Vision misinterpreted

The prediction by Eric Atwell in The Language Machine is unduly alarmist (THES, July 23). The technology required for a machine to translate the spoken word (ie interpret) involves a combination of voice recognition and translation software that will not be within reach within the five years he cites.

Voice recognition software is currently slow and frustrating. It certainly helps those who cannot use their fingers to input material.

Translation software, however, is useful only as a help to either a translator or to a client who wants to decide whether material is relevant and needs to be translated.

Only standard simple letters, forms or questionnaires are successfully translated automatically and even then it is prudent to check the result. I have yet to be impressed.

The long leap from translation to interpreting is more than five years away except again in the case of standard responses (as in simple questionnaires requiring Yes/No answers and the like).

Henry Pavlovich Director Institute of Linguists