This is really a pseudo-liveblog: It was originally written during the event with Lan Li, but grant proposals, end of term events, and similar delayed proofreading and posting until now. Our apologies! Photographs by Lan Li and Amy Johnson.
QWERTY Is Dead! Long Live QWERTY! The Birth of Input in Twentieth-Century China
STS Colloquium 12/9/13 4pm
Speaker: Thomas Mullaney, Stanford
Discussant: Brian Rotman, Ohio State
This is still very much an early draft. Today Tom will run over key points, explain how this fits into his larger book projects, and provide a few examples.
The paper circulated in conjunction with this talk is chapter 8 of Tom’s current book project. Central conceit: The 19th century witnesses the explosion of dramatic, transformative communication technologies. These include telegraphy, typewriters, punch cards, etc. — writing systems based on a radically limited number of letters, syllables, or elements.
This is all well and good for roman alphabets. These technologies can easily be extended from English into Spanish and French — even Cyrillic. Arabic, Hebrew, etc., offer fresh challenges, but none quite at the level of complexity of Chinese.
Information technology is a marriage of input and output systems. There is no such thing as “typing” in China. It doesn’t exist. All that exists is input. Tom defines typing here to mean the unspoken assumption that the depression of a key on a keyboard will result in the near-instantaneous appearance of the corresponding symbol on the screen; that is, symbol on the key will match symbol on the screen. This unity does not exist in China. QWERTY keyboards are used for computing, but no Chinese is visible on the keyboard.
So there’s no matching of key symbol and screen symbol, no immediate appearance on the screen. Tom shows us a keylog of a Chinese headline: gktx zk] vdjii3:dhfgboli clgbji… etc. transforms into Chinese characters: 高铁遭飞鸟撞击：挡风玻璃采购价30万无安全膜. There is no zone of identity, no sameness. And when you don’t assume a correspondence between these two there becomes a lot of room for maneuver and play.
Numerous input editors run on Chinese machines. And there are numerous ways to shape this input experience, both at the level of companies but also at the level of individuals.
The MingKwai (明快 – clear fast) is not the first Chinese typewriter. It’s preceded by traybed Chinese typewriters: Sheffield (1898), Zhou Houkun (1915), Shu Zhendong (1919), Shu 2.0 (1935), and Double Pigeon (1960s). But the MingKwai typewriter works in a way no other keyboard does. The typewriter involves 8 grids, 36 bars, which produces this interior.
Lin, Tom contends, is concerned with retrieval. Lin’s comparative study of looking up words in dictionaries in Chinese and English has him worried: the process takes 1.8 seconds longer in Chinese. This connects to larger fears about the modernity of Chinese, shared with other researchers and intellectuals.
In the MingKwai Lin merges retrieval with inscription, creating what is essentially a retrieval-inscribor machine. Depressing a certain sequence of keys first retrieves the relevant character, and only then inscribes it. This is thus a significantly different form of input than typing. It resonates, too, with an ongoing search for a fundamental order to Chinese. Lin saw this project not just as a theoretical one, but as an ethnographic one — creating, as one advertisement put it, “The Only Chinese Typewriter Designed for Everybody’s Use.”
“One day, a 12-year-old suddenly asked her mother, “What is love?” (How frightening, only 12 and already talking about love!) asking her how to write the character “lian.” At the moment, her mother was in the middle of doing housework, and couldn’t leave her seat. So she explained that the character lian has the character yan in the middle, is flanked on both sides by jiaozi, with one xin on the bottom. As soon as [the daughter] heard it, she immediately wrote without the slightest error. This is the TK of the structure of our writing, and a point which character retrieval researchers should play close attention to. We have never heard of describing the character lian as “top-left corner 0, top-right corner 0, bottom-left corner 3, bottom-right corner 3, 0033 is lian. (Du Dingyou 1925).”
We don’t yet know, Tom suggests, what might be the implications of input as a communicative practice.
Brian offers a brief summary of the paper: For Chinese, all text input is a kind of telecommunication, or communication at a distance from oneself. The user produces characters through variable and individualized middleware. The grand ethnographic claim by Lin was that the MingKwai was the only typewriter that every Chinese could use, expressing an ethnographic desire to universalize literacy in a new way.
What makes the MingKwai so fascinating is that its construction suggests the involvement of a model of predictive text. Its design seems to have been shaped by an idea of what kinds of texts it would be used for.
Brian underscores the moral panic, the fear of backwardness, that inspired comparative measurements of the speed of dictionary use — and how this was connected to larger concerns in the early 20th century about the vapidness of Chinese conceptualizations of the world in the face of a massively powerful capitalist West.
Jiang Yiqian saw attempts to solve the character retrieval problem as a kind of Darwinian struggle between systems. (Many intellectuals in Republican China were influenced by Social Darwinism.) But what’s interesting here is that the system works not with the semantics of the word, but with its form.
This was also driven in part by militaristic need for a simplified form of telegraphy. Brian asks if this system still exists as an alternate form to pinyin (the official phonetic system for transcribing Chinese).
Tom explains that this system has been incorporated as a 5-stroke IME system, which can be used by people not familiar with pinyin. It requires the standard dialect input that even people who speak Cantonese can use. This input system is part of a larger story of standard Chinese and its language ideologies and language planning; e.g.,. the discouraging of televisions shows and other expressive forms in dialect. This input system plays a structural role.
Du Dingyao’s system was completely formalist and was also prefaced by an ethnographic initiative: it sought the kinds of characters that Chinese speakers could understand and organized these into eight types by form, not by meaning. Brian notes that at the same time, David Hilbert was developing a formalist mathematical program, which involved treating mathematical formulae as nothing more than inscriptions, or “meaningless marks on paper” — an important project for mathematical logic and systems.
Brian identifies three kinds of systems in play here: IMEs, Lin Yutang’s typewriter, and traybed moveable typewriters. Traybed typewriters are described in one of Tom’s others papers, in which he examines how Chinese typists developed a predictive-text typewriter during Maoist China (excerpt).
The tray was designed with a geometry of overlapping clusters based on frequency correlations. For instance, in the obligatory rhetoric of the day, “Mao” would be followed by “chairman”; the single character of “Mao” would be surrounded by eight such “related” characters. This is similar to Lin Yutang’s typewriter, in that it attempted to reduce the printer’s labor. “Natural language” was thus brought into the printing process through frequency assessments.
This required profoundly subtle memory practices, a “psychocorporeal” prediction of the next character. Machines were redesigned to resemble humans who themselves embodied machines which were designed to resemble humans… and so on into recursive infinitum.
Brian wants to pull out three points/questions:
1. Are all forms of Chinese communication practices forms of telecommunication? Or just IMEs? Aren’t all forms of writing telecommunication? What about when one writes a shopping list or diary? Isn’t that telecommunication as well? Shopping lists are already themselves code, thanks to abbreviations and appalling handwriting. Brian suggests that there are actually two different writing systems actually going on.
2. Is “coding” really a good word here? What is the relation between Chinese and Western coding practices? Aren’t Western practices deeply alphabetic? How might Chinese coding be different? What is Chinese coding? (Tom explains there is no history of coding in Chinese.) Is this a fruitless search for the ideal input system? An optimal IME?
3. Is predictive text somehow “natural” to the Chinese problem? The IME is predictive on two levels: first with regard to the pinyin, and second in the middleware where the knowledge the predictive ware requires is involved in Lin Yutang’s design. So there must be something about the Chinese view of the problem that makes predictive text come out quite early as part of the solution?
Q & A
The floor is opened for questions, which Tom will tackle in clusters.
Chris Leighton asks how the politics of language and literacy affects understandings of both Chinese and “the everyman.” It seems like the development of the MingKwai occurs in a particularly interesting historical period, in which illiterate people are involved in the production of writing, yet in a system that assumes literacy. What happens to others who were previously involved in such production, but can’t read?
Q: Is it possible to be semi-literate and how does that fit into this picture? It can be possible to recognize something without being able to write it yourself.
Q: Was it helpful to have come to literacy later in life? If you think about how Chinese men of this time period were made literate — through rote memorization — then thinking about a character as a cognitive whole perhaps makes sense. But if you’re a woman — or anthropologist who comes to Chinese later — perhaps you can think of it more as configurable components, with a different solution.
Tom begins with Chris’s question. Each of these retrieval designers in Republican China knows that they are at a disadvantage compared to alphabetical language. There is an orthographic crisis and an ethnographic crisis. The designers see themselves as pursuing the fundamental order of Chinese, searching for Chinese’s metaphorical alphabet. The ethnographic crisis is that the Kangxi dictionary can’t be that alphabet. The Chinese masses who are newly enrolled in the education system are completely confused. The Kangxi System is an elitist linguistic system. This search for linguistic truth — the fundamental order of Chinese — is thus also an ethnographic truth.
Lin Yutang’s system is optimistic about the abilities of the average Chinese user. The user is implicit in the system, which informs the design. For instance, Du Dingyou’s system draws on gestalt theory, assuming that ultimately we are all pattern finders, and thus characters should be thought of as “entities in space.” So when users see 树, they see a whole.
All of these systems except for one is based on the character as pedagogically constructed. It’s only much later, in the 1950s or so, that you start to see the proliferation of phonetic-based phone registers, etc. Without question, engagement with this information system is a different kind of engagement. Its architecture is significantly different, not just in terms of the phonetic character but also its reliance on another symbolic system (the roman alphabet). Moving from graphic to phonetic is a major shift.
In addressing Brian’s questions from the commentary, Tom doesn’t consider Lin Yutang’s machine a predictive-text system. Every act of inscription is a singular, contained unit. In contrast, the typists working on traybeds built archipelagos of language — they were very much concerned with the next character. In the 1970s you start to see the merging of these two. (QWERTY, on the other hand, is a late 80s/90s story.) In 1975, cathode-ray tube systems, the very next word was presented to you. This becomes an explosive mix of inputs and “natural language” database systems.
Q: Lin Yutang proposed three systems. How did the other two differ from this one you’ve been focusing on? And in looking at the inscriptions on the image of the MingKwai, it seems the radical for ming may include a typo?
Tom explains that Lin Yutang was notoriously bad at writing Chinese. But it may just be an artifact of the scan, rather than an actual typo.
Dave Kaiser asks, Why the explosion of concerns? Is this mounting over millennia of dynasties, etc.? Are these only the latest problems of something that has been recurring over history? Or is this somehow related to modernity? Is this a quantitative or qualitative shift?
Chris Leighton asks what font the typewriters work in. They can only be set in one — what does this font tell us about the way designers imagined the typewriter?
In addressing this cluster of questions, Tom explains that Lin Yutang wasn’t thinking about typewriting, but about organizing dictionaries. Think about the materiality of a dictionary: Some categories of radicals are larger than others. For a dictionary, it doesn’t matter if more words start with A than Z, or with one character than another. But in the 1930s and 1940s, when Lin is thinking about dictionary retrieval, there’s a new kind of metal materiality that moves — that isn’t a bound book. As a result, the taxonomy needs to behave differently, too. Now a category isn’t a written radical, but a piece of metal in a region of the machine.
To address this, Lin tries to even out the system. He also literally goes back to the drawing board. Tom interviewed one of Lin’s summer interns who spent the summer drawing shapes in order to create symbols that approximated several symbols with similar visualizations, merging them into a new symbol. He was fighting within a new set of parameters.
There isn’t a more optimal way of doing this. Lin has chosen the optimal number of keys to perform the character selection, if you hold the eight constant. Lin’s starting to think of his taxonomy as moving from one condition of materiality to another.
QWERTY keyboards as they are used by different social actors are fascinating. Programming use is probably the most challenging: ctrl-v, ctrl-c, etc. somewhat approximate this kind of input feature. But remember the scale: We’re talking about every computer, being used by everyone in China. So perhaps “coding” is a good term after all.
We still lack the terminological language to talk about input, in terms of its media effects, semiotic effects, etc. And yet it’s been happening on a phenomenal scale for decades.
Regarding the periodization: The orthographic essence of characters, for example, is a matter of long-standing debate. Preeminent figures in calligraphy identify eight fundamental strokes. Other treatises claim thirty-six essential strokes. The Kangxi system breaks characters into 214 radicals; this only dates back to the late Ming. Others still include 300 or 500 categories. This is an evolutionary experiment between competing retrieval systems. Not only will a system yield faster library card catalogs, phone registers, search functions, but also a fundamental taxonomic structure to the language.
While there are important cases in the Qing and Ming dynasties and earlier, there isn’t the same simultaneity, intensity, and quality of anxious pursuit of language reform. There aren’t the same anxieties over mass literacy. Do you get rid of the entire writing system? How do we build a typewriter, linotype, monotype? Develop a telegraphy system? Certainly this is a moment of challenges occurring at all levels.
This doesn’t happen in quite the same way in Japan because although they have kanji typewriters that look exactly the same as Chinese typewriters (seen in a slide), at the same time, Remington and Underwood are producing Japanese kana typewriters. There is this alphanumeric hegemony.
For comparisons you might want to look at India, Syan, Burma, Thailand, which involve other languages and scripts that befuddle the huge producers of typewriters.
Q: Is there a fundamental cognitive difference between using alphanumeric keyboards and using pictographic characters? This is in reference to the work of Bill Wang (?) and differences in the brains of native speakers who learned to write before learning a phonetic system.
Brian recalls London’s black cabs. Drivers must be able to identify routes to random destinations from anywhere in London. Drivers show enormous amygdalatic function of massive rote learning with thousands of possibilities and without design. Brian suggests that erhaps there are different kinds of neurological imprinting.
Mike Fischer builds on Brian’s question about competing systems. What role does market share play in the stabilization or non-stabilization of systems? And how do you make sense of the transitions between these different kinds of typewriters to contemporary computers and predictive text? Is there an equivalent to predictive text systems of spellcheck, and how is that handled in Chinese?
Tom: In response to one of the earlier questions, there are thousands of kinds of IMEs. Each year hundreds of patents are filed for IMEs. These filers aren’t programmers or Tsinghua engineers, but descendents of character retrievalists; e.g., a WWII veteran who is 90 years old. They think the orthographic or ethnographic question hasn’t been answered. Speed doesn’t matter, it’s about being in possession of an essential truth about the Chinese language or writing. This is an epic mission. In one interview, someone told him the fundamental organization of a particular dictionary was entirely wrong.
Tom is thinking about doing an ethnography dedicated to input.
With regard to font: This also resonates with dictionaries for Hindi or Urdu or Arabic. For these designers are the ones building and sketching letter forms and trying to think about whether these script forms will be taken the right way. In the Chinese case, the Song-ti wins.
With regard to potential cognitive differences based on writing: Tom doesn’t have an answer, but he wants to see responsible research done to address this question. Many Chinese intellectuals argued that Chinese characters were somehow cognitively inhibiting. Not only the character forms, but also grammar–e.g., the idea that there is no counterfactual in Chinese, making some statements unutterable, unthinkable.
For Tom, the starting point wouldn’t be the Greek miracle of an alphabet, but with the process: Yes, there is a different process involved in typing an email in Chinese. And sure, there must be some kind of cognitive process involved. BUT, we need a Hegel alarm that warns of essentializing backwardness, etc. We need to engage with media studies and subjectivity.