems specifically in physics. This paper of mine is about statistical mechanics, applied to gene repertoires, so it’s not statistical physics. On the other hand, this one is statistical physics.

  15. Image Madeline Kalvis says
    October 13, 2026 at 6:53 pm

    I’m aware Burridge is a full professor. Does anyone serious believe he is writing his own code?

  16. Image Y says
    October 13, 2026 at 7:19 pm

    Hat, what FB page linked to it?

    FWIW, the preprint is dated on Sept. 21, and the web domain was created three days later. Its registration info is hidden. Not surprisingly there’s no mention of the website in the preprint.

  17. Image languagehat says
    October 13, 2026 at 9:45 pm

    Hat, what FB page linked to it?

    Sorry, I don’t remember. If I happen on it again I’ll link it in a comment.

Speak Your Mind

*