Computer Science R&D in Japan
There is rising international interest in Japanese scientific and technological R&D, and efforts are underway by Japanese researchers to make their work available to the public, such as via the Internet. This month, Computing Japan takes an introductory look at what is going on inside the computer science research labs at Japan's universities, corporations, and government-sponsored institutions. And starting with the August issue, we will introduce a monthly R&D page that focuses on some of the many astounding, and often esoteric, projects being explored by Japan's researchers.by Steven Myers
In the popular media, Japan has long had the image of a country skillful at adopting and improving upon foreign technology, but contributing relatively little original research work of its own. The past few years, however, have seen a sharp increase in the attention given to Japanese research and development (R&D) by foreign scientists and media pundits, and the global community is starting to realize that significant R&D efforts are underway in Japan. Several organizations and project groups have appeared recently, aiming to expand the collection and dissemination of information pertaining to Japanese technological research. Not surprisingly, a large percentage of Japan's research is related to computer science and the computer industry - especially issues pertaining to networking, distributed operating systems, and human-computer interaction - and involves Japanese universities as well as corporate giants such as NTT, Sony, Fujitsu, and Hitachi.
Rising international interest in Japanese R&D
In November 1993, H. U. Hoppe, a German computer scientist working for the German National Research Center for Computer Science (GMD), reported on his three-month trip to various research centers in Japan. Dr. Hoppe said that his visit would serve as the pilot version for future GMD-sponsored scientific visits to Japan, and in his report he stressed the need for foreign computer scientists to gain "deeper insight into specific research in Japan, in a way that goes beyond what is possible through short round-trips or conference visits."This view seems to be representative of the current general mood regarding Japanese R&D. According to Dr. Hiroaki Kitano, a researcher at Sony Computer Science Lab (CSL) and one of Japan's leading authorities on artificial intelligence (AI), Sony's lab has seen a remarkable increase in the number of foreign visitors in the past two years. The lab currently hosts at least one group of visiting scientists each week from leading universities and research centers around the world. Several computer research labs throughout Japan contacted by Computing Japan, including Sony CSL, say that they are inundated with requests from foreign researchers for "visiting scientist" positions.
Why this sudden surge in foreign interest? It may be that Japanese R&D efforts are at last becoming more visible. In a speech delivered last year at a Japan Information Center of Science and Technology (JICST) conference, Dr. Mary Good, an undersecretary of the US Department of Commerce (DOC) noted that the DOC has long been committed to turning the attention of US industry toward Japanese science and technology information. These efforts are finally paying off in the form of an increase in foreign awareness of such information. Dr. Good says she is greatly encouraged by the increasing number of scientists studying Japanese, and excited about the establishment of a Machine Translation Center in Washington, DC - a joint project of DOC and JICST.
The impact of the Internet on technology management practices in Japanese government and industry has been cited as another factor in the renewed interest shown toward Japanese R&D. The wide-open nature of the Internet is contributing slowly but steadily to a breakdown of bureaucratic barriers that have hampered the spread of Japanese scientific information in the past. Many Japanese are enticed by the capability for the unbiased dissemination of information offered by the Internet, in contrast to the tight control that government ministries have traditionally tried to exert. Today, with the proliferation of Japanese R&D labs offering technical reports on their work through their own World Wide Web (WWW) pages, it is easier than ever before for a foreign computer scientist to obtain up-to-date information on Japanese research developments.
Japanese attitudes toward R&D
Foreign scientists investigating Japan as a source for technological information are naturally curious about the apparent differences in attitude regarding the role of R&D that exists between the "powers that be" in Japan and those in their home countries. Several researchers have commented on the pronounced differences in "style" and orientation of the computer-related research being conducted in Japan. Dr. Hoppe, for example, notes that the necessity for basic research of a general nature (as opposed to applied research related to specific products) is much more accepted in Japan than it is in Europe. "It is not questioned that results of this research are values per se, not just the ensuing industry products," he notes.Researchers in Europe have often been blamed for the failure of the European IT (information technology) industry, Hoppe says, but such arguments are seldom heard in Japan. Computer science researchers in Japan are not only allowed, but encouraged, to take on complex problems in fundamental areas of computing through long-term projects whose outcome is far from certain. Dr. Kitano supports this view, acknowledging that many of the projects currently underway at Sony CSL might not bear fruit for another 10 to 15 years. These projects are nonetheless actively supported by Sony Corporation, however, he notes.
This appears to be indicative of a recent emphasis-shift in Japan, away from product-driven R&D and toward research of a more basic nature. Because of Japan's export-oriented economy and sector-specific interaction between government and industry, the nation has heretofore focused on rapid product development and technology acquisition/transfer rather than on basic technology development. In last December's White Paper on Science and Technology, however, the Japanese Science and Technology Agency (the highest science and technology policy-making body in the government) called on government agencies and private firms to make efforts in such areas as strengthening basic research, developing originality and creativity in their staffs, and placing importance on international aspects.
Government promotion of R&D
The White Paper states that Japan must "take more initiative in contributing to the knowledge base of mankind and in resolving issues of global concern." The Science and Technology Agency identifies more than 180 "fundamental technologies" that should be developed over the next decade in order to advance the frontier of technological R&D in Japan. Among computer-related technologies, the areas emphasized include data standardization and database development, integrated simulation and virtual reality technologies, human-computer interfaces, and autonomous distributed systems.One proposal put forth by the Science and Technology Agency to promote Japanese R&D involves the development of advanced "fact databases." This would require close cooperation between specialized information services, research institutes, and academic societies. The agency also recommended the implementation of a new policy for performance evaluation of individual researchers, one that will give proper credit to those researchers who make valuable contributions to databases. The agency has called on corporate research labs to participate fully in efforts to develop these databases.
The corporate reality
In spite of such visionary recommendations, though, the Japanese government has taken considerable criticism recently over the perceived lack of progress by Japan in computer technology development. As evinced by the recent media frenzy over multimedia and the Internet, as well as by recent books with such (translated) titles as The US-Japan Multimedia War, there is an attitude of fear among many Japanese that the nation is falling behind in a technology race with the United States.A recent Time magazine article advanced the opinion that virtually all of the major elements in Japan's multimedia market are underdeveloped, due to "over-regulation, high prices, and other innovation-stunting problems." The Time article cites several revealing statistics, such as 96% of American homes have access to cable television (a major part of the information-highway infrastructure), versus 19% in Japan, and 52% of personal computers are connected to a network in the US, compared to only 9% in Japan.
It seems evident that a large number of Japanese corporations feel the need to develop technologies and infrastructure that will help to quickly secure the domestic market in the areas of multimedia and virtual reality, rather than focusing on goals that are less concrete. Indeed, according to a recent Science and Technology Agency survey, private firms are increasing their expectations for the results of R&D. Some 58% of the respondents demand their researchers develop products leading to actual production, 40% support the elimination of research that does not yield tangible results, and 39% favor the shortening of their firm's time limit for R&D to produce results.
R&D spending down
Japanese corporate spending on R&D has been declining since fiscal year 1992, when R&D investment dropped for the first time ever. In spite of this continued decline, however, the government is not providing direct assistance to corporations for R&D, nor is such assistance expected. Perhaps surprisingly, from the corporate perspective, government support for R&D should go to promote basic research at universities and national research centers.
Almost 70% of the corporations surveyed stated that even in the economic recession, the relative importance of R&D in the company's overall business strategy has increased. This response, together with statistics showing that the number of corporate researchers has increased, leads the government to believe that the current decline in corporate R&D expenditure is a temporary phenomenon and will soon be reversed.
Sharing the knowledge
A large number of government organizations, academic groups, and other programs have appeared, both in Japan and abroad, to promote the international sharing of Japanese technological research results. Japan Window (a Web site provided by NTT and Stanford University), the University of Arizona's JapanCS project, and the University of New Mexico's US-Japan Center are among the organizations providing such information. (See the sidebar for http addresses.)Online information sources are valuable tools for foreign scientists, and the increase in the number of researchers actually making visits to labs in Japanese universities and corporations is seen as a positive trend (one that is being actively promoted by several organizations within Japanese and foreign governments). As Dr. Hoppe comments, however, there is also a "strong and dense network of personal links between Japanese researchers across different fields of computer science and information technology that can help prepare visits, make new contacts, and make existing contacts more valuable through personal relations."
In the following pages, we introduce just a few of the well-known computer R&D labs in Japan, and describe some of the projects underway there. And starting next month, Computing Japan will feature a monthly report on other labs and projects.
Sources of Japanese computer R&D information on the World Wide Web
Agency of Industrial Science and Technology, Ministry of International Trade and Industryhttp://aist.go.jp
JapanCS Project, University of Arizona
http://cs.arizona.edu
Japan Information Center of Science and Technology
http://jicst.go.jp
Japan Window
http://jw.stanford.edu
National Center for Science Information Systems
http://ncsis.ac.jp
NTT
http://ntt.jp
Real World Computing Partnership
http://rwcp.or.jp
Sony Computer Science Laboratory
http://csl.sony.co.jp
Tokyo Institute of Technology
http://soc.titech.ac.jp
A recent emphasis-shift in Japan [is] away from product-driven R&D and toward research of a more basic nature.
Navicam and Social Agent Projects (Sony Computer Science Laboratory)
Sony Computer Science Laboratory (CSL) was founded in February 1988 for the purpose of conducting research related to computer science. Well known in Japan for having researched and implemented Apertos (a distributed object-oriented operating system in use at many Japanese universities), Sony CSL is currently home to 15 of Japan's top computer scientists, all with doctorate degrees from prestigious universities. Sony CSL's research covers a broad range of topics, including networks, programming languages, human-computer interaction, artificial intelligence, and complex systems. At present, the lab is also hosting two foreign researchers, and is actively involved in joint research projects with well-known computer R&D labs around the world.
When visiting Sony CSL, I was immediately impressed by its open and relaxed atmosphere. None of the research staff have positions, titles, or other signs to indicate seniority. Dr. Hiroaki Kitano, a well-known researcher in the field of AI (artificial intelligence) and recent recipient of the prestigious Computers and Thought Award, explains that - unlike the vast majority of R&D labs in Japan - Sony CSL's compensation system is completely unrelated to seniority; researchers are financially compensated in accordance with their individual achievements. He commented that almost all of the projects underway at the lab are conducted by, at most, two members. Graduate students from schools such as Keio and Tokyo University provide assistance when needed.
Most of the lab's many projects are long-term, and not related to the development of specific Sony products. Two projects that particularly stand out are those of the prototypes for NaviCam and Social Agent. Both are examples of research into the field of complex systems, involving the development of autonomous, intelligent agents.
NaviCam
NaviCam, being developed by Jun Rekimoto, is a tiny, highly portable computer system that displays a high degree of position and situation awareness, as well as impressive speech recognition capabilities. The goal is to develop a system so small and unobtrusive as to be virtually unnoticeable to the user - one that can be used to supply context-sensitive information for a variety of situations. Rekimoto describes this type of human-computer interaction (in which the user interacts with the real world through a transparent computer) as "HyperReality." This contrasts, he says, with the human-computer interaction styles such as virtual reality (in which the computer replaces the real world) and the "ubiquitous computers" approach (in which objects in real life become computers) being explored through Tokyo University's TRON project.
To keep the devices light and small, the system uses wireless communication to connect to a back-end computer, which acts as a server. The server contains the database that stores activity information about the user and real-world information about the current environment, and it can also act as a gateway to other networks.
In a prototype demonstration, NaviCam was used to provide spoken information about the user's surroundings within different parts of the building, based on spoken queries (for example, "Where are we now?" or "Who works in this room?"). Rekimoto predicts that within five years, such computer systems will be as commonplace as the Walkman and other portable audio devices.
Social Agent
The aim of the Social Agent project, being developed by Katashi Nagao and Akikazu Takeuchi, is to create an intelligent computer interface in the form of an autonomous agent that appears on the screen as a human face. This agent is able to follow and contribute to conversation with human speakers (or even other agents), using facial expressions as well as spoken language. The agent will be able to shift its gaze from speaker to speaker and detect communication mismatches.
The current implementation comprises two subsystems: a facial animation subsystem that generates a three-dimensional face capable of displaying facial expressions, and a spoken-language subsystem that recognizes and interprets natural speech (and provides spoken output). The model of the face is composed of some 500 polygons, and muscle movements are simulated numerically. In the prototype implementation, the animation subsystem runs on an SGI320VGX that communicates via an Ethernet network, with the spoken-language subsystem running on a Sony NEWS workstation.
I observed a demonstration in which two humans were discussing their plans for the evening. The "face" of the computer agent followed each speaker, offering occasional suggestions and replying to questions from the speakers about what TV programs were scheduled for the evening. The agent also handled spoken requests for such tasks as making dinner reservations and setting a VCR.
Nagao and Takeuchi say that future plans for the system include the simulation of human-to-human communication in more complex social environments, which take into account such factors as the social standing, reputation, personality, and values of each participant. based on these studies, the pair hope to be able to propose a set of design principles for a society of computer agents.-S. Myers
Multilingual I/O and Text Manipulation System Project (Waseda University)
At Waseda University's School of Science and Engineering, a team of researchers has been working since 1992 to implement a fully "internationalized" computer system - one that can handle all of the world's writing scripts and code sets dynamically, with a minimum of overhead. The project was conceived in 1988 when Yutaka Kataoka, now head researcher for the project, was asked by MIT's Robert Scheifler to investigate methods for incorporating multilingual support into the X11R5 release of the X Window system. While the multilingual capabilities introduced in that release were a major improvement over the X11R4 version, Kataoka says that a true multilingual solution has yet to be realized, although the need for such a system is growing quickly.
The Multilingual I/O and Text Manipulation System Project began in April 1992 and was renewed bythe University in March 1995. (R&D projects at Waseda are generally granted for three-year terms; projects deemed promising enough can be renewed). Kataoka says the project has received little corporate funding (the only initial supporter was Omron Corporation); in fact, it has been strongly discouraged by several big-name computer vendors who would much rather see their own proprietary "localized" systems become the standard. Government support has also been scarce, due in part to Waseda's "renegade" image (which stems from the University's repeated refusals to comply with arbitrary Ministry of Education guidelines for curriculum and degree requirements). Nevertheless, research has progressed, and as papers on the project have been published, more and more organizations, including NTT and JCC (Japan Computer Corporation) are backing the project and offering their support.
Multilingual text-processing issues
Initial research on the project involved the analysis and categorization of global written languages and orthographies. Kataoka confesses that the extreme variation of natural languages posed several challenges to the researchers. For example, many languages do not use blanks spaces to delimit words, and punctuation symbols vary greatly among languages. In addition to classifying languages as phonogrammic (e.g., English) and ideographic (e.g., Chinese), developers must also consider that in some languages (e.g. Arabic or Devanagari), the symbolic forms of written characters vary based on position within a character string. Most current implementations, say the research team, are extremely limited because of a lack of sufficient knowledge about writing conventions and restrictions on their I/O system development environment.
Implementation
The input method of the project's system is implemented in the following manner: When keystrokes are input to an application, they are intercepted by an input multiplexer module, which transfers the keystrokes to the input manager via a communication library. The automaton interpreter within the input manager converts the keystroke sequence to ideograms, and transmits them to the editor module along with information on whether further conversion is needed. The character string is then returned to the input multiplexer. The code set converter provides the input multiplexer with the code set specified by the application, and the input multiplexer then loads the character string values from this code set into the buffer for a function called FGetString( ), which returns the value to the application, acting as the interface between the application and the IM library.
A function called FPutString( ) links the application to the OM (output method) library in a similar fashion. The code set converter converts the string to a sequence of code points for the automaton interpreter, which identifies the language name and generates the font for the characters to be written. Characters requiring additional context-dependent analysis are sent to language-specific modules, then returned to the automaton interpreter. The information thus returned to the application includes the range of code point strings and the font required.
Results of the multilingual I/O research
The Waseda team has already produced a large library of software tools related to the project, including a multilingual version of FORTH, dot matrix and outline font editors, a high-speed multilingual parser, and a multi-device input method server. A multilingual version of Common LISP is also being developed. The researchers have also made extensive modifications to the X11 libraries in order to accommodate their I/O system.
Although the project will continue for a second three-year term (through March 1998), Kataoka says that the implementation of the system will be completed by the summer of 1996. At that time, Waseda will distribute as freeware both the binaries and all of the source code developed throughout the course of the project. - S. Myers
Aizu Supercomputer Project (University of Aizu)
Established in the mountains of Fukushima prefecture in April 1993, the University of Aizu is an international university that emphasizes research in computer software and hardware. The computer science faculty includes a large number of foreign professors conducting research in over 20 different labs, with an abundance of sophisticated computing equipment at their disposal. Many of the foreign researchers are Russian, and they are playing an interesting and significant role in establishing the direction of the university's research. After visiting the University of Aizu, Dr. David Kahaner, a numerical analyst who was then working with the US Office of Naval Research in Asia, commented that "the Russian scientists here are particularly interesting; their isolated system often produced research directions that differ significantly from those in the West or Japan."
Multimedia Center
The university's multimedia center features an array of intriguing virtual reality facilities, including an "artificial worlds" zone and multimedia-aided design system, a human performance analysis system (with multi-modal human interface), and a multimedia networks & groupware system. A key component of this virtual reality system is a research project, headed by Dr. Tsuneo Ikedo and Dr. Nikolay Mirenkov, called the Aizu Supercomputer. The two scientists describe this computer as being both a special-purpose computer to control the equipment of the multimedia center and a general-purpose system to solve virtual reality and simulation problems.
Architecture of the Aizu Supercomputer
The Aizu Supercomputer is made up of a connected set of general purpose microprocessors with a distributed memory arrangement - each processor has its own physical local memory module. A single address space is supported across all processors by allocating a part of each module to a common global memory. The initial prototype of the system will have 1365 processing elements and achieve a speed of more than 100 gigaFLOPS (billion floating-point operations per second) at peak levels.
Each processing element consists of a 64-bit R4400 MIPS RISC microprocessor, two memory management processors, sending and receiving routers, a cache coherence controller for distributed shared memory configurations, and up to 64MB of local SRAM with 15-nanosecond access time. The processing elements (PEs), or nodes, of the Aizu Supercomputer are arranged in a pyramid fashion, with 1,024 nodes on the bottom layer. PEs in the upper layers of the pyramid can be considered as a special-purpose communication network performing operations on "flying" data and supporting global control. Application-oriented processors, such as those for graphics, sound, video, and text, are interfaced with the PEs of the pyramid's bottom layer.
Software approach
The initial software being developed for the Aizu Supercomputer consists largely of tools designed to make programming more visual and sound-oriented. The developers are taking a multimedia approach to programming, using animation clips, icon menus, sound, etc., to make the specification of algorithms a more natural and intuitive process. According to Dr. Ikedo, "Sounds will play a great role for users having a good ear for music; colors will be preferable for users having a good eye for painting; dynamics will be favorable for users liking speed and expressiveness, etc." After a user describes the operations for the program to perform, he or she can watch and listen to the result of the algorithm and thus be able to partially debug it before execution.
Currently, the researchers have an experimental version (which they have dubbed the VIM system) that can present the basic ideas of the technology. They acknowledge, however, that a great deal of work is required before they will be able to develop a practical version of the system. They say they are greatly encouraged by the initial results, and know now exactly what must be developed as well as the best procedure to realize their research goals. A full 1365 PE prototype system is expected to be up and running by the end of the year. - S. Myers
MuSiC++ (NTT Software Corporation)
Large Japanese corporations are well known for their conservative approaches to research, and at first glance, NTT Software Corporation (a wholly owned subsidiary of NTT), might seem to fall into this category with its development of MuSiC++. While this graphical language for creating communications systems seems rather staid, however, the package has broad applicability to the development of complex information networks (developments of which are becoming more and more common).
MuSiC++ is a CASE (computer-aided software engineering) tool being developed for telecommunications system software creation. If all goes as planned, it will have passed from development into production by the time this issue is published. Essentially, the MuSiC++ language is an enhanced version of the ITU-T standard of message sequence charts (MSCs), a method for graphing the control flow of complex real-time and interactive applications. "While, in the past, MSCs were mainly applied to telecommunications systems, as information networks become more complex, this kind of development system becomes more important," explains Hideaki Suzuki, a senior manager of NTT Software. "As multimedia networks and World Wide Web server networks grow, guaranteeing the integrity of the controlling system software becomes a big problem. That's where using the enhanced MSC language can aid a project."
Composing solutions to network problems
When designing network control software, development teams must avoid conflicts (multiple requests for a single network resource), deadlocks (a halt in network flow caused by an unresolved conflict), and other errors or bugs. This need for error-free software development is the driving force behind the movement to create a consistent environment that aids in software design.
For more than 10 years, corporations have been using MSCs to assist in development of software. In 1992, the language of MSCs was standardized by ITU-T and approved by other global standards organizations. The extensions provided by the MuSiC++ language expands what was a development tool into a full-fledged environment for creating and testing the flow control of network and real-time software. By using a simple ladder structure to illustrate precedence and dependence of multiple processes, MuSiC++ allows non-programmers to design functional descriptions of the interaction between various objects. MuSiC++ also can generate a specification and description language (SDL) file that describes in a standard graphical format the detailed specifications of each process and object.
Software for staying in tune
Communications and network software rely on proper timing, and in developing the software, real-time analysis is essential. Whether developing software controllers for telecommunications networks or routing software for distributed networks, assuring that each node behaves correctly is a Herculean task. MuSiC++ is designed to give project managers the ability to create specifications, while its SDL generation function provides software engineers with detailed specifications for development. Extensions to the environment that allow rigorous testing of SDL files by creating test suites in the international standard test suite notation, TTCN, are being researched.
As the "++" hints, the language is based on object-oriented methodologies. The jump from processes that signal each other to objects that send messages to each other was a short one, according to Suzuki. The hierarchical structure of the MuSiC++ toolset enables incremental development of different objects, which allows project managers to be as general or as specific as necessary when creating specifications for software engineers.
Other problems that the developers hope to address with the CASE toolset are those that arise from having multiple development partners working on a single communications or networking project. "We hope to give project teams a common specification language with which to communicate," says Suzuki. Both initial specification creation and later software maintenance benefit immensely from using the new environment.
MuSiC for the masses
NTT Software Corporation is ramping up to market the product. The push is to sell 100 or so systems in the first year, mainly to companies in the US who are designing systems for the telecommunications market. "We see the results of this project as being broadly applicable," stresses Suzuki. "Any process control system can benefit from the MuSiC++ development environment." With its American partner Anonymix, NTT Software Corporation will target developers of telecommunications and network control systems and companies creating software for Internet and Web server networks. - J. Stone
