It was Renato (Beninatto) who reminded me, in the ‘Future’ panel discussion in Dublin, that only eleven years ago (when the TAUS think tank was founded) nobody - in his right mind - would think about using machine translation (MT) technology on any job anywhere. And now? Now MT is everywhere. Insiders say that everyday computers translate 200 Billion words. That is 100 times more than the output of all human translators together. MT is everywhere and always there, except … well, except the professionals seem to have their doubts. That makes me think that the state of the industry could be better.
TAUS may be getting old, but we cannot start resting on our laurels. Just coming back from the Industry Leaders Forum in Dublin, we realize how much work is yet to be done. Too often we are stuck in the past, in the way we have been doing things for a long time. How we price the services. How we plan the process and manage the projects. How secretive we often are about resources and processes. How we hang on to our data. How we ‘know’ quality when we see it. I invite you to read the reports on the sessions and discussions at the TAUS Industry Leaders Forum and the QE Summit in the first week of June in Dublin in the eBook we just published. Your conclusion might be the same as mine: the ideas are there, the technology is there… why are we not moving faster?
If our friends, family and hundreds of millions other ‘ordinary’ citizens around the world happily use the MT button on the internet and share their data, why are we, professionals in the industry, then, afraid to do so? There are good reasons of course. For instance, we have to worry about data security, the quality of MT, how we can best share data and learn from it, and, not least, how we carve out a business and still make a living. But rather than ignoring or denying the future, let’s focus and resolve those matters.
To Cloud or not to Cloud
That is the question. But in fairness, it is somewhat of a rhetoric question, because as much as ‘not to be’ meant suicide for Prince Hamlet, the option of ‘not to cloud’ does not exist for a business that wants to stay alive. We can bemoan all the dangers of the cloud, but it is like lamenting the risks of living. More so perhaps than most other businesses, the language business can benefit enormously from the cloud. The cloud as the delivery channel, the cloud as the host of all our data, the cloud as the vehicle for the Human Language Project. This discussion should be about streaming localization, effective measures for data security and objective comparisons between on-premise installations and cloud-hosting.
Innovation made elsewhere
Translation has entered the public domain. Buttons pop up on every screen and in every app to give us access to content in our own language. The translation industry is bursting with new technologies, platforms and solutions. But as it happens, most of the innovation is made elsewhere, not in our own organizations. In fact, the most disruptive innovation, usually comes from outside an industry. Think about how Google Translate shook up the translation sector. We can’t afford to ignore or deny it. We have to watch the innovation (made elsewhere): in China, in Japan, in digital marketing, and make use of it where we can.
What data would you like to track: machines that learn
Those who still rely (solely) on human judgment and intuition when it comes to finding the best translator for the job, deciding what to translate and what not, which technology to use for which content, will find themselves outpaced by data-driven platforms. The datafication of translation began a few years ago and is now in full swing. In the ‘Machine Learning’ session in Dublin, led by Adam LaMontagne from Moravia, participants enthusiastically listed all the data that we could track to make our daily work more efficient. The desired data range from the number or words and edits per hour to the temperature and weather in the translator’s location. On our to-do list is exploring how we, as an industry, can collaborate identifying data, decide on metrics and extract intelligence.
What’s next in MT: the N-factor
If you thought that MT had reached a plateau where almost no further improvements could be expected, you were wrong. A new breakthrough has announced itself and it is called Neural (or Deep Learning) MT. Where the current generation of Statistical MT engines usually stop learning from phrases or N-grams of no more than three words, the new Neural MT engines take in much longer phrases and learn from context. The result: dramatic improvements in the quality of MT output. You can’t ignore this. Fully automatic translation (without the effort of post-editing) may become useful for more content and in more scenarios that could not be supported until now.
It’s not about ‘You’: the international customer experience
One of the most popular sessions in Dublin was the one led by James Douglas from Microsoft on the ‘international customer experience’. It seems so obvious: it is not about ‘you’ or ‘us’, it is about the end-user and his or her experience. But we have all become so absorbed by our own methods and processes of localization that we have lost touch with that reality. Now, more than ever, just translating the content is far from enough to win customers and build global brands. Whatever industry you are in - software, healthcare, automobiles - razor-sharp marketing to the end-user through direct channels and direct engagement, with personalized content is the way to go. How we converge old-style localization with the latest digital marketing techniques is the new frontier explored by the most innovative companies.
Speech-to-Speech
Robots that speak your language are no longer science fiction, as we all know. But teaching them to take on an attitude as our human companions often do, caused some hilarious moments at the TAUS Industry Leaders Forum in Dublin. Speech-to-speech translation has been on the agenda of TAUS events for years now and its rapid progress cannot be denied. Most outspoken of course with the Skype Translator. What’s needed to push on is data, again: speech data in many more domains and languages. The attendees in the TAUS Industry Leaders Forum in Dublin called on TAUS to expand the Data Cloud and open it up for speech corpora as well.
How we measure
So much has been said about measuring translation, it has become a no-brainer. At the seventh TAUS QE Summit (hosted by Microsoft this time) and at the Industry Leaders Forum participants called for action to become transparent about translation quality and productivity and to adopt the common DQF metrics. Several translation technology companies (SDL, XTM, MateCat) presented their integrations with DQF. A special DQF Enterprise User Group was formed in Dublin that will work together to meet the requirement for data privacy, customized reporting and industry benchmarking for bigger global users of DQF. Good progress but a lot more work to do. Read the reports and sign up for the QE Summit hosted by Intel in Portland on October 26.
Pricing in a freemium world
And yes, we are still discussing pricing models. How much longer should we count words? And how can we put the right price on the word, if the technology component in the solution is becoming so much greater. Is it not about time to shift to a more mature and appropriate model, like ‘managed services’, a subscription or SaaS model? The pricing question is equally challenging for the translation buyers who want to have the right incentive as for the language service providers who need to build a sustainable business. Enough talk about it, you hear some people say in the corridors, somebody has just got to do this - come up with a new and good pricing model - and then the others will follow.
Training talents
The future does not need translators, at least not in the old way. That was the conclusion of the ‘Future’ panel in Dublin. The future may not even need post-editors, as we see the MT engines getting better on the one hand and the audience becoming more forgiving for lesser quality of fast-moving content on the other hand. So, what do we need? By the time the lights in the old translation shop are switched off, what are the talents that we are looking for? We still need literary translators and ‘transcreators’ (don’t expect MT to take on poetry, literature, marketing slogans and copywriting successfully). The human skills that a highly automated localization environment depend on are related to quality evaluation, content profiling, cultural advisory, data analysis, computational linguistics, and yes on post-editing for the time being at least, but indeed less and less so on translating plain text. How universities train talents for the future of our industry is a topic that should appear high on our to-do list.
We hope you enjoy reading the articles written by the TAUS reporters on the various sessions in Dublin. And we hope that you join us in the hard work that lies ahead of us. As I said before, as an industry we are just scratching the surface. If we want to make a real difference and want to help the world communicate better, our focus should be on technology and data. If you find interesting topics and ideas in our to-do list for the industry sketched here above, then check out our agendas for the TAUS Annual Conference and the QE Summit in Portland from October 24-26 and let us know if you like to contribute to one of the discussions.
8 minute read