Language Log

The Large Language Model DistilBert is "a distilled version of BERT: smaller, faster, cheaper and lighter".

A trained DistilBert model is available from Hugging Face, and recommended applications include "text classification", with the featured application being "sentiment analysis":

And as with many similar applications, it's been noted that this version of "sentiment analysis" has picked up lots of (sometimes unexpected?) biases from its training material, like strong preferences among types of ethnic food.

This is a problem that has been troubling colleagues across the country.

"Why fewer university students are studying Mandarin"

Learning the difficult language does not seem as worthwhile as it once did

Economist (Aug 24th 2023)

China | How do you say “not interested”?

Ten years ago Mandarin, the mother tongue of most Chinese, was being hyped as the language of the future. In 2015 the administration of Barack Obama called for 1m primary- and secondary-school students in America to learn it by 2020. In 2016 Britain followed suit, encouraging kids to study “one of the most important languages for the UK’s future prosperity”. Elsewhere, too, there seemed to be a growing interest in Mandarin, as China’s influence and economic heft increased. So why, a decade later, does Mandarin-learning appear to have declined in many places?

"Elusive Ernie: China's new chatbot has a censorship problem"   

By Stephen McDonell, BBC, 1 day ago

It seems that Ernie's favorite response is "Let's talk about something else", particularly when you ask it a "difficult" question.

People, including serious linguists, are beginning to wonder:

John McWhorter, "Are translation apps making the learning of foreign languages obsolete?", NYT 7/25/2023

I remember a time, not too long ago, when John was making a serious effort to learn Mandarin, because he often asked me cogent questions about the language and wanted to know the best methods for learning it.

…

What we can learn from the Tower of Babel

In Europe, nine out of 10 students study a foreign language. In the United States, only one in five do. Between 1997 and 2008, the number of American middle schools offering foreign languages dropped from 75 percent to 58 percent. Between 2009 and 2013, one American college closed its foreign language program; between 2013 and 2017, 651 others did the same.

At first glance, these statistics look like a tragedy. But I am starting to harbor the odd opinion that maybe they are not. What is changing my mind is technology.

From John Rohsenow, via Mabel Menard, comes this bit of Japanglish:

(source)

It's already happening.

Of course, we don't want our students to use AI software to help them write their papers, but the fact of the matter is that some of them, especially among those who have poor English writing skills, are already routinely doing so.  Their attitude seems to be that they will do the basic research and sketching out of the argument, and then they have AI tools make it sound nice.  In some cases, they even ask AI bots to assist them with the data search that goes into their paper.

Some may argue that this is completely unacceptable, that such students should be expelled right away, but where do you draw the line on computer assisted research and writing?  Moreover, this is clearly not the same thing as plagiarism, because the individual who is relying on a chatbot to help him or her is not appropriating the intellectual property of another person.  He/she is utilizing material that he/she specifically requested a machine to create / produce on his / her behalf and at his / her direction.  In other words, the electronic tools are acting as extensions of his / her brain.

Competition among various AI services will spur them to further heights.

ChatGPT, Bing, Bard and DeepL: Which one offers the best Japanese-to-English translation?

by Karin Kaneko, Japan Times (7/18/23)

Kaneko, working with her editors at Japan Times, devised an ingenious test for comparing the quality of several translation tools in different categories of writing.  Since this experiment is so innately interesting and inherently revelatory, I will provide extensive quotations, adding romanization of the Japanese passages from GT (not an easy task for me!).  To be fair to GT, and simply out of curiosity to see how it compares with the newer type of AI translation services, I will also invite GT to translate all three of the chosen passages.  N.B.:  All three of the GT English translations have been added by me.

Whatever that means.

That's what we get when we enter into AI translation software (GT, Baidu, Bing, DeepL) this key term — "双泛" — from this important policy document concerning the governance of Xinjiang issued by the Xinjiang Uygur Autonomous Region Committee of the CCP.

Shuāng 双 is simple:  it means "double".  Fair enough.  But 泛 in this disyllabic expression is notoriously difficult to deal with.  It can be pronounced either fàn, in which case it means  "to float on water; to drift; to spread out; to be suffused with; to flood; to overflow; superficial; non-specific; extensive; general; pan-; careless; reckless", fěng, in which case it means "to turn over; to topple over; to be destroyed; to be defeated; to fall", or fá, in which case it signifies the sound of water.

Or, to put it another way, in the words of Douglas Hofstadter,

Learn a Foreign Language Before It’s Too Late

AI translators may seem wondrous but they also erode a major part of what it is to be human.

The Atlantic (7/13/23)

Hofstadter recounts how he spent years of painstaking, hard labor learning more than half a dozen foreign languages, though he never came close to mastering any of them except French and Italian.

But today we have Google Translate. Today we have DeepL. Today we have ChatGPT—and so on. There’s no need for me to list all the powerful technologies that allow anyone today—a monolingual American, say, who has never devoted a single moment to learning, say, Chinese—to write fluent passages in Chinese. Today it’s a piece of cake to send an email in a tongue you don’t know a word of. You just click on “Translate” and presto! There it is! Or at least, there it is, in a certain sense. Assuming that there are no egregious translational blunders (which there often still are), what you are sending off is slick but soulless text.

Despite the panic over AI we're seeing in many sectors of society, including academia, the juggernaut rolls on, seeming set to crush everything in its way:

"EU gives more power to AI translation machines"

The European Commission has launched a pilot project to translate some press releases without any human oversight.

POLITICO (6/15/23)

Three weekends ago, I delivered a keynote here:

New Directions in Chinese Language Education in the 21st Century

The Eighth International Conference on Teaching Chinese as a Second Language

Swarthmore College, June 9-10, 2023

———–

Abbreviations:

    AI — Artificial Intelligence

    DT — Digital Technology

    IT — Information Technology

    DH — Digital Humanities

    AGI — Artificial General Intelligence, where machines supposedly can accomplish any intellectual task that a human can (to me that's a pipe dream)

(given for present and future reference and use)

Title "Aspects of AI and digital technologies in Chinese language teaching"

Abstract

In recent decades, language processing hardware and software have progressed at an astonishing rate, one that is geometric rather than arithmetic.  The opportunities these advances offer and the challenges they pose require our thoughtful attention and careful response, lest the machines get out of control and affect our students in detrimental ways.  DeepL, ChatGPT, and other constantly evolving technologies possess enormous power to manipulate language, power that we can utilize for the enhancement of Chinese language pedagogy.  On the other hand, we must monitor and adapt this potential in such a manner that it fits our purposes and meets the needs of our students. 

A couple of weeks ago, John Hansen tried "an experiment to see if I could successfully combine random and seemingly unconnected topics into one poem", and reported the results on Medium. This experiment was quickly reproduced by Adrian CDTPPW, Block Wife, and Robert G. Longpré.

The buzz about Large Language Models has re-ignited interest in Alan Turing's famous 1950 article "Computing Machinery and Intelligence". Two interesting recent discussions: Jessica Riskin, "A Sort of Buzzing Inside My Head", NYRB 6/25/2023, and Mustafa Suleyman, "The Coming Wave: Technology, Power, and the Twenty-first Century's Greatest Dilemma", Random House 9/5/2023.

Suleyman's book won't be released until 9/5/2023, so it's interesting that several outlets have blurbed one of its ideas ten weeks early: Brad Stone, "AI Leader Proposes a New Kind of Turing Test for Chatbots", Bloomberg 6/20/2023, and Sawdah Bhaimiya, "DeepMind's co-founder suggested testing an AI chatbot's ability to turn \$100,000 into \$1 million to measure human-like intelligence", Business Insider 6/20/2023.  Based just on Business Insider's title, Suleyman's proposal puzzled me, since we don't usually think of machine-trading systems as measuring intelligence — at least not the intelligence of the system rather than its designer. But in fact Suleyman has something different in mind, more along the lines of  an extended "shark tank" competition:

In describing his proposal, Suleyman argues that there’s a misplaced focus in the tech industry on the distant possibility of achieving artificial general intelligence, or AGI: algorithms with cognitive abilities that match or exceed humans’. Instead, he said the more achievable and meaningful short-term goal is what he calls artificial capable intelligence, or ACI: programs that can set goals and achieve complex tasks with minimal human intervention.

To measure whether a machine has achieved ACI, he describes a “modern Turing test” — a new north star for researchers — in which you give an AI \$100,000 and see if it can turn the seed investment into \$1 million. To do so, the bot must research an e-commerce business opportunity, generate blueprints for a product, find a manufacturer on a site like Alibaba and then sell the item (complete with a written listing description) on Amazon or Walmart.com.

Suleyman expects AI will pass this more practical threshold sometime in the next two years. “We don’t just care about what a machine can say; we also care about what it can do,” he writes. And when that happens, he says, “The consequences for the world economy are seismic.”