As often in such situations, researchers are encouraged to do basic research, and then asked to produce realistic, nearly preoperational prototypes, two mutually incompatible activities. Other problems, specific to ATR, are that researchers usually stay for less than 3 years, which severely limits the "memory" of the team; and that the proportion of trained linguists is very small, which leads to research being done more on computational problems than on problems specific to MI. Given these circumstances, and the expectations of the scientific and industrial communities, what course could ATR-IT take for the second half of the allotted time of 15 years?
This report has been prepared during a short stay at ATR, on suggestions by MM. Kurematsu, Morimoto and Iida. It contains an account of detailed discussions held with a number of researchers at ATR Interpreting Telephony Research, followed by an analysis of recent research at ATR, based on these discussions and the reading or rereading of relevant literature. Finally, some directions for future basic research directly related to the overall (engineering) goal of MI are suggested.
Keywords & abreviations
Machine Interpretation (MI), Machine Translation (MT), Example-Based MT (EBMT), Dialogue-Based MT (DBMT), Natural Language Processing (NLP), Artificial Intelligence (AI), Bilingual Knowledge Banks (BKBs), Speech Recognition (SR), Speech Synthesis (SR).