Lost In Translation : Google Faces Significant Challenges To Build Babel Phone

The news that Google was working on a phone that could one day become the equivalent of the Hitchhiker's Guide to the Galaxy's Babelfish, opened some tantalising possibilities.

It was hailed by the Sunday Times as a technology that could eventually transform communication amongst users of the 6000 or so languages and dialects worldwide. It could even help sustaining and reviving dying ones as well.

But the reality is slightly more sobering.

For a truly translation smartphone to become commonplace, Google will have to solve two fundamental obstacles that have thwarted the efforts of some of the best engineers worldwide for decades.

First, there's the issue of voice recognition, something that Google and others are working hard to crack. While the main languages are fairly well recognised, we are quite some way towards a ubiquitous voice recognition engine that can not only recognise the thousands of voices but also the countless variations in accents and pitch that make one's voice unique. Nuance is currently the market leader after having acquired assets from Spinvox and IBM.

Then there's the task of actually converting the text into another language. Everyone who has used either Google Translate or any dictionary knows that while they work well most of the time, their usefulness is limited to word-by-word translation.

Anything too complicated and the translation engine goes berserk and delivers content that's very often out of context. There's also the fact that only one percent or so of languages are catered for, that's still 3600 possibilities. Getting the 6000 or more languages in the set would imply nearly four million permutations.

Of course, the phone will also have to deliver the goods, that is, convert the text into speech via a voice synthetiser. Electronic pocket dictionaries do a fairly lousy job at it and unless you invest in a complex but expensive language tool, the results are likely to be below your expectations.

If that wasn't difficult enough, all the calculations will have to be performed in quasi real time by a mobile phone whose computing power is roughly equivalent to a 10-year old computer.

This can possibly be resolved - at least in some places - by making use of Google's infrastructure but you'd need to be connected to access it. Bearing in mind that there's no mobile network reception in areas like Lake District, that would only be a palliative solution.

To make real-time, universal translation happen therefore would require a massive improvement in processing power on smartphones. Before this ever happens, one can expect translation and text to speech solutions to appear on sites like Youtube.