The Hazards of Machine Translation and Beyond
The Jackie Chan bus stop, a restaurant called Translate Server Error, wife cake, children sandwiches, wide-boiled aircraft – they sound like comical lines at a stand-up show. But, in fact, they’re real-world examples of glaring mistakes and the hazards of machine translation.
For big firms, poorly translated text can have major consequences such as the risk of offending potential customers or losing business. Although we often hear promises of machine learning tech that will process language translation effortlessly and naturally, just as a real [human] translator would.
But when will such technology be available to businesses?
Skype and its real-time translation upset
Last January, Skype launched its real-time translation tool which instantaneous voice-to-voice translation in seven languages. However, it suffered heavy blows when users complained that it turned ordinary Mandarin words into obscenities. The glitch came to light during the shooting of a Skype commercial in China; apparently even the simple phrase ‘it’s nice to talk to you’ translated into offensive swear words.
Google Translate, the traditional approach
Translation tools like Google Translate have traditionally been built around phrase-based statistical machine translation. Machine translation works by analyzing a back catalogue of texts already translated, such as academic papers and glossaries.
The texts are analyzed in parallel – both original and target languages. Using statistical probabilities, it selects the most appropriate translation to the phrase submitted; the better the quality of the original language, the greater its effectiveness. But it’s prone to howlers, like the ones mentioned earlier. Often the translations sound mechanical and dull.
The End of Machine Translation?
Alan Parker, director of engineering language technology at Facebook, recently commented on statistical machine translation reaching “the end of its natural life”. It has been said that translation technology is on path towards artificial neural networks much like the neural pathways of the human brain.
These neural networks are structured similarly to the brain, using complex algorithms to select phrases appropriate to the translation. Astonishingly the sophisticated network can learn metaphors, idioms and the subtle meaning behind language. This will effectively transform language translation today – rather than direct literal translation, the neural network can translate the same meaning to a different culture avoiding any possible offense.
While Facebook and Google have reported that they will switch over to neural network translation this year, they have not publically announced specific dates.
Auto-translation, not perfect yet
Despite the two tech giants rolling out plans to use neural networks, there are still major hurdles to cross before we’re quite there yet. According to Professor Philipp Koehn: “there are very hard problems with semantics and knowledge representation that have to be solved first, and that we are not close to solving.” Professor Koehn hints at less explicit information in the source language such a gendered nouns and verbs in languages such as Portuguese, Italian and German. Prof. Koehn is a computer scientist and expert in translation technology at the University of Edinburgh.
‘Chinese doesn’t use plurals, verb tenses or pronouns as we do in English, which makes exact translation very difficult’, Prof. Koehn added.
The Hazards of Machine Translation Tech and the Future
Albeit, translation technology has come a long way and provides decent literal translations; there is the need for a tech that speaks the real language of the end user.
Machine translation technology is a handy tool, but don’t rely on it entirely.
The STAR Team
Source: BBC Business News