Language is what separates us from the rest of the animal kingdom. Not a particular language, like English or Bengali, but the ability to learn a language and use it to communicate with others and even with oneself. This last possibility is very important, because we probably use language for thinking too. Do animals think? I don’t know, because animals don’t talk, so we can’t ask them.

The obvious way to translate a source language to a target language is to find the equivalent word and substitute it. Different languages have differing word-order. For example, the red box in English becomes the box red in French. Never mind, re-order the words according to a fixed set of rules, and hey presto! You have a translation.

When I was working on a computer in the 1950s, my erstwhile professor came to visit me in my lab, as erstwhile professors are wont. I was a little bit in awe of the great man, and I knew he was working on a project for doing computer translation. Needing a subject to open the conversation, I asked how the work was progressing.

"We nearly have the problem licked. Our computer only has 4,000 words of memory. If it had 8000, I think we could do it."

Space forward 60 years, and computers now have a million times as much memory, but the problem is as unsolved as ever.

The reader will be astonished. He will protest that computer translation is now an established fact. Google translate is on my computer screen, with an invitation to click on it.

An item in yesterday’s newspaper describes a court case, resulting from the use of a fork-lift truck by an unfortunate Polish worker, who caused himself serious injury because he misused the device. His employers had indeed given him the safety instructions, but they had entrusted the translation of those instructions from English into Polish to a program, which had resulted in gibberish. The victim claimed that he had never been warned.

All this is not surprising when you consider the nature of a translation. Far from being a word substitution process, a translation is what would result from someone unacquainted with the target language, communicating in the source language with a translator, who having understood, then communicates the idea he has received to the target listener, who understands only the target language. The translator must understand before he speaks.

To make a translating machine, it is therefore necessary to have an understanding machine. We don’t have one yet.

Perhaps a couple of concrete examples are in order: The Director General of CERN, the European science centre sent a circular to his staff congratulating one of the scientists on a brilliant research breakthrough, and in particular praising his entrepreneurial effort in setting up and executing the experiment. A French translation of the circular appeared on the back of the sheet. Unfortunately, the word "entrepreneur" doesn’t mean the same in French; it means a fixer or operator, with pejorative implications, The good director had to send out a second memo, with an apology for the translator’s gaffe.

I once received an offer for an item from a Swiss firm. It quoted a price "including tangled mass". A quick check with Swiss colleagues elicited the surprising fact that Wertumsatzsteuer abbreviated WUST, is the local word for Value Added Tax. The unfortunate salesman had looked up wust in his dictionary and found it to be the word for what a kitten does with a ball of wool. I can imagine an English salesman doing the same thing by quoting his prices "inclusive of barrel"!

Even human translators aren’t infallible!

I would not be so foolish as to maintain that mechanical translation of languages is impossible. We humans are very clever. But to maintain that you can do it marks you as someone who doesn’t understand what it is you are attempting. Before you make a translating machine you have to make an understanding machine. And we don’t know how to make one of those.

Frank Beck