Language Log

[This is a guest post by IA]

Let's set the stage. A writer has drafted an essay for a publication that has specified a maximum word count of 5,000, and a preferred range of 1350 to 2700. The draft totaled 4,869 words, so it was within the limit, but not in the preferred range. Facing an imminent deadline, and knowing that I have a GPT-4 subscription, the writer asked me to try using ChatGPT to produce a (draft of a) shorter draft.

Last night I tried, with results that were both interesting and frustrating.

The goal was not to test GPT-4, but (perhaps) to speed up the creation of a shorter draft. And I'm not going to comment on the content of the edited version — which I gather was mostly good enough to be useful,  but sometimes wrong, misleading, or meaningless.

The full dialog is below. Commenters will no doubt notice my poor-quality "prompt engineering", but still, the interaction suggests that counting words is not one of GPT-4's strengths…

Jichang Lulu briefly alluded to work on languages of Italy in the dialectometry thread (here [the whole comment is well worth reading, as are the comments by Jonathan Smith [here — this one on an earlier thread, here, here, and here] on that post). He also thought that Language Log readers might find of interest some comments in this paper by Mauro Tosco.

"Measuring languageness:  Fact-checking and debunking a few common myths", DIVE-IN

“Interestingly, the more traditional classifications are marred by purely sociolinguistic analyses – and quite often their accompanying political and ideological underpinnings – the more they are proven wrong when dialectometry is applied.”

(Tosco:  homepage; International Research Group on Contested Languages)

"They are known as one of archaeology’s great enigmas – hollow 12-sided objects from the Roman era with no known purpose or use."

So begins this article by Jessica Murray in The Guardian (4/29/24):

Mysterious Roman dodecahedron to go on display in Lincoln

There are no known descriptions or drawings of object in Roman literature, making its purpose unclear

Roman bronze dodecahedron found in Tongeren, Gallo-Roman Museum, Tongeren

Zoe Greenberg, "Are we saying 'Passyunk' wrong?", The Philadelphia Inquirer, 4/23/2024:

In this time of widespread division and chaos, The Inquirer decided to unite all Philadelphians by documenting the definitive way we pronounce “Passyunk.” Were we motivated to act by a random New Yorker article confidently declaring this word is pronounced “‘passion’ with a ‘k’”? Absolutely. But our quest grew far beyond that.

The effort left some of us, and those we interviewed, questioning who we were and what we know on a fundamental level. One woman interviewed by The Inquirer, for example, claimed to pronounce the word exactly the same as her husband, who proceeded to pronounce it completely differently.

In our perennial discussions on the supposed mutual intelligibility of the countless, so-called "Chinese dialects" of the allegedly monolithic / monolingual Hànyǔ 漢語 ("Sinitic"; my colleague IA calls it "Hannic"), we seldom take into account the actuality of what these innumerable lects sound like on the ground / street.  Let's take a listen to this 4-year-old kid from Tianjin, which is close (70 miles) to Beijing, singing in the local Muttersprache, here.

Serious questions about "religion" in Sinitic

Below the fold is for advanced specialists in Chinese philology, theology, and lexicography.  Even for them, it is recommended that readers prepare themselves by reviewing "Spelling with Chinese character(istic)s, pt. 4" (7/4/16).

[This is a guest post by IA]

BREAKING: University of Florida students chose to breakdown their encampment after being handed this of Allowable Activities and Prohibitive Items and Activities.

Look at those Consequences for Non-Compliance

University of Florida's chapter of Young Democratic Socialists of… pic.twitter.com/l4jYjrSVcr

— Stu (@thestustustudio) April 26, 2024

Every individual's speech is variable. And when we look beyond the individual, we see variation across space, time, style, and social structure — among other dimensions. And these variations are generally gradient rather than abrupt, although standardization efforts by national or regional governments may try to eliminate the variation.

For millennia, scholars have noted and catalogued these patterns of variation — and for the past couple of hundred years, this study has been called dialectology. But until 1970 or so,  people interested in this topic faced an uncomfortable choice: you can either pretend (falsely) that the variation can be put into a few well-defined boxes; or else you can limit your research to compiling very large lists of who said what where, when, and why.

About 50 years ago, some European researchers began trying to get past this dichotomous barrier, under the banner of "dialectometry". For a recent survey, see Martijn Wieling and John Nerbonne, "Advances in dialectometry", 2015 (from which I'll quote a long explanation):

This is a recent instagram post of the Vietnamese singer Suni Ha Linh.

View this post on Instagram

A post shared by SUNI (@sunihalinh)

Jennifer Rubin, "Has Trump’s family abandoned him? I’m answering your questions", WaPo 4/24/2024:

Q: Are Republicans the party of no? Why can't Republicans say yes? Instead of getting a border deal in exchange for Ukraine funding, they got nothing.

A (Jennifer Rubin, Opinion Columnist):
Yup. They are the masters at cutting off their noses despite their faces. Remember that they really do not want to solve the problem. They want an ongoing crisis they can use against Biden. This is all to deny Biden a "win." Their obligations to their constituents and to their oaths evaporate in the face of performance politics.

Speaking of Korean translation and AI, as we did in recent posts (see "Selected readings"), let us take a look at the latest developments in Korea:

New AI-based translation tools make their way into everyday life in Korea

AI equipped with natural language processing software, which allows it to interpret human language in various contexts, is gaining the most attraction among mainstream users among all AI services

Jung Yu-gyung, Hankyoreh (2024-04-23)

More and more, AI is becoming a part of daily life:

On Monday, SK Telecom unveiled its AI-based translation program “TransTalker,” which offers real-time interpretation for 13 languages, including Arabic, Russian, Vietnamese and Indonesian. Lotte began testing the translation service on Friday through its information desks on the first floor of Lotte Department Store's Avenuel Jamsil and on the first floor of Lotte World Mall. Both locations receive over a thousand visits from foreign tourists every day. Lotte has reported that the majority of users are surprised at the effectiveness and clarity of the interpretative service.