LSP Perspective: Real-Time Translation Tools, The State of the Industry

This is another in a series of LSP perspectives, this time from Tomedes, on MT and real-time translation technology. I do not always share the opinions of the guest writers, but I like to see differing views in this forum as they often help to understand the perceptual reality that we live in. For example, if people in power believe that Sadam Hussein has WMD, then it is easier to justify going to war even though actual data suggests the existence of WMD is not a reality. I generally do not edit these opinions except sometimes for minor typos and basic reading flow related edits. I may sometimes highlight some statements in bold to highlight something I think is central to the view being presented. I invite other LSPs who might want to use this forum to share a personal opinion on translation technology. I do not expect opinions to be the same as mine, and only ask that the post meets some basic quality requirements.

This is also an opportunity to share some feedback from the AMTA (American Machine Translation Association) conference which is sort of related to the title of this post. Daniel Marcu (formerly of SDL/Language Weaver) released a survey he did for an MT commercialization panel. Keep in mind the problems associated with most sampling based deductions as I pointed out in the last post. This is especially true for conferences where the same people seem to be talking amongst themselves.  Some of his findings perhaps ring true but are somewhat obvious anyway, and I was not quite sure what the point of several of the questions was. Hopefully, the panelists, especially Chris Wendt of Microsoft were able to steer the conversation to a more constructive review of what is happening with customized MT and professional use of MT. More than 60% of an 185 person sample in this survey were MT researchers or translators/PMs and the details are available at the link. My advice to anybody reading conclusions from any survey – examine and understand the sample first, as conclusions are really only reflections of the sample. I thank my statistics teachers for drumming this into me, early in life.

The most useful question in the survey for me was the first one: What are the top barriers to broader MT adoption in the Language Services industry? As you might guess, better MT quality and more consistent MT output were the primary barriers chosen from the available list of answers you can see below with your sharp eyes. I would add that predictability is a much more important driver i.e. can you foresee/predict what is likely to happen with your MT engine BEFORE you build it. Professional users do not want to gamble, thus, having the foreknowledge and the skill to understand what might happen a priori with an MT deployment, is critical in a professional setting. Also, overall competence with something like MT is something that is built over time with experience, unfortunately, this is something many agencies expect the technology to magically compensate for. As easy as 1-2-3 is a myth, and an instant dump-you-data-in and boom here-is-your-system is very unlikely to produce an actual business advantage.  If you want instant results with MT, stay with the big generic engines. Remember that drunk drivers will have accidents with even the safest and most reliable cars, and that there are no cars yet that are drunk driver proof. 

 Some other survey highlights:

• In the aggregate, participants believe that technology savvy professionals will eventually replace those who are not (tech savvy).
• Most participants (54%) believe that increased MT quality will lead to the industry being consolidated with competent MT practitioners gaining a long-term business advantage.
• Language Technology and Services Executives feel strongly that MT startups should focus on developing end-to-end solutions that tolerate less than perfect MT (47% and 58% respectively).

-------------------------

It feels like every week or two another big company is claiming to have conquered real-time translation, or at least produced a tool that is ‘almost there’ in terms of machine translation rivaling professional human translation. If that’s really the case, then translators around the world should be dashing to join their local unemployment queues. The fact that many translation professionals are still gainfully employed seems to argue against the ‘revolutionary’ real-time translation advances made of late, so what is the current state of the industry? And how long until real-time translation actually does what the name implies without producing blundering errors?

Giant tech corporations are racing to be the first to reveal flawless, real-time translation capabilities. Google has put a lot of time and energy into the Google Translate app, which impressive (if not yet flawless) results. The app allows you to type in words you want translated, scan an image using your phone’s camera (through the app’s Word Lens feature) or speak into the phone. The results are displayed in both the original and the target language and you can listen to the being read out in both languages as well as seeing them on the screen.

Having put Google Translate to the test with a variety of simple sentences spoken aloud, which it translated accurately from English to French and English to Portuguese, I tested it out with some more complex translations. I read from business letters and other work-related documents, curious to know how Google Translate would hold up in a professional capacity. I was pleasantly surprised to find that, other than the occasional word which the phone ‘misheard,’ the translations were largely accurate. The results from scanning business letters were equally impressive, although the process took a little longer to complete than simply reading the letter into the phone.

Google Translate’s Word Lens feature and real-time voice translation tool were first revealed back in early 2015. At the time Barak Turovsky, Product Lead of Google Translate, claimed:

 “Today's updates take us one step closer to turning your phone into a universal translator and to a world where language is no longer a barrier to discovering information or connecting with each other.”

 Google wasn’t claiming to have mastered real-time translation in its entirety, but the app certainly stands up well in terms of reducing language as a barrier.

Microsoft’s Skype Translator was announced just before Google’s updated Translate tool, with a pre-release version covering spoken Spanish and English available to Windows 8.1 users from December 2014. Since then, Skype Translator has been expanded to include eight languages for voice calls and over 50 languages for instant messaging. The roll out to Windows users was complete by January 2016, but sadly the Mac version hasn’t yet arrived, meaning that I haven’t been able to test it out personally.

The preview version left some users a little underwhelmed, leading Digital Trends to summarise it as, “promising, but definitely still in beta.” When the Mac version arrives, we’ll see whether Skype Translator has managed to live up to that initial promise.

Microsoft has also produced Microsoft Translator, which it announced in early November 2016 would be able to translate multi-lingual group conversations in real time by the end of the year. Quite a claim, particularly given that it didn’t fare too well when put through its paces during the research for this article. After the slick interface of Google Translate, the Microsoft Translator app felt distinctly lacking. While simple spoken sentences were translated well from English to French, the app struggled with longer sentences and the more formal language used in business letters. Several times, the recording stopped halfway through a sentence being read out and some of the translations were less than ideal.

Microsoft demoed the Microsoft Translator multi-lingual group translation feature at Microsoft Future Decoded in London and it was fair to say that the speech recognition element was far from perfect. Olivier Fortana, director of product strategy for Microsoft Translator, has commented:

“The idea is to say 'Everybody has a smart device, a smartphone or a tablet. What if we could harness the power of those smart devices to enable real-time, multilingual conversation translation for an in-person situation?'”

It’s a great idea, but testing has shown that further work is needed before such a tool can deliver on its promise.

Nor is it just the biggest players who are aiming to conquer the real-time translation market. A host of companies have sprung up offering software and apps that translate beautifully. iTranslate Voice is one of the best of the bunch, with excellent speech recognition and very good translation. In fact, based on testing specifically for this article, iTranslate Voice did better than any of the other tools tested when it came to speech recognition.

Aside from device-based apps, The Pilot System from Waverly Labs looks to be one of the most exciting wearable translation devices on the market. If it actually exists. So far, the internet has been buzzing with comparisons to the legendary Babel Fish from Douglas Adams’ The Hitchhiker’s Guide to the Galaxy and the potential of The Pilot, but the in-ear translation device has yet to materialize. The latest estimate from Waverly Labs is that the cutting edge technology will be available in May 2017, with the company’s Indiegogo page showing 4,329 orders of the product (which is currently listed for US$199 plus shipping).

Asked outright by Forbes if The Pilot was really a scam, Andrew Ochea, CEO and Founder of Waverly Labs, responded:

“We are a tech startup in New York City, we are a team of six…there is no elaborate marketing scheme behind it, we are not out to scam anyone. We spent about 18 months proving that we could build it before we even put any kind of information or promotional materials to generate interest.”

The Pilot isn’t yet available for public testing and after a lengthy phone call with Waverly Labs, Forbes remained unconvinced that the device would live up to the hype. CEO Andrew Ochea stated clearly:

“This thing isn’t the Babel Fish, it is only going to work in certain conditions.”

In summary, it’s clear that real-time translation has come on leaps and bounds, particularly in the last two years. However, while there are some good tools out there, none of them are flawless and we’re likely looking at a couple more years at least before any one company can sincerely claim to have achieved the real-time translation goal of which so many organizations are in hot pursuit. It seems that traditional translation companies don’t need to start panicking quite yet!

======================

This post was authored by Louise Taylor.

 

Louise Taylor is a freelance writer who has had a passion for languages since an early age. She holds qualifications in Latin, French, German and Spanish, as well as her native English. She is also well on her way to speaking Portuguese fluently. As a long-term freelancer, she understands both the positive parts and the pressures of working in the freelance sector.