Language Log

With this post, we are already acquainted with Inspur's Yuan 1.0, "one of the most advanced deep learning language models that can generate coherent Chinese texts."  Now, with the present article, we will delve more deeply into the potentials and pitfalls of Inspur's deep learning language model:

"Inspur unveils GPT-3 equivalent for Chinese language", by Wei Sheng, TechNode (1026/21)

The model is trained with 245.7 billion parameters—the number of weights in an artificial neural network, according to the company. This is more than the Elon Musk-backed GPT-3 language model for English, which has 175 billion parameters. Inspur said the Yuan model was trained with 5 terabytes of datasets.

…

New article in EnterpriseAI (October 21, 2021):

"Language Model Training Gets Another Player: Inspur AI Research Unveils Yuan 1.0",  by Todd R. Weiss

From Pranav Mulgund:

Valerie Hansen is Director of Undergraduate Studies for East Asian Studies at Yale.  Yesterday she was talking to a sophomore who had taken 1st and 2nd year Mandarin online and is about to start 3rd year.  Valerie writes:

After a while, she told me that she did have one worry about taking 3rd year: she had never written a single character and she wondered if her teacher would expect her to know how to write characters.

Brian Merriman ran into this article and device when researching electronic typewriters from the 1980s:

Article by Holly Else in Nature (8/5/21):

"‘Tortured phrases’ give away fabricated research papers

Analysis reveals that strange turns of phrase may indicate foul play in science"

Here are the beginning and a few other selected portions of the article:

In April 2021, a series of strange phrases in journal articles piqued the interest of a group of computer scientists. The researchers could not understand why researchers would use the terms ‘counterfeit consciousness’, ‘profound neural organization’ and ‘colossal information’ in place of the more widely recognized terms ‘artificial intelligence’, ‘deep neural network’ and ‘big data’.

Further investigation revealed that these strange terms — which they dub “tortured phrases” — are probably the result of automated translation or software that attempts to disguise plagiarism. And they seem to be rife in computer-science papers.

Research-integrity sleuths say that Cabanac* and his colleagues have uncovered a new type of fabricated research paper, and that their work, posted in a preprint on arXiv on 12 July¹, might expose only the tip of the iceberg when it comes to the literature affected.

[*VHM:  Guillaume Cabanac, a computer scientist at the University of Toulouse, France]

Phil H wrote these comments to "Uncommon words of anguish" (7/18/21):

The anguish is very real. My wife had a character in her name that most computers will not reproduce ([石羡]), despite it being relatively common in names in our part of the world, and has been refused bank accounts, credit cards, and a mortgage because of it. In the end she changed her name rather than continue to deal with the hassle. The character is in the standard, but it was too late for us.

…there have always been ways to get the character onto a computer, but any given piece of bank software might not recognise it, and any given bank functionary might be unfamiliar with them. We then had trouble when some organisations used the pinyin XIAN in place of the character, but that then made their documentation inconsistent with her national ID card (which had the right character on it) and so yet further bodies would not accept them… It was the standard "mild computer snafu + large inflexible bureaucracy = major headache" equation.

An anonymous correspondent, a computer scientist, sent in the following remarks:

Phil H is talking about a character which is in a "supplementary plane" in Unicode (and similarly in GB-18030).  Unfortunately, an awful lot of software was only ever tested on Basic Multilingual Plane characters.

From a manual for a thermal printer:

Dǎyìn kòngzhì bǎn nèizhì GB18030 Zhōngwén zìkù, chèdǐ miǎnchú shēngpì zì de kǔnǎo

打印控制板内置 GB18030 中文字库,彻底免除生僻字的苦恼

Printer control panel built-in GB18030 Chinese character, thoroughly remove the uncommon words of anguish

(courtesy of Amy de Buitléir)

A more accurate English translation would be:

Printer control panel with built-in GB18030 Chinese character font, thoroughly removing the anguish brought about by uncommon / obscure characters

"GB" stands for "guóbiāo 国标" ("national standard"), and is used for many technical terms in the PRC (another instance of encroaching digraphia, for which see here and here [with extensive bibliography]).

生育还是生肓？ pic.twitter.com/GBPx07QSaT

— Chenyu_Liang (@chenyuliang) May 31, 2021

Article in official CCP media source:

"Chinese lawmaker proposes removing English as core subject"

By Liu Caiyu, Global Times (3/5/21)

Coming from GT, the hyper-nationalistic tabloid, this attack on English is not unexpected, and similar anti-English proposals come up every year around the time of the national meetings of the Liǎnghuì 兩會 (Two Sessions), annual plenary meetings of the national People's Congress and the national committee of the Chinese People's Political Consultative Conference that have just concluded in Beijing (March 4-11).

Here we go again:

Is English really that important? A Chinese lawmaker at the two sessions has proposed removing English as a core subject for Chinese students receiving compulsory education, triggering heated discussion on Chinese social media.

The proposal was made by Xu Jin, a member of the Central Committee of the Jiusan Society and also a member of the Chinese People's Political Consultative Conference (CPPCC). It has also been proposed by other lawmakers in previous years.

Last week in our Dunhuangology seminar, a student wanted to type "wǔ 武" ("martial; military") into the chat box, but instead out popped "nián 年" ("year").  I immediately said to her, "I'll bet you were using a shape-based inputting system", which left her a bit surprised.

Ever since information technologists began to wrestle with the problem of inputting, ordering, and retrieving Chinese characters in computers during the 70s, I have been intensely interested in the theoretical and practical obstacles they faced.  To better understand the overall situation with regard to characters in computers, I organized an international conference at Penn in 1990 on the computerization of Chinese characters that resulted in Victor H. Mair and Yongquan Liu, eds., Characters and Computers (Amsterdam, Oxford, Washington, Tokyo:  IOS, 1991).

Ben Hull writes:

In our Computational Linguistics class we were discussing different methods of segmenting Chinese character texts. Today I came across a terrific example of the problems of segmenting left to right, in the first sentence of the attached image. I hope you find it as amusing as I did.

This has been making the rounds:

1. Go to Google Translate.
2. Set the input language to Spanish.
3. Paste in "soy milk"
4. Set the output language to English or X language.
5. Hilarity ensues.

The obligatory screen shot:

This is all over the Chinese internet:

(source)