Got a little downtime, but forgotten your knitting? Then try this fun activity, which is all the rage among some internet users with far too much time to kill. Call up an online translation page, think up a phrase, and have it translated first into another language and then back into English. Then sit back and enjoy the results.
‘Motherhood and apple pie’, for example, comes back as ‘Maternity and meat pie of apple’ on one site. ‘Mary, Mary, quite contrary, how does your garden grow?’ produces ‘Do Mary, Mary, completely contrary, how your garden develop?’ on another. It might not be up there with playing ‘Call of Duty: Black Ops’ on your Nintendo DS for excitement value, but it makes a change from updating your Facebook profile when you need a break from work.
But can machine translation be relied on for anything more serious than light diversion? According to the European Patent Office (EPO), it certainly can. Last November, the EPO announced an agreement under which it will use Google’s translation technology to translate patents into all the languages of the 38 countries it serves.
Translations for all
The aim of the collaboration is to remove the existing need for patents to be registered in one of the EPO’s official languages of English, French or German, and then translated at the applicants’ cost into the languages of all the countries in which they wish the patent to apply, an expensive, time-consuming and complicated hurdle to registering a patent in Europe.
That’s clearly a worthwhile ambition, but can the technology that produces ‘meat pie of apple’ be expected to come up with the precise, unambiguous translations essential to patents? Judging from the examples above, it could easily go horribly wrong, replacing procedural complexity with confusion and serious misunderstanding. Can machine translation ever be as accurate and useful as translation by a real, live human? Or, as one online translation service asks, ‘La traduction automatique peut-elle jamais être aussi précise et utile que la traduction par un vrai, de phase humain?’
It’s been a long time since my last French lesson, but dim echoes from back then tell me this is not a bad effort, particularly from a free online service. And of course, actual translations will be done only once, in one direction, avoiding the kind of glitches thrown up by translating one way and then back again.
Similar tests on sentences taken from genuine patents produce equally solid results, despite the fact that, with some patents, the original English itself could do with a lot of improvement.
Cutting costs
Of course, computer-generated translations will never be as good as hiring a trained professional, but, for those without the budget to do so,automated translation may soon be alow cost alternative. There will no doubt be teething troubles; however, the results of computer translationsare already reasonably good and are likely only to get better, becausethesystems improve as they have more material to refer to, which is precisely why Google signed up to the deal with the EPO in the first place. So instead of producing long lines of indecipherable strangeness, the result is likely to be simplicity, efficiency and lower costs.
And, in a way, that’s a bit of a shame. If automated translation were as unreliable as the examples above might suggest, it wouldn’t be long before the patent archives filled with bizarre, mistranslated, unworkable creations. New applicants and officials would be baffled about whether a new idea bore any similarity to an existing one, andthere would need to be a lengthy unravelling process that could provide many years of lucrative work for IP professionals. Instead, the ‘meat pie of apple’and other indigestible innovations will just have to remain a dream.
Richard Brass is a columnist for The Times. He writes on business issues for The Daily Telegraph and is a former editor of Punch.
