The Character of Chinese Computing

Maura Elizabeth Cunningham reviews Thomas S. Mullaney’s “The Chinese Computer: A Global History of the Information Age.”

By Maura Elizabeth CunninghamDecember 1, 2024

• Science & Technology

• History

EVERYTHING ABOUT LEARNING Chinese felt difficult when I began studying it in 2003. I struggled to pick up Mandarin’s four tones and spent hours listening to tapes of my professor reciting phrases, then recording my own attempts at mimicry. I could memorize the romanized version of the language, known as pinyin, with relative facility, but I didn’t enjoy the same ease with characters, which I read slowly and wrote poorly. Ting shuo du xie—listen, speak, read, write—each of these four building blocks necessary for learning Chinese bedeviled me throughout that first class.

When I started typing assignments in Chinese a year or so later, though, I needed only minutes to become comfortable. With Chinese-language software added to my computer, I could smoothly toggle between English and Chinese entry on a regular QWERTY keyboard, tapping in romanized pinyin words to bring up a selection window of potential character matches. The candidates changed as I typed, the software responding to context and drawing on its memory of characters I had used frequently in the past. For example, if I typed the pinyin “shi,” the software would begin by suggesting the most common characters pronounced “shi” like 是 史 詩 十 事 or 時 (there are many more characters pronounced “shi”). It assigned number values to each character, allowing me to quickly choose the one I wanted.

The program often saved me from mistakes, suggesting a correct character when I might have otherwise written the wrong one. Sometimes you know a word is pronounced “shi” but you forget what the character looks like. Ting shuo du xie still weren’t my strengths, but with da—type—I had finally found something in Chinese that came to me easily.

It was only by chance that my first encounter with Chinese-language entry software (known in technical circles as input method editors, or IMEs) came in the early 2000s, when programs built on a foundation of phonetic pinyin input had recently become dominant. Prior to that, the hundreds of IMEs designed by linguists and computer scientists were all structure-based: typists used combinations of keyboard entries to signify elements of a character’s shape and create a pool of candidates from those. Learning a structure-based IME would have required me to have excellent recall for what characters look like, as well as the ability to memorize entry codes corresponding to each of those character components. If I’d started typing in Chinese five or 10 years before I did, using a structure-based IME would have been yet another arduous task in learning an already challenging language.

The creation of character input methods, and how thinking about them has changed our relationship with language itself, is the core subject of The Chinese Computer: A Global History of the Information Age, by Stanford University history professor Thomas S. Mullaney. In this new sequel to his 2017 book The Chinese Typewriter: A History, Mullaney covers the story of Chinese computing from the late 1940s through the present day, describing experiments in hardware and software that failed more often than they succeeded but have nevertheless left behind enduring legacies for the entire computing world. These technological palimpsests break through every time Microsoft Outlook attempts to complete a sentence for me, or whenever Google suggests the end of a search term while I’m still typing.

Mullaney begins with a bridge between his previous volume and the present one: the electric Chinese typewriter that inventor Chung-Chin Kao began pitching to IBM and Mergenthaler Linotype in the early 1940s. Instead of a QWERTY keyboard, the machine employed four rows of Arabic numeral keys. Kao composed a reference book of 6,000 Chinese characters, each corresponding to a four-digit code; when the typist depressed the four numbers of a given code on the keys, the machine retrieved and printed the selected character. After completing a prototype in 1946, Kao organized a series of public demonstrations in the United States and China to prove the viability of this enterprise. Smartly dressed young women—Mullaney was able to interview one, Lois Lew, before her death—sat at the machine and took “dictation” from the audience, entering four-digit codes from memory as onlookers called out various characters.

Although this number-based entry method was rooted in that of Chinese telegraphy, which had been in use for decades, the bosses at IBM and Mergenthaler were skeptical that a machine requiring such enormous capacity for memorization would prove commercially successful. Learning Morse code or similar for the 26 letters of the Anglophone alphabet was one thing, but retaining thousands of character codes, and inputting them at an efficient typing speed, was quite another. As Mullaney points out, Kao’s invention asked American corporate leaders “to contemplate a model of human-machine interaction that had no precedent, nor parallel, in the Western world at that point.” A civil war in China further added to their hesitation, and the 1949 victory of Mao Zedong and the Chinese Communist Party (CCP) made the country an unstable place in which to build a new business sector at the dawn of the Cold War era. To Kao’s disappointment, IBM and Mergenthaler Linotype exited the Chinese computer business before it even got off the ground.

For a period in the 1950s, however, US military leaders saw in Chinese computing the opportunity for a Cold War win. An MIT professor named Samuel Hawks Caldwell had devised the “Sinotype,” the first true version of what we would recognize as a Chinese computer. Though he did not speak or read Chinese (indeed, the creation of Chinese computers seems to have often attracted men who were interested in the engineering challenge, not the language), Caldwell worked with a team of native speakers to figure out how entering the structural components of a character could aid in suggesting potential matches—the basis of structure-based IMEs. 

Caldwell’s initial attempts to secure financial support for Sinotype based on claims that it would add to global engineering knowledge were unsuccessful. Caldwell, Mullaney writes, figured out that he needed to present Sinotype “less as a project meant to benefit foreign countries than one meant to augment the reputation of the United States on the global stage.” Once he reframed his request in national security terms, the money started rolling in from both private foundations and the US military. As the years of work dragged on, though, military leaders exhibited the same hesitation that Kao had encountered. Was Chinese computing really possible? How many people would actually want a Sinotype? Would the utility of such a machine justify the cost of developing it? Support fizzled, and Caldwell’s death in 1960 brought an end to the Sinotype project.

Yet the legacy of Sinotype lived on. Caldwell and his colleagues had introduced two lasting elements of Chinese computing: the preview window, which showed the typist any characters that fit the parameters they had input, and auto-completion, which determined the most likely characters based on the minimum number of inputs. (To illustrate this, Mullaney draws on the English example of the word “xylophone,” which is easy to identify and complete as the letters x-y-l-o-p … are entered.) To this day, both structure- and phonetic-based IMEs employ the technologies that the Sinotype team first devised in the 1950s.

Other innovations proved more immediately successful but ultimately less enduring. In the 1960s and ’70s, engineers moved away from designing QWERTY-based machines, seeking to streamline the input process by bringing Chinese typewriters and their tray beds full of character slugs into the computing age. IPX, created by Taiwan-born inventor Chan-hui Yeh and his Ideographix company, was an early Silicon Valley product that put nearly 20,000 characters within reach of typists. (There’s no easy way to summarize how IPX worked; Mullaney needs more than a page to walk the reader through the basics, and he concedes that, on its own, the machine appeared “unprecedented, even outlandish.”) First limited to use by the Taiwanese military, which had provided funding for its development, IPX then enabled significant productivity leaps in generating telephone bills and directories, tax documents, and newspapers across Taiwan. Yet, when talking with Mullaney, Yeh admits, “We ran ourselves out of business.” IPX was such a success in Taiwan that the market quickly became saturated, and Yeh’s attempts to court customers in mainland China ran into lingering political tensions between Beijing and Taipei. By creating a machine uniquely tailored to the Chinese language, Ideographix had backed itself into a corner when it found no room for expansion.

Custom-built devices were rendered obsolete by the software and PC revolutions of the 1980s, which took computing out of the hands of experts and made hardware cheaper. This coincided with the opening-up of mainland China during the early years of the reform era. Although disconnected from Chinese computing advances in both Taiwan and the United States during the Mao era (1949–76), engineers on the mainland had worked on similar projects, even during the upheaval of the Cultural Revolution (1966–76). By the 1980s, Chinese developers were primed to make their own contributions to the field, turning back to QWERTY-keyboard devices and the IMEs that would make them viable options for Chinese-language users. By the mid-1980s, there were 400–450 structure-based IMEs available to users, without a single one emerging as dominant. With so many different input methods to choose from, Mullaney observes, “pure accident—and in other cases relentless self-promotion—played as large a role as merit.”

To run any of these IMEs, the PC had to be tailored to the needs of Chinese users, with technicians engaging in a flurry of modifications, “piracy,” and hacking of foreign computer systems. This was not to copy them or lower their cost, Mullaney stresses, but simply to make them compatible with the demands of installing Chinese-language software, which required enormous amounts of memory—well beyond what computers came with out of the box at the time. “[T]he history of early Chinese personal computing has been retrospectively imagined as one in which the Western-built computer had always been language-agnostic, neutral, and welcoming,” Mullaney writes, while describing extensive alterations that prove the opposite was true.

By the end of the 20th century, software engineers had built on Sinotype’s auto-complete function to create ever-improving predictive text. Early versions suggested the next character in a compound; over time, programs began to suggest which word would come next, based on context, character repositories, and statistical frequency analysis. Even more than in previous decades, the skills needed to develop IMEs were not linguistic but mathematical, as programming them required knowledge of statistics and probability theory rather than character structure. The expansion of predictive text technologies finally overcame the problems posed by pinyin’s many homophones and made pinyin-based phonetic-input IMEs reliable enough to crack the market previously dominated by structure-based programs.

For all the hours that I’ve spent typing in Chinese, I never made the connection between the preview window filled with characters and my phone attempting to complete a text for me in English. They’re the same technology, I now recognize, but while the English version is an add-on that I often ignore, the use of auto-complete and predictive text is what makes typing in Chinese smooth and efficient. Mullaney identifies my attitude as that of many Anglophone computer users, who have “left on the table” much of the potential of their devices. “In contrast,” he argues,

engineers focused on non-Latin computing have spent incalculable hours trying to figure out ways to reimagine human-computer interaction and harness the computational power of microcomputers to make their input experiences more intuitive or efficient—a pursuit driven by baseline necessity.

Only the needs of Chinese computer users (and those typing in Japanese, Arabic, Burmese, and other languages that require the mediation of an IME) drove engineers to develop work-arounds that have subsequently become global, and essential, technologies. Without the host of modifications and innovations developed for Chinese computing, Mullaney posits, “no Western-built computer could have achieved a meaningful presence in the world beyond the Americas and Europe.”

Mullaney writes in an engaging, conversational tone, even when describing complex character-input processes. He has a skill for picking out memorable people and anecdotes that enliven his compelling prose. The Chinese Computer is more complicated to follow than The Chinese Typewriter, which benefits from having fewer technical details to navigate and a deeper consideration of questions about language and modernity that Chinese intellectuals debated in the early 20th century. Yet this volume offers readers a rich examination of how modification, experimentation, and more than one technological dead end resulted in today’s Chinese-language computer users sitting down to type at a QWERTY keyboard instead of a custom device. At first glance, that seems like a concession, or an example of Euro-American hegemony steamrolling over the need for local variations elsewhere. As Mullaney concludes, however, the software inside those QWERTY-keyboard machines is where the true importance of Chinese computing can be found.