Reads

Posted by Victor Mair

https://languagelog.ldc.upenn.edu/nll/?p=70937&utm_source=rss&utm_medium=rss&utm_campaign=the-many-myths-about-the-chinese-typewriter

https://languagelog.ldc.upenn.edu/nll/?p=70937

Julesy tells it like it is: "The Impossible Chinese Typewriter"

In less than 20 minutes, Julesy gives us a more accurate and complete introduction to the history, nature, and workings of the Chinese typewriter than a couple of recent authors specializing on the Chinese typewriter do in hundreds of pages.  Unfortunately, they have reaped an enormous amount of publicity, which means that the American public (and the English-speaking public worldwide) have gotten a terribly distorted understanding of what the Chinese typewriter is all about.  (See J. Marshall Unger, "Triple review of books on characters and computers" [8/23/24].  Although the book by Uluğ Kuzuoğlu is not exclusively about the typewriter, he has the most sensible things to say about the Chinese writing system and information processing that are relevant to Chinese typewriting.)  Therefore, one can only hope that this concentrated video presentation by Julesy will dispel the misapprehensions of the popular publications about the Chinese typewriter of recent years.

It's nice to have Julesy making these excellent videos about Chinese language and script.  I usually agree with practically everything she says, and the bulk of this presentation is no different, except for the rushed ending (I think she wanted to finish within 20 minutes; she's a real pro, and completes the video with 4 seconds to spare!).

During the last two minutes, she says that, in the 1800s and 1900s, when so many revolutions in communications and information processing were taking place — telegraph, typewriter, and computer, all of them tailored to the alphabet — many people believed that the Chinese writing system was inimical to such modern developments and that it was on its way out.  We have had numerous posts about script reformers who advocated that, unless characters be dropped altogether in favor of the phoneticization (Latin-based) of Chinese writing, the country would completely perish in the modern world, and Julesy alludes to that trend in her closing remarks.

Here's her final sentence:  "But it's those, like the protagonists of today's story, firm believers of the Chinese language, that helped Chinese characters survive to this very day, and, for that, I couldn't be more grateful."  I think she's being disingenuous, particularly in the concluding clause, both because it runs counter to the rest of what she carefully lays out in this video and the sum total of the rest of what she says in her videos, as well as because of the slightly evasive way she says it.

In fact, it amounts to an affirmation of the profound truism that William Hannas pointed out in his linguistic treatises (see in the "Selected readings" below), namely, that the makeshift computerization of Chinese characters bought time for them to survive beyond what the script reformers had projected.  In other words, computers ironically staved off the ultimate demise of the antiquated, archaic logographic writing system of sinographs (cf. Sumero-Akkadian cuneiform, Egyptian hieroglyphs, Tangut syllabographs, Vietnamese Chữ Nôm, etc.).

Lin Yutang (1895-1976) was an outstanding writer and brilliant humorist, but his vaunted MingKwai (míngkuài 明快 ["clear and quick"], a reputation to which Lin ardently aspired for it, but was light years away from ever achieving) typewriter was doomed from the start.  Existentially and conceptually, the Chinese writing system, with all of the stupendous challenges and obstacles it presents to mechanical simplification so eloquently and elegantly pointed out by Julesy in this video, simply cannot be effectively and efficiently reduced to an object the size of a Remington or Olivetti typewriter.  The MingKwai typewriter weighs 50 pounds (NYT {7/24/25]) and measures 36 cm × 46 cm × 23 cm (14.2 in × 18.1 in × 9.1 in) (source).  It was a clunker.  A portable Remington typewriter, depending on the model, weights from less than 16 pounds to 20-25 pounds, while a desktop model might weigh around 30 pounds, and they were solid and steady.

When I went to college, my father gave me the precious gift of an Olivetti portable typewriter that weighed 8.2 pounds and measured 27x37x8 cm.  It was a thing of beauty.  I carried it with me up into the mountains of Nepal for two years, then across the seas to Taiwan for two years, then to graduate school in Boston.  I kept using it until I got an Apple Macintosh Portable in 1989.  It was heavy, especially with its big battery, weighing about twice as much as the Olivetti, but I took it with me even into the deserts of Central Asia.  I'll never forget crossing the sandy border between Kazkhstan and China, with miles of barbed wire and a guard with a machine gun accompanying the passengers on the small public bus I took from Almaty to Ürümqi, clutching that big, black computer case all the way.

Lin bankrupted himself over the MingKwai, but its sole useful legacy was to promote separation of lookup and entry from clumsy methods such as radicals plus residual strokes and total stroke counts and replace them with a more spelling-like analysis of different types of strokes.  That was only a stopgap at best.  The verdict is already in.  The vast majority of entry, lookup, ordering, and grouping / organizing of character-based material is done through romanization.

A byproduct of Lin's research on the MingKwai typewriter, whose finding system is based on the principles he developed for the designation of individual characters:  top left, bottom right (that's a simplification of the many other niceties for mastering the system) was his Chinese-English Dictionary of Modern Usage (1972).  In terms of content, translation equivalents, grammar, usage notes, sensitivity to colloquialisms, semantics, and so forth, this is a good dictionary, as one would expect from a highly educated bilingual linguist.  It incorporated a lot of the invaluable scholarship that went into the superb Gwoyeu Tsyrdean 國語辭典 (Dictionary of the National Language), comp. Wang Yi (1937-1945), which I still frequently use.  Despite its strengths in other respects, almost no one took the trouble to learn the two-corner "Instant Index System", relying instead on the supplementary alphabetical list of characters that was published in 1976. 

The four-corner method or four-corner system (sì jiǎo hàomǎ jiǎnzì fǎ 四角號碼檢字法 ["four corner code lookup character method"]) is a character-input (and finding) method used for encoding Chinese characters into either a computer or a manual typewriter, using four or five numerical digits per character. 

The four-corner method was invented in the 1920s by Wang Yunwu, the editor in chief at Commercial Press Ltd, preceding Lin Yutang's MingKwai typewriter and his Chinese-English Dictionary of Modern Usage.  Its original purpose was to aid telegraphers in looking up Chinese telegraph code numbers in use at that time from long lists of characters. This was mentioned by Wang Yunwu in an introductory pamphlet called Four-Corner Method, published in 1926. Cai Yuanpei and Hu Shih wrote introductory essays for this pamphlet.

The four-corner method itself was inspired by the seminal study and system of Russian Orientalist, Sinologist, and scholar of Japanese Studies, Otto Julius Rosenberg (1888-1919) on a system of classification of Chinese characters by the shapes of their strokes at the top left and bottom right corners. Based on ideas of Russian Sinologist and Buddhologist Vasily Vasilyev (1818-1900), Rosenberg developed a method of classification of Chinese and Japanese characters. He published the results of his research in a dictionary in English and Japanese and which formed the basis of the so-called "Four Corner Method", which is still employed today in the creation of Chinese and Japanese dictionaries. This volume is extremely rare and, according to the world's largest bibliographic database, WorldCat, it is only found in six libraries worldwide. While a graduate student, I was privileged to use a copy held by the Harvard-Yenching Library.

For more information about Otto Rosenberg, see John [S.] Barlow, "The Mysterious Case of the Brilliant Young Russian Orientalist… – part 3 and finish", Bulletin of the IAO (International Association of Orientalist Librarians), Vol. 44 (2000).  Barlow, a medical doctor with a deep interest in brain research with a focus on electrophysiological function, practicing what he preached about Chines lexicography, compiled this very impresive dictionary: A Chinese-Russian-English Dictionary Arranged by the Rosenberg Graphical System (Mudrov's Chinese-Russian Dictionary with an English Text and Appendices) (Honolulu:  University of Hawaii Press, 1995), xxiii, 830 pp.

The Rosenberg lookup system was also adopted by the monumental Chinese-Russian dictionary in 4 large volumes (1983-84) edited by Ilya Mikhailovich Oshanin.  The first volume contains three huge indices for looking up the characters by Kangxi radicals and residual strokes, by the four-corner method, and pinyin, in addition to the shape-based stroke system of Rosenberg, the key to which is found on each page of the other three large volumes which contain approximately 250,000 terms.  If I can't find a colloquial or classical term in one of the other dictionaries that I usually rely on, this is where I turn next.

Thus, Lin Yutang's MingKwai was the heir to a Russian lexicographical ordering system for Chinese characters that is still in use today, but not in China.

Selected readings

Julesy videos

• "Fundamental Sinitic linguistic issues solved through analysis of Chinese rap" (8/13/25)
• "Tones and intonation in Sinitic languages" (53025)
• "The conundrum of singing with tones" (5/30/25)
• "Japanese lexical influence on other East Asian languages" (7/23/25)
• "Words, morphemes, collocations, characters" (7/23/25)

Previous Language Log posts on the Chinese typewriter

• "Chinese Typewriter" (6/30/09)
• "Chinese typewriter, part 2" (4/17/11) — featuring the old Chinese typewriter at the University of Kansas
• "Chinese typewriter redux" (11/5/16) — this and the preceding two posts all talk about my favorite M. C. Hammer performance, "Chinese Typewriter Dance" ("U Can't Touch This")
• "Another chapter in the history of the Chinese typewriter" (8/14/21)

The Chinese writing system

• "How many more Chinese characters are needed?" (10/25/16)
• "Writing characters and writing letters" (11/7/18)
• "Sinological suffering" (3/31/17) — the agony of looking up characters in a Chinese dictionary; see especially my remarks about the Russian lookup system in the  Bol'shoi Kitaisko-Russkii Slovar'.
• Victor H. Mair, "The Need for an Alphabetically Arranged General Usage Dictionary of Mandarin Chinese: A Review Article of Some Recent Dictionaries and Current Lexicographical Projects" (pdf), Sino-Platonic Papers, 1 (February, 1986), 1-31.
• "Grids galore" (11/19/23) — brilliant post that is essential for anyone who wants to understand the nature and construction of Chinese characters, though many of the comments wander off in aimless / fruitless directions.
• J. Marshall Unger, "Handling Chinese Characters on Computers:  Three Recent Studies", Sino-Platonic Papers, 354 (August, 2024), 1-18.  (free pdf) — of the three books reviewed by the author, the one by Uluğ Kuzuoğlu gives the clearest headed account Sinographic issues related to the Chinese typewriter.
• ________, Ideogram:  Chinese Characters and the Myth of Disembodied Meaning (Honolulu:  University of Hawai'i Press, 2004)
• William C. Hannas, Asia’s Orthographic Dilemma (Honolulu, Hawai’i: University of Hawai’i Press, 1997),
• ________, The Writing on the Wall: How Asian Orthography Curbs Creativity (Philadelphia: University of Pennsylvania Press, 2003).

https://languagelog.ldc.upenn.edu/nll/?p=70937&utm_source=rss&utm_medium=rss&utm_campaign=the-many-myths-about-the-chinese-typewriter

https://languagelog.ldc.upenn.edu/nll/?p=70937

Posted by Mark Liberman

https://languagelog.ldc.upenn.edu/nll/?p=70961&utm_source=rss&utm_medium=rss&utm_campaign=gpt-5-can-parse-headlines

https://languagelog.ldc.upenn.edu/nll/?p=70961

Philip Taylor sent a link to this Guardian article "West Point cancels ceremony to honor Tom Hanks as ‘outstanding US citizen’", with the comment

It was only on reading the article that I realised that West Point was/were not cancelling the ceremony in order to honour Tom Hanks (as I had originally thought/believed) but were in fact cancelling a ceremony intended to honour Tom Hanks …

I've been meaning to test GPT-5's parsing ability, ever since I discovered its surprising ability to represent semantic scope ambiguities in correct predicate logic (see "No X is better than Y", 8/13/2025, and the details of its analyses).

So I simplified the Guardian's headline slightly, in order to make the analyses fit the page better, and tried it.

Me: Draw a parse tree for the sentence "They cancelled the ceremony to honor Tom Hanks"

GPT-5:

Thought for 25s

Great sentence—there’s a classic attachment ambiguity.

Here are two clean parse trees (Penn-Treebank style with ASCII):

#1 NP-attachment (most natural: the ceremony’s purpose is to honor Tom Hanks)

S
├── NP
│   └── PRP  They
└── VP
    ├── VBD  cancelled
    └── NP
        ├── DT   the
        ├── NN   ceremony
        └── S    (non-finite modifier of N; “reduced relative/purpose”)
            └── VP
                ├── TO   to
                └── VP
                    ├── VB   honor
                    └── NP
                        ├── NNP  Tom
                        └── NNP  Hanks

Bracketed:
(S
  (NP (PRP They))
  (VP (VBD cancelled)
    (NP (DT the) (NN ceremony)
      (S (VP (TO to) (VP (VB honor) (NP (NNP Tom) (NNP Hanks))))))))

#2 VP-attachment (less plausible meaning: they cancelled, in order to honor Tom Hanks)

S
├── NP
│   └── PRP  They
└── VP
    ├── VBD  cancelled
    ├── NP
    │   ├── DT  the
    │   └── NN  ceremony
    └── S    (purpose clause modifying the VP)
        └── VP
            ├── TO  to
            └── VP
                ├── VB  honor
                └── NP
                    ├── NNP Tom
                    └── NNP Hanks
Bracketed:
(S
  (NP (PRP They))
  (VP (VBD cancelled)
      (NP (DT the) (NN ceremony))
      (S (VP (TO to) (VP (VB honor) (NP (NNP Tom) (NNP Hanks)))))))

Entirely correct! (or at least close enough…)

See the full conversation, where I go on to ask for for dependency-grammar versions, which GPT-5 gets formally correct but is not very good at drawing. I didn't try to get GPT-5 to draw pretty syntax trees, rather than the ascii and bracketed versions, so I did it myself via jsSyntaxTree for those who might care.  The NP-attachment version:

And the VP-attachment version:

So it seem that GPT-5 can analyze (some instances of) syntactic as well as semantic ambiguity, and can use Penn Treebank and Universal Dependency representations correctly to describe the syntactic alternatives. This deserves more investigation than I have time for this morning, including exploration of two obvious questions:

• Is GPT-5 recognizing when syntactic and logical analyses are needed, and turning the task over to special "experts"? If so, how do those subsystems work?
• Does GPT-5 use such analyses in its own training and question-answering? If so, that's quite different from just predicting the next token in a text sequence. (Though thinking for 25 seconds about the analysis of one sentences suggests that its probably not yet doing this to much of its training text…)

https://languagelog.ldc.upenn.edu/nll/?p=70961&utm_source=rss&utm_medium=rss&utm_campaign=gpt-5-can-parse-headlines

https://languagelog.ldc.upenn.edu/nll/?p=70961

Posted by Mark Liberman

https://languagelog.ldc.upenn.edu/nll/?p=70939&utm_source=rss&utm_medium=rss&utm_campaign=is-the-decline-of-writing-making-journalism-dumber

https://languagelog.ldc.upenn.edu/nll/?p=70939

No. At least, there've been plenty of dumb articles over past decades and centuries, and plenty of smart ones recently. But I have some complaints about one particular recent article in The Economist, "Is the decline of reading making politics dumber? As people read less they think less clearly, scholars fear", 9/4/2025.

I should start by saying that the quality of articles in The Economist is generally very high, in my opinion, and its articles about language are especially good. So why was I disappointed in this one?

Here are its first two paragraphs:

The experiment was simple; so too, you may have thought, was the task. Students of literature at two American universities were given the first paragraphs of “Bleak House” by Charles Dickens and asked to read and then explain them. In other words: some students reading English literature were asked to read some English literature from the mid-19th century. How hard could it be?

Very, it turns out. The students were flummoxed by legal language and baffled by metaphor. A Dickensian description of fog left them totally fogged. They could not grasp basic vocabulary: one student thought that when a man was said to have “whiskers” it meant he was “in a room with an animal I think…A cat?” The problem was less that these students of literature were not literary and more that they were barely even literate.

My first complaint: there's no  link to the referenced experiment. We're not even given the title of the publication documenting it, or the names of its authors.

Here's why that matters. Internet search reveals what the publication was: Susan Carlson et al., "They Don’t Read Very Well: A Study of the Reading Comprehension Skills of English Majors at Two Midwestern Universities", CEA Critic 2024. And checking that publication reveals several relevant facts:

• Although the study was published in March 2024, the study was done in January to April of 2015, more than 10 years ago.
• The 85 subjects in the study came from two Kansas regional universities.
• Their average ACT Reading score was 22.4, which is "low intermediate level",
• The authors divided the subjects' Bleak House explanations into three categories: problematic, competent, and proficient.
• Their discussion focused on the students in the "problematic" category: 49 of 85.

In other words, they discuss the worst students in a sample with low scores to start with.

Why did they do that? As they explain,

The 85 subjects in our test group came to college with an average ACT Reading score of 22.4, which means, according to Educational Testing Service, that they read on a “low-intermediate level,” able to answer only about 60 percent of the questions correctly and usually able only to “infer the main ideas or purpose of straightforward paragraphs in uncomplicated literary narratives,” “locate important details in uncomplicated passages” and “make simple inferences about how details are used in passages”.  In other words, the majority of this group did not enter college with the proficient-prose reading level necessary to read Bleak House or similar texts in the literary canon. As faculty, we often assume that the students learn to read at this level on their own, after they take classes that teach literary analysis of assigned literary texts. Our study was designed to test this assumption.

So the study was designed to test the university and its faculty, not the students. The conclusion, basically, is that these students entered the university incapable of reading canonical literature; the university and its faculty failed to fix the problem; and the students didn't fix the problem on their own.

I'm not convinced that being able to read and understand the first seven paragraphs of Bleak House is an appropriate measure for the reading ability of modern American youth. That novel's many words and phrases from the 19th-century British court system make it hard for a modern American reader to grasp the context. I'd be more impressed if the students failed to understand the start of Great Expectations, Emma, Gulliver's Travels, Jane Eyre,  Tom Sawyer,  Alice's Adventures in Wonderland or etc.

But let's grant that Carlson et al. have proved their point, and just note that The Economist's writer badly mis-read (or maybe mis-represented?) their work, by presenting it as evidence that today's university-level literature students can't read Dickens.

My second complaint is that The Economist's writer goes on to use the Flesch-Kincaid readability measure:

We also analysed almost 250 years of inaugural presidential addresses using the Flesch-Kincaid readability test. George Washington’s scored 28.7, denoting postgraduate level, while Donald Trump’s came in at 9.4, the reading level of a high-schooler.

See my 2015 post "More Flesch-Kincaid grade-level nonsense", which points out that different choices of punctuation strongly modulate the Flesch-Kincaid index, as in this example from one of Donald Trump's speeches, which was used in a stupid newspaper article to prove that Trump operates at a 4th grade level:

It’s coming from more than Mexico. It’s coming from all over South and Latin America. And it’s coming probably — probably — from the Middle East. But we don’t know. Because we have no protection and we have no competence, we don’t know what’s happening. And it’s got to stop and it’s got to stop fast. [Grade level 4.4]

It’s coming from more than Mexico, it’s coming from all over South and Latin America, and it’s coming probably — probably — from the Middle East. But we don’t know, because we have no protection and we have no competence, we don’t know what’s happening. And it’s got to stop and it’s got to stop fast. [Grade level 8.5]

It’s coming from more than Mexico, it’s coming from all over South and Latin America, and it’s coming probably — probably — from the Middle East; but we don’t know, because we have no protection and we have no competence, we don’t know what’s happening. And it’s got to stop and it’s got to stop fast. [Grade level 12.5]

That post closes this way:

It's uncharitable and unfair of me to imply that the author of the Globe piece might be "stupid". But at some point, journalists should look behind the label to see what a metric like "the Flesch-Kincaid score" really is, and ask themselves whether invoking it is adding anything to their analysis except for a false facade of scientism.

That's enough complaining for now. But since The Economist's article also frets about secular changes in sentence length, let me refer interested readers to the slides for my talk at the 2022 SHEL ("Studies on the History of the English Language") conference.

https://languagelog.ldc.upenn.edu/nll/?p=70939&utm_source=rss&utm_medium=rss&utm_campaign=is-the-decline-of-writing-making-journalism-dumber

https://languagelog.ldc.upenn.edu/nll/?p=70939