Information Technology Research Center
Kyoto, 617, Japan
Date Visited: 26 March 1998
WTEC Attendess: B. Croft (report author), R. Reddy, B. Davis-Brown, G.Mendel, R. Chellappa, R. Larsen
Omron is an international, diverse company with about $6 billion in annualsales and more than 22,000 employees. The main products are in the areas ofcontrol components and systems (more than 50% of sales), electronic fundtransfer systems, public information and traffic systems, medical andhealthcare equipment, and open systems. Products related to digital informationorganization technology would come under the last category. The four coretechnologies originally identified by the company are fuzzy technology, sensingtechnology, computers, communication and control technology, and life sciencetechnology.
Recently, "humanmedia" or "flexible intelligence" technology to support"human-oriented computing" has been identified as a new direction. Thistechnology is being developed both to support other Omron products, such asticket sales machines, and as stand-alone software products. The goal ofOmron's humanmedia technology is to create "human-oriented equipment" toachieve the "Optimization Society," where people can enjoy work and privatelives suited to their individual needs. Although digital libraries are ofpotential interest to the Multi Media Group within the information technologyresearch center, Omron currently has no products in that area or plans todevelop such products. Some Omron software products, however, are importantcomponents of current commercial retrieval systems in the Asian markets.
In terms of government funding, Omron receives funding from NEDO, whichprovided ¥22 million for the FY 1998 and was expected to provide ¥24 millionduring FY 1999.
Omron's "humanmedia" technology is a combination of language processingtechnology, speech recognition technology, image processing technology, and"kansei processing technology." Kansei is a term that the WTECteam encountered frequently during this visit and is related to the developmentof intuitive interfaces and to the concept of affective computing proposed bythe MIT Media Laboratory. Each of these technologies is summarized in thefollowing sections.
The topics that were discussed here include morphological analysis,multilingual input and language recognition. SuperMorpho-J is a Japanesemorphological analysis system that is used for word segmentation andpart-of-speech tagging. The system has high segmentation accuracy (about 98%)and can process more than 1 GB of text per hour. The system can also deal withsome issues specific to Japanese such as word variants and word breaks at theend of the line. Versions of this system have been developed for Korean andChinese. Supermorpho-J is being used in the Japanese versions of Infoseek andVerity's search engine. Although segmentation and different character encodingsare often described as significant problems for Asian languages, it was clearfrom this discussion that these issues have been largely dealt with and are notan obstacle to multilingual retrieval systems.
Wnn is a well-known input system for Japanese and Chinese that wasoriginally developed in the UNIX environment. This system has been steadilydeveloped as a product and ported to Windows and Java. The system now alsosupports multiple languages in one document and many encodings, includingUnicode. Once again, from the point of view of multilingual digital librarysystems, it is clear that input will not be a major issue.
A speech recognition project based on word spotting was described. Wordspotting was preferred to large vocabulary language model approaches due to thepotential for better robustness in noisy environments, such as informationkiosks or ticket machines. Omron is now focusing its technology on Japaneselanguage, and a detailed explanation was not provided.
An image retrieval project was described as Omron's part of a nationalproject supported by NEDO. More than 13 companies, laboratories, anduniversities are involved in 3 working groups. The group that Omron wasinvolved in is "kansei agent and human media database." The imageretrieval project is viewed as a means of developing a better understanding ofkansei rather than being leading edge research in image processing. Thedefinition of kansei used by this group is "subjective criterion inhuman information selecting." Images from a database of postage stamps areindexed using four types of composition templates: horizontal, vertical,circle, and radiation. Combinations of these templates are being used forretrieval experiments, and excellent results have been obtained for the domain(postage stamps).