Using the system, the speaker would speak in his or her own language and the speech would be translated into the language of the listener. According to Docomo, the processing time for the translated sound bytes is approximately two seconds. At present the translation system covers Japanese, Korean, Chinese and English.
Regarding translation between Japanese and English, the system is around 90 percent accurate in understanding the Japanese side of the conversation, and about 80 percent accurate in understanding the English side. Those results might sound impressive, but there is an enormous amount of misunderstanding that can arise in that 10-20 percent that the system cannot understand. To be used practically, conversations will need to be kept simple.
Docomo envisions potential applications as being tourism, retail, health care and education. While these sound like applications that could be complicated, the language requirements, for the most part, can be kept simple. Listing a range of symptoms to a doctor is a fairly difficult exercise in a person’s non-native language–but when the unfamiliar vocabulary is taken out of the equation, the basic grammar structures used are very simple. Where Docomo’s translation technology is likely to fall down is when it is applied to general conversation–where language tends to be a little more nuanced and a lot less direct.
At present, the technology is being tested by 400 monitors. Tourist facilities, retail companies and hospitals are also involved in the trials. Tests end at the end of March. Depending on the results, we may see the technology opened to the general public after fiscal 2012 (which in Japan means after April 2013).