Acorn User


Computer Translation Of Natural Language

Publisher: Sigma Technical Press
Machine: Archimedes A3000

 
Published in Acorn User #083

Universal Talk

Computer Translation Of Natural Language

This is more than just a book about machine translation. It contains a specification and an implementation of the translation system and a full program and sample dictionary listings to run in Basic on a BBC Micro-computer. Anyone who wishes to become more involved with machine translation can register as a user of this system and may even assist with further developments.

Part 1 of the book is a readable, informative and amusing overview of machine translation by Kelly. The reader is taken from where machine translation was 30 years ago (making interesting mistakes such as translating 'Hydraulic Ram' into the equivalent of 'Water Goat'!), through the ALPAC report (1966) which virtually killed off all machine translation research for a few years to the present day.

Part 2 by Goshawke details the 'Spoken Languages Universal Numeric Translation'. SLUNT has been specifically designed to run on any digital computer in the world by using an intermediate language or interlingua which consists entirely of the digits 0 to 9.

What does become clear in this section is that SLUNT is not yet complete and finalised. The basic framework has been specified but there is still more work to be done for translating to and from English, to and from other languages and for the handling of specialist vocabularies. Future developments will be co-ordinated and controlled by the authors, care of the Language Translation Group of the British Computer Society.

Part 3, the International Communication System (ICS), by Wigg, is an implementation of a sub-set of SLUNT, designed to run, primarily, on a BBC microcomputer. Some simplifications have been made to allow ICS to work within the limitations of a 32K machine. ICS is written in BBC Basic and a full listing, with comments, is given in the book. With minor modifications, which are also detailed, the program can also be run on an IBM PC. A checksum listing of the datafiles and programs are available on request or for the modest sum of £10 a BBC or IBM PC format disc can be purchased. For this price, the recipient will also receive at least four ICS newsletters, which will detail further developments with SLUNT and ICS, and a more advanced version of the software when it becomes available.

I have tested the software on a BBC microcomputer and within its limitations it works - albeit rather slowly. Loading the program from disc takes around 43 seconds, translating from number language into French in around one minute 15 seconds for two of the short example phrases supplied on the disc, and about one minute 18 seconds for translating the phrase 'George is speaking to the teacher' into number language.

This is, however, a limitation of the BBC rather than the software; a compiled version running on an IBM PC AT runs extremely rapidly and for those of you lucky enough to own an Archimedes 440 (hard disc version) those timings become 2.3 seconds, 4.1 seconds and 6.2 seconds respectively (and that's just interpreted Basic!).

One feature I did find inconvenient was the fact that the program is case sensitive; 'George' cannot be input as 'george' or 'GEORGE'.

I would say that this book is essential reading for anyone involved with or contemplating becoming involved with machine translation. The overview is readable, informative and entertaining! The formal description of SLUNT may be a little heavy going for the average hobbyist, but the listing are well documented and the software worked well.

The implementation is crying out for a more powerful computer to speed things up and to allow for some of the limitations to be relaxed. I would definitely recommend this book and software to anyone exploring this area of computer applications. However, if you don't have access to an Archimedes or IBM PC AT (or compatible) then bear in mind that the implementation is going to be slow to run and limited by memory.