In five self-help steps
Author: Jaap van der Meer, Director and Founder of the Translation Automation user Society (TAUS)
Communication across the world’s many spoken languages is a problem. The technology that can help solve this problem is getting better and better. But the professionals, needed to use and improve the technology – at least the vast majority of them – reject the technology and deny its advancements because they fear for their jobs. A good Dutch expression jumps to mind: “Man suffers most from the suffering he fears, but never appears.”
In this prefatory note I argue that to develop the MT industry further we need to push users to overcome their fears – MT technology is already quite good in its current state – and become supporters and enthusiastic users. I am proposing “five self-help steps” with links to the TAUS web site for more support to users who are ready to adopt MT technology in their daily practice.
The translation industry is caught in a dilemma: to automate or not to automate. Dammed if you do, dammed if you don’t. Using machine translation technology feels like a curse to everyone who spent years to study other languages and become a professional translator. But if you don’t, it’s becoming harder to stay in business. Customers want translations faster and cheaper. Besides, the volume of information requiring translation is beyond imagination. The worldwide population of a few hundred thousand professional translators is just scratching the surface. Machine translation technology can help increase efficiency and open new business opportunities.
TAUS recommends five steps for every buyer or provider of translation who is caught in this dilemma.
The first step is to ‘get real’ about the technology, ask the right questions. Since MT has been made available by Google and Microsoft and millions of people started using it every day, have we seen a decline in the mainstream translation market? Have translation jobs been taken over by computers? Market analysts report solid growth rates of the translation industry from 2005 to 2011. The need for translation is so great; it is becoming impossible to meet demand without the use of automated translation technology. Did other industries decline or suffer when automation technologies were introduced? Think of the banking industry or the travel industry. Admittedly jobs changed, but overall opportunities have been on the rise and industries started to flourish when new technologies were introduced. The first step is therefore to open up, get over the fears for change. TAUS has published a series of articles and white papers about changes in the translation industry and language business innovation. We welcome you to surf through this library of publications (http://www.translationautomation.com/articles/)
The second step is to try it out. After all, the proof of the pudding is in the eating. However, don’t start trying MT before you have successfully concluded step 1. Too many people have already trialed MT with the sole intention to prove that it isn’t working, that it’s ugly and often laughable. You want to try it looking at the potential and the positive. TAUS recommends a simple do-it-yourself test before engaging vendors and consultants. There are many options available to try machine translation technology. You can check out the TAUS Tracker (http://www.taustracker.com) online directory of MT engines to see what is available in your preferred language pair. You can translate some of your own text using online free MT tools, but you will in most cases not be able to feed these engines with your own terms and phrases to help them improve the results. It is worth buying a desk-top license of a commercial MT package. For a small investment (usually under 1,000 Euro) you get access to a rich set of customization features, often not much different from the features offered with the server-based systems. If your team is equipped with some software engineering skills and you are prepared to set aside some time, the other option to gain MT experience is to download open-source MT software, such as the Moses engine. Moses is gaining popularity very quickly now among service providers and buyers of translation. TAUS has published a (free) online Moses tutorial (http://www.tauslabs.com/open-source-mt/mosescore/moses-tutorial) , dozens of use cases and best practice reports (http://www.translationautomation.com/reports/) for training open-source MT systems.
Now you are on your way to become a user of MT, you know you cannot avoid going through trials and errors. The third step we recommend, to not make too many of the mistakes others made, is to learn from others. TAUS has launched an online knowledge base for MT users: the FAQ Forum (http://www.tauslabs.com/open-source-mt/faqs) Questions from users are posted, intelligence is collected from all other users and TAUS is undertaking further research. Responses are documented and reviewed. Another source of intelligence is the TAUS YouTube channel (http://www.youtube.com/user/TAUSvideos) with presentations of use cases. Finally if you can’t find what you are looking for you can always send an email to firstname.lastname@example.org. Only after you have completed your discovery of others’ trials and errors, we recommend that you start documenting your goals and addressing the fundamental questions about budget, vendors and business models.
If, after you have learned your lessons, you decide to proceed and implement MT technology in your organization, the most crucial step (of all five) is to define your goals. Which content types do you plan to apply the technology to and what do you want to achieve in terms of quality and productivity. Many large and small enterprises struggle to formulate these goals, not just for the use of MT, but even for human-based translation. The lack of criteria and clear evaluation metrics leads to uncertainty, friction, disputes, and loss of time and money. It is often opinions, rather than measurements, that lead to the rejection of MT system and the firing of vendors or translators. In consultation with many of the large enterprise members TAUS has developed the Dynamic Quality Framework (DQF) (http://www.tauslabs.com/dynamic-quality/about-dqf). DQF allows users to profile their content types, based on a measurement of three factors: utility, timeliness and sentiment (UTS). The resulting UTS score is then linked to a recommended approach to evaluating the quality of the translation of this particular content. DQF is a knowledge base documenting seven different quality evaluation approaches. In addition DQF provides industry-shared tools that allow users to compare and benchmark MT productivity and translation quality scores, such as adequacy and fluency. The TAUS Dynamic Quality Framework brings credibility and transparency to translation business. Without the methods and the knowledge to set goals, to measure and to benchmark against these goals, users of machine translation technology are like hunted game.
The fifth and final step that TAUS recommends is strategic and is aimed at sustaining growth and innovation in the translation industry. Just like you have learned from others at step 3, we recommend that you share the lessons you have learned, so that others can learn from you. Do not fear to give up an advantage, or you will soon find that it’s you who is left behind. Only if we decide to be truly open and ensure complete interoperability in the systems we build and use, we will reach the maximum growth opportunities of the translation industry. TAUS has launched a translation web services API (http://tauslabs.com/interoperability/taus-translation-api) to facilitate seamless exchange of translation jobs on the web. But we take sharing even a step further. TAUS advocates that all organizations share their language data in order to improve the performance of their own translation technologies and to stimulate innovation and growth in the translation industry overall. More and more companies and government organizations are following this vision. In 2008 TAUS launched a data sharing platform. Today the TAUS Data repository (http://www.tausdata.org/) contains already 50 Billion words of shared translation memories in more than 2000 language pairs. These language data are available for every user of MT to train and improve their engines.