Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
4 The Research and Development Policy Agenda The discussion above highlighted the special needs of U.S. users of Japanese to English machine translation for high-quality, general-purpose systems and the research and technical challenges in developing such systems. What should be done and who should do it? The sections that follow review a range of perspectives on these issues discussed at the symposium and lay out alternative approaches to the formulation of a policy agenda. THE SKEPTICS Despite the apparent growing need for translations of Japanese technical documents, and significant efforts by members of Congress and others to expand U.S. government efforts in this area, skeptics question whether there is real demand for such information. A number of commercial ventures to provide translated Japanese technical literature have failed and many question whether the U.S. research and business communities will overcome a "not invented here" syndrome the assumption that technical information from abroad is second rate. At the same time many suspect that the information being made available is out- of-date and of little technical interest. A number of observers question whether the United States government should attempt to mount a major initiative on machine translation. The case against doing something, particularly at the U.S. government level, is based on uncertainty about the need to target this set of technologies for special attention. 30
31 Despite a decline in U.S. funding for machine translation, the United States maintains a solid research base in machine translation. Skeptics, moreover, question whether Japan has really taken the lead. Instead, they say that the Japanese have Bushed prematurely to market machine translation systems that need much more work and would not be attractive to the typical U.S. user. Therefore, the argument goes, there is no justification for a U.S. government effort. If the demand is there and the technical problems are amenable to resolution, they say, U.S. industry should take the lead rather than calling on the government to support costly and uncertain research.33 Furthermore, support for R&D on machine translation must be seen In broader context. Some believe other technologies deserve more attention and may have quicker commercial payoffs. In an era of tight federal budgets, how can the case be made that machine translation deserves special attention? Nor is machine translation a panacea for widening U.S. access to Japanese scientific and technological information. It is one piece of a large picture. Efforts to expand Japanese language study, harmonize patent systems, and reward young scientists and engineers who spend time in Japanese laboratories are also important, sometimes competing themes in the national debate. Professional translators, moreover, worry that advocates of machine translation believe that it will be possible to eliminate "the most expensive link in the communication chain"-the skilled bilingual.34 Even those most committed to work on machine translation note the sociological problems that plague this enterprise. Viewed from this angle, no amount of R&D funding will by itself resolve the fragmentation of the research, development, and user communities in the United States. If machine translation is going to be supported, the skeptics say, it should be as part of a larger national commitment to building networks of interaction among researchers, business people, and professional translators. THE ADVOCATES In response to those who question whether the demand is "real" for translations of Japanese technical literature, advocates of new policy initiatives note the lack of follow-through in previous initiatives. Some believe that He 33 Skeptics note that most machine translation systems in operation today are based on logic programming, the expert system approach. Marshall Unger argues that recent breakthroughs in artificial intelligence (those obtained in the neural-net paradigm, also called connectionism) offer the most promising direction for future research and development. 34 As discussed above, it would appear that such concerns are no longer valid in view of changing perspectives on machine translation.
32 Office of Japanese Technical Literature, created by law in 1986, has been downgraded within the Department of Commerce and given inadequate funding.35 Federal agencies could, using machine translation, do a better job disseminating existing data on Japanese science and technology to the public, research, and business communities alike. Machine translation, used as a gisting tool, does work that is simply beyond the capacity of human translation. If machine translation were developed into a systemized service, the result could be to stimulate new demand for Japanese technical information. If users do not know it exists or that it can be obtained at reasonable or no cost, demand will remain low. It is not enough to say: let the marlcet be the judge. Markets can fail to meet the national interest, providing a legitimate rationale for government action. In Japan, government and business are cooperating to support a long-term machine translation development effort. They are, in effect, creating a market. Because the needs of users in the United States are different from those in Japan and cannot be fully met by the Japanese to English machine translation systems now in operation, some believe that our needs cannot be met by simply buying from Japan.36 There are broader motivations for advancing support for machine translation in the United States. In Japan and Europe, machine translation is seen as an important component in the new range of information technologies for the 21st century. As a technical challenge, Japanese to English machine translation represents an engineering stretch that may spin off progress in other areas of information technology, one that requires integration with other fields of information technology R&D. Viewed from this perspective, machine translation is a test bed for solving natural language processing problems in information technology. The question, advocates say, is whether the United States can afford not to pursue this technology. The potential impacts of a focused effort on machine translation extend beyond information technologies. Bnnging together researchers, developers, and users in a long-term project would provide an opportunity for experimenting with new approaches to R&D organization. There is a striking fragmentation of machine translation efforts in the United States ineffective communication between scientists working on basic research and the engineers who apply it; weak links between the developers and the potential users. This lack of coordination makes machine aanslaiion a very fragile technology in the United 35 Comments by Congressman George E. Brown Jr., on the "Federal Role in Accessing Japanese Technical Information" at the Symposium on Japanese to English Machine Translation. 36 At least one participant in the symposium, however, argues the case for simply buying from Japan. As long as the technology is readily available to U.S. customers, the argument goes, it makes sense to take full advantage.
33 States today. There are also deeper issues at stake such as whether we should be satisfied with a situation in which the United States produces the best research papers and the Japanese produce the products. WHAT MIGHT BE DONE? In developing a policy agenda for machine translation in the United States, there are a number of levels at which initiatives can be taken. Initiatives could be taken at the international level. Second, a national policy effort could be defined. And third, individuals and organizations could define new approaches. The suggestions outlined below are not mutually exclusive, however, policymakers who see the need for action will also need to determine where scarce resources can best be invested. International cooperation in machine translation is desirable in order to develop expanded and up-to-date bilingual dictionaries. It also makes sense in view of the large expenditures required and the "precompetitive" nature of some of the more basic research needed to lay a solid foundation for product development. Dr. Makoto Nagao, a leader of Japan's machine translation efforts, has intitiated an International Association for Machine Translation that will bring together researchers' developers, and users in North America, the European Community, and Japan. The purpose of the organization would be to collect and compile information on machine translation so that users have a better understanding of what types of systems are available, how they can be used, what the experience of others has been, and improve evaluation of machine translation systems. For the developers, the benefits would include cooperative approaches to dictionary building, database construction, and sharing of texts as well as expected improvements in theories of machine translation and in machine translation systems. Users and developers alike could benefit from the development of standards for input, design of controlled languages, and evaluation. Conceived as a federation of regionally-based organizations, the international organization will be supported by membership dues (including both individual and corporate members) and other sources, including government funding. In the context of U.S.-Japan relations, cooperation in machine translation has special significance. Machine translation has been identified by both countries as an area for potential cooperation under the United States-Japan Agreement for Cooperation in Science and Technology. Machine translation appears to be well suited for U.S.-Japan cooperation for a number of reasons. First, the precompetitive aspects of the R&D challenge are significant. Second, a long- term effort is needed. Third, the costs are high and the commercial payoffs over the horizon. Finally, the United States could potentially learn from the efforts already under way in Japan, provided that the participants include senior
34
35 . US US Japanese Univer- US Industry Industry sities Machine US Translation GoVt Firms motivation resources - financial - organizational 0 expertise - core machine ~I translation technology - machine translation production development user-friendly interfaces user feedback develop pitfalls large-scale databases o format handling - integration ~A o platforms networks o database management systems graphics 0 scanners - high-quality MT ? ? technology * Note contnbuiions are also made by Japanese universities and government but Japanese industry has all of these resources. FIGURE 8 Sources for J-E development SOURCE: Bernard Scott. Technical Information Service for accessing and translating Japanese language documents could serve as a foundation for this effort. Machine translation could be a test bed for a focused, U.S. national initiative in information technology development. Developing procedures for rapid processing of requests to use copyrighted materials is also essential, as well as agreements that encourage timely dissemination and low cost or free usage of jointly developed dictionaries and other tools. At another level, policy initiatives can also be developed at the organizational or corporate level. Those with long experience in providing Japanese language technical information argue that a commitment is necessary to create and maintain demand at the organizational level. MCC's International Liaison Office,
36 the information specialists at Bell Laboratories-these are the people and places where demand can be developed internally. Building better ways to organize, disseminate, and utilize Japanese technical literature provides, therefore, the context for machine translation. Building a constituency of users for translated documents means working directly with the users to tailor the search, selection, and output to their needs. Unfortunately, only the largest organizations today have invested the resources needed and only a few have made expanding access to foreign technical literature a high priority. CONCLUSION ~ range of options is available to policymakers (public and private) who determine that initiatives in machine translation should be pursued. These options international collaboration with the European Community and Japan, a U.S. national project, government support for stronger linkages between university research centers of excellence and potential developers, a focused government procurement strategy designed to build machine translation expertise in the United States are not mutually exclusive. It is, however, not clear that pursuing an international cooperation strategy would by itself build a fully integrated research and production base in the United States or a new generation of machine translation systems that fit the special needs of the U.S. user community. Machine translation, broadly defined, offers interesting research and development challenges in building new information technologies. It also offers significant potential rewards to businesses operating in global markets, to researchers and engineers who need to know what is going on in Japan, as well as to translators whose work can be augmented by the use of machine translation tools. Machine translation is not a panacea or a fully developed technology-but it is a dynamic and challenging area worthy of serious policy consideration.
