Site: Toshiba Research and Development Center
1 Komukai, Toshiba-cho
Saiwai-ku, Kawasaki 210, Japan
Date Visited: May 26, 1995
Report Author: T. Sheridan
Toshiba was founded in 1875 as Tanaka Engineering Works, became Shibura Engineering Works, combined with Tokyo Electric Company in 1939, and became Toshiba Corporation in 1978. At the time of the JTEC visit, it employed 74,600 persons and had net sales of $31.6 billion. The fraction that is information and communication systems as compared to heavy electrical apparatus has continued to grow and was 54% of operations in 1993. Toshiba is a principal supplier of boiling water nuclear power reactor systems. It employs 1,500 scientists and engineers, including 200 women. R&D expenditures are 8.5% of sales.
Sadanori Yoshimura presented the center's development of an operator training support system for nuclear power plants (Ohtsuka et al. 1995). Trainees undergo emergency response exercises in a full plant simulator. Operator responses, alarms, and plant parameters are evaluated in combination with respect to a computerized knowledge base. By preestablished criteria, errors are counted and stored in a database for later analysis or playback for operator training.
Yoshimura made a second presentation of what Toshiba calls an "ecological interface" for nuclear power plants (Sakuma et al. 1995). By ecological is meant an interface that accords with a taxonomy of human mental functions developed by Professor J. Rasmussen of Riso Laboratory, Denmark, who used the term to characterize what are asserted to be natural hierarchical relationships in human cognition in relation to task performance -- a mapping between high-level goals, through functional abstractions such as energy and mass balance, and finally to physical phenomena and task motor actions. It amounts to a philosophy for designing and evaluating procedures, displays, and controls.
Dr. Takebayashi demonstrated the center's speech recognition activity (Takebayashi 1992; Takebayashi et al. 1991; Takebayashi et al. 1993; Kanazawa et al. 1994; Kanazawa et al. 1995). This is a major effort, dating from 1992, that constructs a spontaneous speech dialogue system. This system consists of a noise-robust wordspotter, a syntactic and semantic parser, a user-initiated dialogue manager, a multimodal response generator, and a speech response chancellor. Dr. Takabayashi and his associates demonstrated the system in the context of ordering a hamburger at a fast food establishment (!).
They also discussed briefly the application of some of these ideas to document retrieval, using keywords in relation to semantics. A prototype system automatically generates abstracts based on the user's inputs.
Toshiba has had a Japanese language processor since 1977. The company is very interested in applying this technology to groupware.
Finally Miwako Doi demonstrated a virtual reality simulation of an operator or a team of operators performing procedures in a nuclear power plant, presumably useful either for engineering development of procedures or control room design (displays and controls) or for training of operators.
Kanazawa, H., S. Seto, H. Hashimoto, H. Shinichi, and Y. Takebayashi. 1994. A user-initiated dialog model and its implementation for spontaneous human-computer interaction. International Conf. on Speech and Language Processing, Yokohama.
Kanazawa, H., M. Tachimori, and Y. Takebayashi. 1995. A hybrid wordspotting method for spontaneous speech understanding using word-based pattern matching and phoneme-based HMM. IEEE 0-7803-2431-5/95.
Ohtsuka, T., K. Yahuchi, S. Yoshimura, Y. Takada, and K. Noji. 1995. Development of an operator training support system for nuclear power plants. In Proc. HCI International '95 (6th International Conference on Human-Computer Interaction, 9-14 July 1995, Yokohama, Japan).
Sakuma, A., J. Itoh, E. Yoshikawa, and K. Monta. 1995. Simulation study of an ecological interface for nuclear power plants. Proc. HCI International '95 (6th International Conference on Human-Computer Interaction, 9-14 July 1995 in Yokohama, Japan).
Takebayashi, Y. 1992. Speech recognition based on the subspace method: AI class-description learning viewpoint. J. Acoust. Soc. Jpn. (E)13(6):429-39.
Takebayashi, Y., H. Tsuboi, and H. Kanazawa. 1991. A robust speech recognition system using word spotting with noise immunity learning. Proc. ICASSP 91 (Intl. Conf. on Acoustics, Speech, and Signal Processing, 14-17 May 1991, Toronto, Canada).
Takebayashi, Y., H. Tsuboi, H. Kanazawa, Y. Sadamoto, H. Hashimoto, and H. Shinichi. 1993. A real-time speech dialog system using spontaneous speech understanding. IEICE Trans. Inf. and Syst. E-76D(1)(Jan).