Software engineering has come a long way in the brief history of computers. But one problem still posing a major challenge to programmers is that of natural language translation - training a computer to handle English and other languages.

The topic came under examination at a conference at Cranfield Institute of Technology recently. It's still largely the preserve of universities but the interest of business and the possibilities offered by micros mean that it's gradually breaking out.

Most of the development work to date has been done in universities, and the field is jealously guarded by academics who seem bent on keeping knowledge shrouded in mystery.

Little progress seems to have been made in the last decade, mainly because there's no agreement over approach. No-one's been able to define clearly just how humans solve the problem. This isn't actually essential; you can get a computer to simulate complex activities without using directly analogous methods, but it would certainly be a help.

The latest trend is towards using a 'bridge' or intermediate language. This would mean that text would be translated in the bridge first, then into the target language. Such an approach would greatly reduce the number of programs needed for multi-lingual translations.

Moreover, it would provide for some form of standardisation. Several institutions are looking at using Esperanto as a bridge, while an English system called SLUNT uses a numerically-coded system.

Academia seems to be a brake on progress and it looks as if the most significant practical achievements will come from individuals and companies which want results.