It’s the computer speaking ...

Press, 13 May 1986, Page 24

It’s the computer speaking ...

From the “Economist” London

Having finished technical tests on the next generation of telecommunications equipment at Mitaka, NTT engineers are starting on the generation after that. First on the blocks is a computer database that can read out the information it contains over the telephone. Eventually, NTT hopes that it can use this technology to make things like computerised reservation services.

At the heart of NTT’s show-off database is a collection of newspaper clippings from “Nikkei Shimbun,” a leading financial newspaper. By pressing various keys on a

digital telephone, a user can search through the database for all articles that mention, for example, aardvarks, — or any other key word. Pressing other keys then instructs the database to send the article to a fax machine, a personal computer or a machine that will read out the article (or just its headline) to somebody listening on the telephone. Nobody at NTT expects much demand for a machine to read financial articles over the telephone, but the “Nikkei Shimbun” database is handy for experiments. NTT does expect a lot from the database’s underlying technology.

Because the telephone system of the future will hook up so many kinds of terminals in one network, it will breed increasing demand for equipment that will convert data from one machine’s format to another. And if the telephone could be made just another sort of terminal, whose information is freely interchangeable with fax or computers, it would eliminate much of the worry about getting computer terminals into people’s homes, or teaching them how to use the new technology. But translating from computer data to voice is not easy — particularly in Japanese. Unlike Western languages, Japan’s Kanji

characters are not pronounced phonetically. Moreover, pronunciation of Japanese changes with meaning, and the meaning of a string of characters can be completely altered by the last one.

To get round these problems, NTT’s voice telephone parses each sentence into its constituent words and phrases. Potentially tricky ones are identified, and the computer tries to work out from simple rules about context whether a given string of characters is, for example, the white gold a jeweller uses in his craft or the jeweller’s family name, and to pronounce them accordingly.