Language Log

[This is a guest post by Diana Shuheng Zhang]

(click to embiggen)

During my visit to the Luoyang Museum 洛阳博物馆, I found something amazing in its museum store. It is a set of 24 postcards, corresponding to the 24 solar terms 节气 (jieqi) in the traditional Chinese lunar calendar for agricultural purposes. What’s special about this set of cards is the design: every disyllabic lexeme for a solar term is coined into one single Sinitic “character”. I intended to attached multiple photos as examples but the last email was not successfully sent — the “size” was too big as an email. Therefore, here I’m only attaching one photo, that depicts the whole scene of the set of 24 cards.  By clicking on the photograph, you will be able to enlarge it sufficiently to enable you to see the details of the artwork and the writing.

On first blush, I thought perhaps the person pictured had a double chin, and by cropping the photo this way they were trying to hide it.  On second blush, it was clear that they had misinterpreted the name of the famous 2nd c. statesman, general, and poet, Cao Cao 曹操 (ca. 155-220 AD).

Most of the population of China cannot afford or obtain a VPN (Virtual Private Network).  For the privileged portion who do manage to purchase and install a VPN, does that solve their global internet access problem?

Some people claim that, with a good VPN, you can see anything on the global internet in China.  But I don't think that's quite true.  Sometimes I direct my students (all of whom have VPNs) to various Wikipedia and Wiktionary articles, and they say, "Sorry, Prof. Mair, I can't access that in China" (blocked without warning or explanation twenty-some years ago*).  The problem is even more acute with YouTube.

YouTube has all sorts of stuff:  richly informative, inflammatory, educational, political, DIY, linguistic, etc., etc.  Practically anything one can imagine, except probably outright pornography.  The Chinese government is deathly afraid that its citizens might see YouTube content that is critical of Xi Jinping, the CCP, and so forth.  So YouTube is a no-no, and that means nature films, music, poetry, art, archeology, chemistry, physics, recipes, Chinese / Japanese / Indian / Iranian / etc. culture.  How impoverished the Chinese people are because of the benighted policies of their government!  The nearly 15 BILLION videos on YouTube are off limits to Chinese citizens.  And that's just YouTube.  Think of all the internet riches that are unavailable to the people of China.  No wonder so many of them are desperate to go abroad to study where they have free access to the internet.

Hannah Al-Othman, "‘Mad fer it’: Greater Manchester Aldi to keep Aldeh name in tribute to Oasis", The Guardian 7/24/2025:

An Aldi store in Greater Manchester rebranded Aldeh in honour of Oasis is to keep the new name, the supermarket chain has said.

The new sign was erected at the Prestwich store before the Oasis homecoming gigs earlier this month. The band played five sold-out shows at Heaton Park, which is near the store.

It was meant to be a temporary name change, but the sign has been a massive hit with tourists and local people, with Oasis fans queueing outside for selfies.

The sign has even been listed on Google Maps as a cultural landmark, with a string of glowing five-star reviews calling it “the Stonehenge of a generation” and “the greatest rebrand of all time”.

"Tom Lehrer, Musical Satirist With a Dark Streak, Dies at 97", NYT 7/27/2025.
Nancy Friedman's skeet underlines who and what he outlived:

He outlived Henry Kissinger and New Math. RIP Tom Lehrer.
(Gift link) www.nytimes.com/2025/07/27/a…

[image or embed]

— Nancy Friedman (@fritinancy.bsky.social) July 27, 2025 at 1:14 PM

Today's Tank McNamara:

The most interesting LLM research that I've seen recently is from Alex Cloud and others at Anthropic and Truthful AI, "Subliminal Learning: Language models transmit behavioral traits via hidden signals in data", 7/20/2025:

ABSTRACT: We study subliminal learning, a surprising phenomenon where language models transmit behavioral traits via semantically unrelated data. In our main experiments, a "teacher" model with some trait T (such as liking owls or being misaligned) generates a dataset consisting solely of number sequences. Remarkably, a "student" model trained on this dataset learns T. This occurs even when the data is filtered to remove references to T. We observe the same effect when training on code or reasoning traces generated by the same teacher model. However, we do not observe the effect when the teacher and student have different base models. To help explain our findings, we prove a theoretical result showing that subliminal learning occurs in all neural networks under certain conditions, and demonstrate subliminal learning in a simple MLP classifier. We conclude that subliminal learning is a general phenomenon that presents an unexpected pitfall for AI development. Distillation could propagate unintended traits, even when developers try to prevent this via data filtering.

From Alex Baumans:

Hyeri needs translation by Haneul while talking with Kiss of Life! #kissoflife #키스오브라이프 #혜리

https://www.youtube.com/shorts/zv6hFvQyiDM

I've recently noticed two compound neologisms, both involving metaphors about foot-related self injury.

The first one was in an article in Medium on 6/27/2025, "Why Google is Betting 8 Years on a Programming Language That Doesn’t Exist Yet". That article explains that

In 2022, Google introduced Carbon, a potential successor to C++. Unlike Go or Rust, Carbon wasn’t ready for prime time. In fact, it was barely out of the conceptual phase.

And among the reasons given for the effort [emphasis added]:

C++ has steep learning curves and footguns.

The second foot-related compound was in a 7/24/2025 TPM article, "Why is Jeff Bezos rakestomping the Post?".

From Glenn B.:

I just spotted a pair of recently introduced resolutions in the New Jersey legislature that might be of interest to Language Log. SJR 167 and the identical AJR 230 would (if adopted) recognize Sanskrit "as one of the world languages."

Not all of the claims made on behalf of Sanskrit seem kosher to me, particularly the claim that "Sanskrit has a unique origin, not created by man," but I'd love it if Language Log were able to provide a more authoritative discussion.

This is an odd-looking word that I encounter fairly frequently, especially in my publishing ventures.  Since I don't understand how / why "queue" should be pronounced like "cue", which is also a variant spelling for the same word, I'm especially cautious about "queue" when I approach it.  Moreover, since I'm steeped in pinyin, I'm tempted to pronounce "queue" as "chyueyue" (!).  Consequently, I always have to slow down when I spell / type it:  "q-u-e-u-e", which I seldom have to do with other words except "Cincinnati", which I still haven't mastered.

Other than "its / it's", "queue" is probably the most frequently misspelled word I know of, even among educated persons.

I also am somewhat perplexed why "queue" means both "line" and "tail".

The word "queue" is used to mean a line, particularly in British English, because of its etymological origins. "Queue" comes from the French word "queue," meaning "tail," which in turn comes from the Latin word "cauda," also meaning "tail". This connection to "tail" makes sense when visualizing a line of people or objects, as they often form a linear arrangement reminiscent of a tail. The term "queue" is also used in computing to refer to a data structure where items are processed in a first-in, first-out (FIFO) manner, similar to how people are served in a line.

(AIO)

More Julesy:

Why 50% of modern Chinese vocabulary was made in Japan

The PR campaign for the Unitree GI Robot now comes in at least three local variants: the "Uncle Bot" in China,  "Jake the Rizzbot" in Austin, and a gay version of Jake in Los Angeles.