No doubt, the future is here: with the release of Microsoft Translator for iOS, Android, Apple Watch and Android Wear smartwatches, you can now talk into the little magical device on your wrist and have it translate whatever you said into another language.
Well, sort of.
The app, a challenger to the Google Translate app that’s been offering speech-to-speech and on-the-fly text translation for a while now, is reasonably accurate. But by now, everyone knows not to expect too much from machine translation — for every surprisingly accurate result, there are still too many garbled misunderstandings or out-of-context translations.
But how does the new kid on the block stack up against Google’s app, which has just recently added 20 new languages to it’s also very sci-fi instant visual translation feature?
Android Police answers this question with an in-depth comparison between the two apps. In short, Microsoft Translator offers a prettier interface and the smartwatch versions, but lacks many other features that the more mature Google app has been offering for long time now: offline use, live conversation mode, more comfortable copying of results, and the visual/camera translation.
As far as accuracy is concerned, the jury is still out. We’d love to hear from you about your experiences with both apps — how do they stack up in your language pair? Let us know in the comments.
We’ve seen American politicians playing fast and loose with translations before — but that last incident, involving a specially targeted Spanish-language version of President Obama’s State of the Union Address — was strictly domestic. Now, Foreign Policy reports on an international case of high-profile, low-quality translation of a particularly explosive document.
As you might have, heard, Arkansas Senator Tom Cotton sent a letter signed by 47 Republican Senators to Iranian leaders explaining the US Constitution and warning them about signing a nuclear arms deal with President Obama. The Senator also tweeted a Farsi version of the letter.
The letter was widely criticized as a partisan attempt to undermine the President’s foreign policy — the headline in the New York Daily News read: “TRAITORS“.
If that wasn’t bad enough, it turns out that the Farsi version of the letter “reads like a Middle Schooler wrote it”, according to Foreign Policy. Here’s an excerpt from the translation, as it would read to a Farsi speaker:
The meaning of these two articles of Constitution is that we any agreement related to your armed nuclear program is not passed by the Congress is that not anything more than an agreement between President Obama and Ayatollah Khamenei don’t consider. Whoever replaces the president could remove such a chairmanship agreement with the movement of a pen and future Congresses could modify the terms of the agreement at any time.
Foreign Policy suspects that Senator Cotton relied on Google Translate instead of hiring a high-quality human translator. It’s something we recommend even when the stakes are somewhat lower than international nuclear negotiations.
From Skype to Google, we have previously reported on the ongoing struggles of machine translation to deliver reliable real-time results. While this branch of software has been making astonishing strides, it still seems a long way from being a viable alternative to a high-quality human translation.
The latest example comes from the BBC, which took a few of the latest mobile translation apps on a field test to Bilbao, Spain. Armed with a to-do list of goals that included visiting the Guggenheim museum, taking a selfie with someone and asking a stranger to tell him the story of their first kiss, Kevin Rawlinson tried to navigate the Basque city with the help of Google Translator and a few other apps.
Again and again, Rawlinson came up short. Even under optimal conditions, the apps kept misunderstanding or mangled their output to an incomprehensible mishmash. (Our headline is from Google Translate’s attempt to translate “Can you speak slower, please,” which it rendered as “Can you speak Spanish, Big Show.”)
From field tests like these, it’s clear that real-time translation apps still have a long way to go — but it’s also clear that companies will keep trying, given that the language interpretation industry generates about $37 billion of annual sales. In a quickly globalizing world, demand for reliable translation, localization, and interpretation services is as high as ever, and it’s appealing to think that computing power could fill the demand. It appears that for the time being, though, human translators are still the only way to get the job done right.
We’ve been following Skype Translator since the summer, when the company revealed a demo video of the real-time translation software. It seemed like science-fiction then, but now a beta version has been rolled out and the first reviews are coming in.
Back in July, we predicted that no matter how powerful Skype’s babel fish software was, it would probably still take some time before human translators and interpreters should start looking for new jobs. A first test on Gizmodo seems to bear this out: “feels like a language assistant,” not like a true translator.
According to Darren Orf, who tested the English-to-Spanish live translation, the software takes some getting used to. You have to speak slowly and make pauses at the right places. Once you get used to this, the recognition is “really quite amazing,” but apparently the machine translation needs some help: the results are “broken” and take some time to decipher. Orf gives a particularly dangerous example of a misunderstanding.
From the sound of it, it’ll be a while before Skype Translator will be ready for professional use — but we’re curious to see this technology progress, and if you’re itching to give it a try, too, you can sign up for the beta now.
It’s an old dream that pops up in science fiction time and again: on-the-fly language translation that lets speakers of different languages communicate as if they understood each other directly. In Star Trek, Kirk and Spock have their tricorders, and in Douglas Adams’ Hitchhiker’s Guide to the Galaxy, a “babel fish” burrows into your ear to send a translation directly into your brain.
Skype Goes Multilingual
Now Skype, the super-popular video call service recently acquired by Microsoft, unveiled that they are working on a similar technology — no fish or Vulcans required.
Back in May at the Code Conference in Rancho Palos Verdes, California, Microsoft unveiled a prototype for a new Skype feature called Skype Translator. Microsoft CEO Satya Nadella demonstrated the “near real-time” translation service by calling a German friend and conducting a conversation with him — in German and English.
The feature listens for pauses in the speech, recognizes and translates it on the fly, and serves a text and text-to-speech output to the other party. “It is early days for this technology but the Star Trek vision for a Universal Translator isn’t a galaxy away and its potential is every bit as exciting,” said Gurdeep Pall, corporate vice-president of Skype.
English-German to Launch First
The company, which connects 300 million users every month for over 2 billion minutes of calls, says it’s been working on real-time machine translation for over a decade. The first language pair to be good enough to work is English-German, and Skype Translator is planning to launch the service this fall with a beta app for Windows 8 — but not necessarily for free.
“It’s been a dream of humanity ever since we started to speak and we wanted to cross the language boundary,” said Nadella.
Time for Translators to Look for Other Work?
Does this mean it may be time for professional translators and UN interpreters to hang up their jobs and look for other work? Not so fast: like all machine translation, Skype Translator may one day become a useful tool for casual conversations — but the risks for using MT for business and professional use remains high.
As amazing as this science-fiction inspired technology may be, some argue that it will never be able to understand every nuance of human speech. Sharp-eyed observers will even spot minor but telling mistakes in Microsoft’s demo video (hint: take a look at 0:46), making it clear that professional translators won’t need to fear for their jobs any time soon.
It has been almost twenty years since IBM’s Deep Blue beat reigning world champion Garry Kasparov in a game of chess. Computers can outsmart humans–we all know that, right? Well, Germany’s Federal Association of Interpreters and Translators (BdÜ) has recently staged a different kind of battle between man and machine: a competition between translators.
Given the popularity of free online tools such as Google Translate, it’s understandable that traditional, human translators feel like they want to measure themselves against the artificial challengers head-on. That’s what the BdÜ did: long-time, state-certified financial translator Ralf Lemster and Google Translate went head-to-head, translating a newspaper article from English into German.
The results were then checked and compared by Uwe Reinke, a linguistics professor from Cologne. While Google Translate spits out a finished text in seconds, Lemster takes about 20 minutes — but the difference in quality speaks for itself. “Shorter sentences are semi-intelligible,” says Reinke about the machine translation. But on more complex sentences, Google failed completely — it’s hard to make out any meaning.
“GT isn’t capable of accurately capturing context, background, or even industry terminology in some instances,” explains Lemster. He also points out that customers appreciate it when their texts are reviewed by a human translator who can give feedback.
For Andre Lindemann, the President of the BdU, the test once again demonstrates the weaknesses of machine translation. To him, it’s useful for personal use, as a quick fix or initial point of orientation, but only into a language the user knows so he or she can catch mistakes.
“It’s completely unsuitable for business use,” says Lindemann.
He also points out that privacy is an issue with cloud translation services. “”Nobody knows where the data ends up,” he says, adding that confidential documents should never be translated with an automated online service such as Google Translate.
The test shows quite clearly that the Deep Blue of machine translation is still nowhere in sight. We can only agree with the video’s conclusion: if you want accurate and professional translation, you have to rely on human intelligence.