People these days take Google Translate for granted. But believe it or not, automated translation hasn’t even been offered for a decade.
Researchers have been trying to develop machines that would produce a translation with 0% added human since the 1950s, when IBM cobbled together a system they cheerfully claimed would soon be able to translate 2 million words of Russian per hour. We’re in 2016, and the world still needs Russian translators. If anything, there is a serious shortage of professional Russian linguists.
A good language learner sees the challenge as a series of patterns. Automatic platforms also look for patterns, but they rely on computer databases. The translations come about as a result of statistical analysis, rather than application of grammatical rules – that is, they look for patterns between the source and target language that are statistically significant.
Because they rely on probability, using a Machine Translation is always a gamble. It is worth noting that these crucial databases consist of the millions upon millions of documents that have already been translated by human professionals.
Of course, any exasperated language learner can also tell you that to every pattern in a language, there is an exception. So the accuracy of machine translation is a percentages game – a case of infinite typewriting monkeys rather than skilled polyglot elves at Google HQ. So while the first generations of ‘internet native’ humans get more and more comfortable with trusting computers to perform such tasks, it’s worth having a look at what Machine Translation can and can’t do.
1) Google Translate
Reader, Google, I believe you’ve already met.
• Google Translate is free and fast.
• It is integrated with other Google services, like its in-browser translator for Chrome.
• Coverage is probably Google’s big strength. In February 2016, the tech giant announced that it was adding 13 new languages to its service – such as Amharic, Pashto, Kyrgyz and Xhosa. This gives a total of 103 different languages that Google claims to translate to and from, which largely accounts for its dominance.
• You can download language packs for offline use when abroad.
• Source-language auto-detection.
• Since the acquisition of Wordlens, the software now allows instant camera-detected translation for some languages.(Image courtesy of Google)
• Google Translate is only as good as its database. And that database varies massively between language pairings. So the chances of a good translation of English into Japanese are much lower than a good translation into English, because Google is not the big-name search engine in Japan. And if you’re trying to translate your business brochure from Hungarian into Hindi, well, just give up and go home. Since when did people take quantity over quality?
• There is no speed-slider on its loudspeaker function. If you didn’t catch it, you’ll have to make do.
• While it copes with short sentences, it often gives you word salad when you add in winding phrases, ambiguity and, in particular, humour. All of which are, sadly for Google, an inescapable facet of human language.
• The output can be quite embarrassing. Also, you can’t get feedback from the translator, nor can you edit it to suit your style and goals. For example, one goal of your company might be to provide accurate information on your product to avoid liability.
Seems fine, right? Well, it depends how careful you need to be. “Des noix” is not the generic term for nuts in French, which is “des fruits à coque” (‘shell-fruits”). It refers specifically to walnuts. As oversights go, that’s a pretty dangerous one. See you in court.
Microsoft’s inevitable foray into the murky world of Machine Translation
• Like Google, it is free and also instant.
• It is integrated into all new Windows operating systems and Windows phones.
• Users can vote on the accuracy of the translations given.
• Source-language auto-detection.
• While the 56-strong language list is somewhat smaller than Google’s, their selection is nothing if not eclectic. Think Hmong Dau, with 4 million speakers in provincial China, or Queretaro Oromi and Yucatec, two Native American languages with perhaps a million speakers between them. This might seem like a gimmick in your Anglophone eyes, but if you happen to be a Yucatec Mayan, Bing is a godsend.
• Like Google, it delivers maddeningly erratic quality across all of its language pairs. Pairs not including English are likely to be very volatile.
• It may cope with simple language. With anything more complex like nuanced meaning, slang, or specialised terminology, you’re in trouble. Bing rarely even agrees with its own translations; just take a look at the Translation Party widget, which runs Bing’s own translations back through the system into source and target language.
2.5) Skype Translate
This doesn’t get its own number because it is Microsoft powered – but Skype Translator goes one step beyond with the same technology. What it effectively seeks to do is to connect people by being an interpreter – bringing real-time text and voice translation to your desktop.
• It offers a selection of important world languages – English, Spanish, French, German, Italian, Mandarin, and as of March 2016, Arabic – the pan-Arab Modern Standard form to get around the issue of diglossia.
• There is an onscreen ‘third-party’ – the translated text is generated real time to complement the voice synthesis.
• As more people use the software, the system should adapt with more efficiency to the idiosyncrasies of human expression. In theory, at least.
• The speech synthesis comes with two speaker options – ‘Jane’ and ‘Bob’. How glorious a thing choice is.
• There is an inbuilt profanity filter with several degrees of strictness that can prevent your foul language from reaching delicate foreign ears. Although I’m not sure if this isn’t actually a joykill.
• The technology is still only in preview mode, so the quality of the translations across certain language pairings will be shaky to say the least. Nonetheless, it is an impressive feat they are attempting.
• Translating real time speech is a much more significant challenge than a plain text, because of the nature of our conversational habits; our phrasal units are rarely full sentences, and we often rely on subtle variations of interruption, tone and rhythm to convey meaning that must be particularly difficult for an AI machine to identify as significant.
• Therefore you have to be kind to the machines, and speak to your interlocutor as if they were a confused child.
• The speech synthesis option is, predictably, lacking the dulcet tones of real-life human voice.
• Only available to Windows 8.1 users.
The lapidary summary of the MIT Technology Review captures it perfectly:
“In theory, Skype translation could be transformative. It’s like a version of the discreet live translation that world leaders enjoy when visiting the United Nations. In practice, though, it can be more like having Apple’s Siri constantly interrupting your conversation and talking over you.”
The fact remains that no matter how rough and ready the output may be, the ambition and realization is fairly remarkable given the limitations of a free service.
3) Panasonic’s “Megahonyaku” megaphone
A portmanteau of the words megahon (megaphone) and honyaku (translate), this is a very cool concept indeed. It was dreamt up in 2014, after officials struggled to communicate with stranded travellers in the wake of devastating floods. Panasonic’s prototype, which has been tested by airport officials in Japan, is designed to produce real-time translations from Japanese into English, Mandarin, and Korean. Yes, you speak in Japanese, and the gadget makes you quad-lingual. Sort of.
• You can communicate with whole crowds who don’t speak the same language.
• You will win in the futuristic street-cred stakes.
• You might use it to have a polyglot dinner party, with no chance of guests talking over each other.
• It’s not commercially available, and they’re only aiming to use it during the 2020 Tokyo Olympics. Don’t bother asking how much it will cost; if you need to ask, you can’t afford it.
• It is presumably quite good at translating phrases like “Please wait”, and “Your bus will be here soon”. But don’t expect to see this kind of hardware replacing interpreters at international summits anytime soon.
• One input only. You need to be able to speak Japanese to use it. Sayonara!
In sum, Machine Translation is here to stay, and it certainly has some uses. Future generations will never again have to gesticulate wildly about “el chips por favor” at bemused Spanish waiters, nor will they miss their flights because of indecipherable Korean subway signage. But those situations are quite distinct from the service that translation professionals provide. What Machine Translation is replacing there is the individual’s language learning for pleasure. Human language is full of complexity, nuance, and beauty.
Its potential lies well beyond the frontier of Artificial Intelligence. The idea that machines could one day replace human professionals is risible because the stakes are always too high for the risk of terrible translations.