According to The Asahi Shimbun, the megaphone-shaped automatic audio translator was developed by Panasonic Corp. It takes the gadget only about a second to translate a phrase.
“Passengers with large luggage are advised to take elevators,” a station employee speaking in Japanese said into a megaphone by a ticket barrier at Tokyo’s Haneda Airport International Terminal Station.
A screen on the megaphone then displayed the same sentences translated into English, Chinese and Korean.
At the push of a button, the phrases were replayed in the foreign languages one after another. “If you have a suitcase ...,” an English voice began saying.
The scene was part of a public test that Keikyu Corp., the railway company that operates the station, conducted in early October.
“As far as travel phrases are concerned, the device could get a score of 600 or so in the Test of English for International Communication (TOEIC),” a Panasonic official said.
“We are hoping to enhance the precision to an equivalent of a TOEIC score of 700 by 2020, while at the same time reducing the time requirement for translation.”
On-the-spot speech translation is on the cusp of becoming an everyday occurrence in Japan, with an increasing number of public transport operators testing out the state-of-the-art technology.
The public and private sectors are working jointly to enhance the technology’s precision ahead of the 2020 Tokyo Summer Olympic Games, which are expected to draw an increasing number of visitors to Japan over the coming years.
The technology is expected to achieve a level where translating simple sentences, such as travel phrases, causes little inconvenience, sources said.
A number of other public transport operators have also introduced automatic translators to cope with non-Japanese tourists.
Tokyo Metro Co. has been keeping tablet computers with an automatic audio translation app installed at ticket gates and offices of most of its stations since August 2015. The app, which handles about 30 languages, can translate, convert and play a phrase in audio in one-to-two seconds.
Keikyu provided almost all of its stations with similar devices in February this year, and Keisei Electric Railway Co. followed suit in March.
All those devices are based on automatic audio translation technology of the National Institute of Information and Communications Technology (NICT).
With the expectation that the Tokyo Olympic Games will prompt a growth in the number of tourists from abroad, the Ministry of Internal Affairs and Communications, which supervises NICT, has defined a five-year term through fiscal 2019 as a research and development period, and has so far been spending more than 1 billion yen ($8.8 million) every year on the project.
Members of a council, which organizes 152 entities including NICT, public transport operators and manufacturers, have repeatedly conducted demonstration tests with the aim of improving accuracy and achieving practical operability.
The basic process of automatic audio translation can be summarized as follows.
Suppose that Japanese speech has to be translated into English.
First, voice recognition technology is used to convert the Japanese speech into text, sentence by sentence.
Next, a “dictionary” of sorts, containing a huge amount of equivalent Japanese and English text pairs, is used as a reference to convert every single word in an original sentence into a corresponding English word, and to rearrange the word order to form a sentence that is the most likely to represent a correct English word order. The resulting sentence is finally synthesized into speech.
Automatic audio translators are already performing almost as well as human translators in subject areas where extensive dictionaries are available, sources said.
“Our goal is to achieve a level where there will be no problem in everyday conversation in the fields of travel, health care, disaster management and living in 10 languages, including English, by 2020, the year of the Tokyo Olympics,” said an official with the research and development office of the communications ministry.
“Our dictionary currently has only about 25 million text pairs, which is far from sufficient,” said Eiichiro Sumita, director of the NICT Multilingual Translation Laboratory.
“About 500 million translated sentences are being generated annually in Japan. If they could be collected into a dictionary, that would help create a society where automatic translation will suffice, except with people who need to use foreign languages in their work and for other purposes.”
If you want to read this article in Japanese, please see the following link:
Subscribe to our English Newsletter