Long Term Research

Back to Contents of Issue: September 2001


Since 1989, the Advanced Telecommunication Research Institute near Kyoto has conducted some of the world's most significant, long-term research in human-machine communications. Now the Institute is being restructured, and more than a decade of quiet research is coming to fruition ...

by Alex Stewart

THE ADVANCED Telecommunication Research Institute, situated in the futuristically named Kansai Science City, has exercised a pull on my imagination since I first read about plans for building it in the mid-1980s. This was when the Bubble economy and yen spending power conferred invincibility on any project Japan undertook. Despite living on and off in Kansai since that period, I never actually set foot inside this fabled region. One reason could be that it is literally in the middle of nowhere: You do not pass through it, you go to it. Thus you must have a reason to go. And this I subsequently discovered could be one of its largest handicaps.

The nearest "experience" I had to entering the ATR Labs before my actual visit was through the pages of techno-cyber writer Howard Rheingold's book Virtual Reality, in which he devotes some 20 pages to a visit in 1990, only one year after ATR opened its doors. Futurist though he is, Rheingold could not foresee how Japan and its reputation for technology would change over the ensuing 10 years, and reading it now, the writing seems more ironic and eccentric than it did back then. Ten years ago, he tip-toed into ATR expecting to meet the future Godzilla to conquer the US lead in human-machine interface research. Ten years later he would stride in without any such preconceived fears N though no one, I decided by the end of my visit, should write off the ATR's quixotic adventure, either: There could be some hidden gems that will surface in a new science-techno wave led by Japan after all.

The question I set out to determine on this odyssey to the far-flung land of Kansai Science City, both as local resident and businessman helping venture companies inside and outside Japan, was if I was missing something under my nose, as my vague memory of Rheingold's description 10 years earlier had suggested could be the case -- memories of extraordinary research into technologies for projecting 3D virtual tele-images down broadband fiber have never quite left me.

The catalyst to exorcise this ghost was the announcements in the papers over the last two years about wearable real-time translation systems ATR had developed. As a poor speaker of Japanese, this struck me with more than usual force, and it certainly seemed worthy of ATR's reputation, in my mind, that it should be associated with such a futuristic development. With this in mind, I met first with ATR researcher Dr. Nick Campbell, a Brit, to talk to him about it (see interview).

The meeting with Dr. Campbell and the discovery of the wider research community operating in the laboratories in what is called Kansai Science City prompted the idea for a broader article. Initially it began as a feature to explain what the "ecosystem" around ATR consists of. However, the encounter raised wider philosophical issues about Japan's ability to do research and to spend money wisely, about regional competition, and about waste. In fact, this turned out to be a case of looking into the Pandora's Box of Japanese regional funding and big infrastructure projects and coming away distinctly underwhelmed -- though not hostile to the initial premise on which ATR was founded, nor to its potential to add considerable value and benefit to Japan's output in the critical area of human-computer interface design.

The ATR Laboratories are a child of the 1980s mindset in Japan, when a strong need was felt to contribute to fundamental research at an international level and dispel the image of Japan as borrower (even sneaky thief) -- taking the West's research ideas, turning them into unbeatable products, and destroying western industries. ATR was established as a private/public-funded institution, with the private sector contributing 30 percent of all invested funds. The private contribution must have seemed a relatively small "donation" when the Bubble produced money on trees, and the outside world was fulminating against Japan's "unlevel playing field"...

The ATR Laboratories were always intended to be the centerpiece of Kansai Science City and the biggest draw for research labs coming to the area. Created in 1986, ATR began operation at its existing site in 1989. The building provides 25,000 square meters of research space, which is on a par with other large research sites around the world. It provides facilities for 260 researchers, not counting support and administrative staff.

The building is surrounded by a considerable amount of undeveloped land. Rheingold on his 1990 visit speculated that this would leave room for expansion. I asked the same question, but it seems there are no plans. In fact, this is not the time to ask about expansion, as ATR is now undergoing the biggest transition in its history. Until July this year, the center was owned 70 percent by the government and 30 percent by the private sector. NTT provided over 50 percent of the private sector's contribution. A total of 136 private companies "invested" funds each year. The government's share was managed by the Key Technology Center, but day-to-day management was devolved to a body called ATR-International (ATR-I), staffed by mainly NTT and Ministry of Posts personnel.

Since July, a new law has come into effect that passes supervision and control of ATR from the Key Technology Center (now disbanded) to the Telecommunication Advancement Organization (TAO), a management agency affiliated with the Ministry of Public Management, Home Affairs, Posts, and Telecommunications. As such, it brings to an end several years in which the private sector has chafed at the annual cost of supporting the ATR's research budget. As Masanobu Higashida, a vice president and director of corporate strategy at ATR, notes, "Research people only spend -- they do not return funds to investors."

The annual budget thus far has been JPY8 billion (about $75 million). Corporate irritability is partly due to the economic downturn and shortage of funds, but also because too much of the research carried out by the ATR is perceived by companies as blue-sky. One researcher in a government-related media research project said, "I've attended the annual Open House, when all the research activities are on display to the public, but I wonder if the public or business people can see anything there they can understand. I think it's too general, and a waste of our tax money."

The TAO, rather than perform basic research, has always managed telecommunication applications, notably satellite services and operations, on behalf of the government. Thus there is a subtle shift under way to require ATR to reduce its research horizons and make the output of its research more accessible and practical. One way TAO can exert pressure is by requiring ATR to submit funding proposals in which it will compete for access with other bodies, including private research labs. This brings it into line, belatedly, with the way governments allocate research grants in the West. However, being Japan, it may still be a case of plus a change, plus c'est la mme chose. As one researcher put it, "It's not the first time for us to face a major challenge. The chances of everything staying unchanged has fallen from 100 percent to 98 percent." Nevertheless, a salaryman working in Keihanna commented, ironically, "The company has lost its former operating status" -- it is a sign of failure, not success."

In the new structure, all research will be managed under ATR-I, and the existing laboratories, or business units, will cease to operate. Under the former system, there were four main laboratories. Of these, the Human Information Processing group was disbanded last year; the ATR Media Integration and Communications Lab was due to complete its research activities ahead of schedule and begin focusing on the commercialization of its research results from the middle of this year. The two other labs, Adaptive Communications Research and ATR Spoken Language Translation Research, are also due to suspend their activities by the fall of this year. They will then be assigned to new research projects, based on their successful applications to TAO for funding. All research staff will now be directly employed by ATR-I.

After the Human Information Processing research group was disbanded, those belonging to the companies and institutes that had sent them on a two- or three-year contract returned, while those employed as direct research staff, usually the managers of the projects, were absorbed into a general unit called the Information Sciences Division. When I visited, there was a barbecue on the grounds outside the cafeteria at which members of the different departments and research sections were taking time out to socialize as part of the preparations for working in this new structure.

The goal of the research at the Human Information Processing (HIP) laboratory was to enhance human-machine communications. This involved research on natural language processing (speech synthesis, automatic speech recognition, and coding), neurological processing of voice patterns, visual information processing, scene and pattern recognition, and other related human information processing tasks. The research covered most of what other parts of ATR do as well. In this way it helped to integrate and focus some of the work in other laboratories.

Research on the HIP program ran for nine years and received JPY16 billion in funding over its life (around $145 million). This would be heaven for most research institutes, first for the level of freedom to plan research over such a long period, then for the size of the budget. One of the positive results of such largesse is that it attracted numerous foreign researchers to join ATR for contracted periods. As a result, the ATR laboratories are unique in Japan, and possibly the world, in enjoying both a long-range research focus and a large complement of researchers from around the world to work on large-scale, interdisciplinary, complex, human-machine communication issues.

This makes ATR not just a national institute carrying out research at the behest of government agencies, but rather an extension of a wider international research community, with its branches mainly in the US and Europe. As part of this international research effort, ATR published over 1,000 papers in 1999, over 400 of them outside Japan. Besides this, it has lodged between 10 and 20 international patents every year. Its major contribution internationally is almost certainly in speech technology. For example, in 1993 it initiated a worldwide experiment for machine translation, and it now serves as the chair of C-STAR II, an international research consortium for spoken language technology.

Compared with research centers in the Tokyo area, however, ATR and Kansai Science City have failed to attract investment from world-class labs. At least one European research lab considered the Keihanna area around ATR for its Japan base, but settled on Tokyo. There, the Yokosuka Research Park, 25 km south of Yokohama, has much greater commercial relevance, especially for foreign companies that want to keep abreast of wireless technology in Japan.

However, Yokosuka and the Kansai's ATR are looking at different research problems. Yokosuka is developing products along one- or two-year timetables, while ATR is looking at issues that will not be turned into products for five or 10 years. To underline this, ATR's Higashida reserved his greatest admiration for the Santa Fe Institute in New Mexico, a long-range think tank remote from the buzz of urban activity, which he and a team from ATR visited at the end of last year. He cited the informality, the comparatively large amount of funding it attracts (all on a voluntary basis from corporations), and its long-range research focus -- very fuzzy subjects -- as reasons for his approval. "You don't have to worry about the funding application process each year", he adds. Higashida himself worked at NTT's famous ETL (Electro-Technical Laboratory) in Yokosuka before coming to ATR in early 1999. He was an AI specialist focusing on speech recognition for many years, and, while he's an administrator now, evidently his heart still lies in the field of his research.

Perhaps the most important benefit of ATR's activity is that it provides stimulation for Japanese corporate researchers who join the labs on two- or three- year research tenures. During this time, they come into contact with top brains from outside Japan in an atmosphere that, by Japanese standards, is laid back and highly creative. About 30 percent of the researchers come from private domestic companies. However, one former researcher from a government institution describes many such researchers as vacationers: "It's an opportunity for them to take time out: They get to visit the US and attend seminars, and when their contract time is up they are often reluctant to go back to their company". Yep, like the musical asks, "How can you hold them down on the farm when they've been to Paree?" Still, most do go back, and they add to an ever-growing "old boys" network, which after 15 years in operation must number well over 400 research staff around Japan, many of them in the Kansai area. It must also create at some point, if it has not done so already, a critical mass of know-how. If so, the results are likely to emerge over the next few years, as the blue-sky ideas trickle and then flood into commercial applications.

Some of the areas likely to see greater commercialization include:
* Wearable translation systems. Matsushita will probably have a lead here; its labs are directly across from ATR.
* Voice portal and voice recognition applications, perhaps linked to car navigation systems. Omron is staking out a strong position here and also has very close links with ATR.
* 3-D avatars to create an intelligent agent interface experience on wireless or PC applications. Sharp, down the road in Nara, has developed an avatar with complete body and facial movements. It can deliver email messages as speech. Check out Sharp Space Town (www.spacetown.ne.jp) -- it's in Japanese -- for a glimpse into an AI-mediated future worthy of ATR's HAL-style research interests.

All of these companies are in the Kansai region. ATR may thus take some credit for helping to create a new area of competitive advantage for the consumer electronics champions of the region. It's also certain to be part of the major trend toward ever friendlier, cuter, cuddlier[!], more anthropomorphic, and simply odd user experiences. Japan is likely to commercialize this area faster and more effectively than other countries, due to the ease with which its people co-exist with machines and adopt synthetic lifestyles in general.

Despite its potential, ATR is not the most effective marketer of its own research. Each laboratory has a marketing arm set up to operate for 20 years, in order to maximize the return on investment from research patents and applications. However, as vice-president Higashida candidly admitted, "Truth to tell, none of our research products are in very strong use in the private sector." One of the exceptions is CHATR, which Dr. Campbell (page 22) helped to lead. Almost all the other examples of commercialized results are also in the speech processing field. For example, Matsushita is adopting ATR's MATRIX system in its notebook PC speech translation system. ATR also has a multilingual document translation system that companies are incorporating into products. It has developed a voice recognition system expected to have applications in car navigation systems. There is also quite a bit of its research inside NTT's applications. "Our research is more advanced than NTT's. They focus on their immediate needs. If they need deeper research, for example in areas based on neuroscience or gene analysis, they will ask us," says Higashida.

A trawl through the Nikkei database over the last year produces half a dozen or more research applications from ATR that have attracted commercial interest. A "wearable musical instrument" helps the body play music when pads wired on clothing are rapped. It seems the idea is to use the instruments for virtual jam sessions, using wireless communications to connect users in different locations. A robot pet responds with different musical rhythms and tempos depending on how it is handled. ATR researchers have contributed significantly to the understanding of how robots can learn and visualize their surroundings. ATR's Robovie is a humanoid designed for use in daily life. Nevertheless, when Sony went looking for software that simulates the movement of humans to help it develop humanoid robots, it found it in a US spin-off from the MIT Artificial Intelligence Laboratory in Boston, not ATR.

In research centers elsewhere in the world, one can expect cases of new business spin-offs, which also serves as a measure of their health and vibrancy. In ATR's case, there is currently only one example. I asked Dr Campbell over coffee about the state of venture activity. He could only cite two former foreign researchers who had made it big on their return to the US, though in areas unconnected with their immediate research. The absence of this venture spirit is not surprising, though, given the backgrounds from which the researchers come -- mainly companies that promote the lifetime employment system, or universities and research institutes that foster the same cradle-to-grave mentality.

Enter former researcher Kennosuke Wada, member of the disbanded HIP project. Last year, he became the first researcher at ATR to leave and set up his own company. He learned of a METI initiative to promote individual innovators, rather than companies (in order, as an NHK program reported earlier this year, "to find the next Bill Gates"). It was conceived and managed by a young METI bureaucrat, Takahiro Hagawara, also something of an angel in Wada's eyes. He trapped JPY5 billion of government funds to provide 56 "super creators" with the seed capital to turn creative ideas into businesses. Wada described the ATR labs as a wonderful place to work, but acknowledged he could get into a rut. The alternative -- returning to a university to continue research -- would be like "going back to kindergarten." The other reason Wada could make the jump was that last year Kyoto Prefecture opened "an incubator facility" in the Keihanna Plaza opposite the ATR Laboratories. ATR provided him with most of his furniture. He does not have staff, though. His wife, who also worked briefly in the HIP Laboratory, now assists him, "because I was worried he was working too long hours." His mother, a matriarchal figure seemingly always in a kimono, and as different from the image of a high-tech CFO as one could wish, runs the business side. Mr. Wada and three other "super creators", two of whom live in different parts of Japan and one who works elsewhere in the Plaza, write the code. They congregate from time to time in his incubator for marathon sessions, he explains, sleeping cross their desks in the evening, waking at midnight to pound out code fiendishly through the night.

Wada is the archetypal creator, the type Japan's research community produces once in every thousand tries, as opposed to say once in every hundred in the US. By background he is a geneticist. The HIP project, in its own words, involved "psychology, medicine, dentistry, physiology, mathematics, biology, and others," so it had a nicely eclectic mix for a Renaissance researcher.

At his new company, Untrod, Wada's research is focused on writing computer programs to accelerate the process of creating the physical movements of anime characters in cartoons. He uses mathematical formulae to compute how a figure will move on the screen, using the humanoid and genetic knowledge he acquired in his research at ATR. This then automates part of the movement sequence -- an almost Pinocchio experience. He also demonstrated a program for linking the actions of different characters across a network of PCs -- which would be ideal for networked games. He is working on confidential areas involving real-time voice communication and a robot avatar with an AI engine, both clearly outgrowths of research at the HIP Laboratory. Another "top secret project" concerns a major advertising agency in Osaka, which has commissioned him to design a new kind of system for providing more graphical interpretations for analyzing marketing survey data. It is a sobering thought that it has taken more than 10 years for the first "super creator" to emerge as a venture innovator from this science utopia. Seen thus, the emergence of Mr. Wada from the ivory tower of ATR's research sanctuary is a small step on what is still a long road.

Despite this poor record of research output and venture activity, in defense of ATR's existence, one might say that the kind of research it's done in human-machine interaction is actually essential to the nation. This is because so much of Japan's future economy rides on the ability to make machines truly friendly, so that the Japanese, with their kana-kanji keyboard phobia, dislike of working in isolation, and innate love of cute products with zany interface designs, can ride through to a new future, comparable to the one that was promised when the ATR idea was first proposed in 1985. In this emerging techno-science paradigm, Japan could still have it all to play for. Then ATR's image could enjoy a dust-off, re-jigged from Bubble spendthrift to cutting-edge research center of the new Japan Information Revolution. After all, Japan was going to win the technology race hands down, so it seemed when Howard Rheingold wrote in 1990 the usual paeans to Japanese research organization and industrial competence. Just about everyone agreed, or did not dissent. Now, the reverse judgement is held: Japan's day is over.

However, from this overlooked corner of Japan's former high-tech hopes, a revolution may just be brewing. And, who knows, in a few years' time writers like Rheingold might be descending again to find out the "secrets" of Japan's success. Having personally experienced a few cycles of hyped expectation followed by disappointment over the years of watching Japan, I'm prepared to allow the benefit of the doubt to ATR, and to Japan's techno-renaissance.

Note: The function "email this page" is currently not supported for this page.