People-powered translation at machine speed

Cloud computing has dramatically changed how we do business and the way we work. It has given rise to remote workforces and global collaboration. At the same time, businesses are leveraging large crowds of workers to perform tasks at unprecedented scale, speed and quality. Translation is no exception. So how is translation by the crowd, in the cloud, changing the landscape of content globalization? 

Cloud computing relies on sharing resources over a network, using converged infrastructure and shared services. The cloud also focuses on maximizing the effectiveness of all of these shared resources, updating and reallocating resources as needed. The cloud is a natural place for translation, in line with the Translation Automation User Society (TAUS) vision of “translation as a utility.” The cloud enables translators to do their work anytime, anywhere, from a variety of devices. Using the cloud means freedom. 

Crowdsourcing, on the other hand, is the practice of obtaining services, ideas or content from a large group of people, especially from an online community. The “crowd” sources what you might otherwise get from a traditional employee or supplier, typically in small increments, where each contributor voluntarily contributes to the greater result.  

In a world where there are vastly increasing amounts of content, splitting translation work among a large pool of contributors can enable us to keep up and to share it with the world quickly and affordably, something we wouldn’t have been able to do via traditional methods. The crowd is collaboration on a vast scale. 

Combining the freedom of the cloud with the collaboration of the crowd has many areas of application, from pooling scientific observations to locating missing people, as well as translating massive volumes of content. For example, Wikipedia is a fantastic example of translation by the crowd in the cloud. Wikipedia relies on volunteer crowdsourcing, known within the industry as “community translation,” enlisting thousands of volunteers to translate their articles into hundreds of languages. Translators without Borders also uses an unpaid volunteer model. These are nonprofits, and their use of community translation is a pillar of their strategic mission.  

In the past, for-profit companies faced with translation needs had two options: machine translation (MT) or traditional translation agencies. MT is a great solution for long, predictable texts written in a uniform voice. Long technical manuals go with MT like bread with butter. But what if the content requires a lively, human voice and possibly even a sense of humor? If the content is brand-critical, like marketing promotional pieces or product user interfaces, traditional translation agencies are a good choice — they have highly specialized translators, painstaking quality processes and can even do transcreation. However, these high-end services can be costly and time consuming. Today, there is a third option: paid crowdsourcing. 

For-profit companies have also experimented with inviting passionate and knowledgeable users to become volunteer translators. Facebook is one of the most famous examples and several other well-known companies have followed suit. The potential time and cost savings of this approach are very appealing. Oftentimes, however, it can be difficult to get volunteers to go the whole distance. It is not unusual to hear that a community manager was able to get the translation 70-80% completed, but got stuck there. It is also a common experience that community translators prefer exciting and visible content; it may be less interesting for them to work on support content or routine product descriptions.  

Paid crowdsourcing fills the gap between MT and traditional agency translation. It is ideal for content that needs a human voice, but has a dynamic nature and high volume that makes traditional translation too expensive and slow. It combines the speed and scalability of community translation with an economic incentive that ensures work is completed and quality standards are maintained. Paid translation crowdsourcing is all about getting the right quality for a project at the right price.  

The concept of “right” quality has been discussed with increasing frequency at globalization events over the past few years. Companies have realized that content with a short shelf life doesn’t have the same quality requirements as highly-trafficked, curated content. Social content often gets read one day and superseded by a ton of fresh material the next. As long as the human tone and persuasiveness is there, the quality doesn’t have to be flawless. For much of user-generated content, the source is not even perfect, so the market expectation is that translation does not have to be either. 

We know that consumers buy more when product information is in their own language. Online retailer Rakuten, for example, saw a 16% increase in conversion rates by using human-translated product descriptions. In “The six essentials of ratings and reviews,” online reviews provider Reevoo has also identified a direct correlation between the number of user reviews on a site and increase of conversion rates. Being able to translate this more casual, but very persuasive, content is becoming business-critical. It has to be affordable and it has to be fast. The right quality is an enabler of both.  

The right price, of course, is also a factor. There is so much more content in the world today than there used to be. Think of all the online video, e-commerce catalogs, community forums, product reviews, user-generated reviews and other social content. Online content is growing exponentially and user-generated content has already passed the 70% mark, or 900 exabytes. These volumes of content quite simply cannot be translated at traditional rates. Measured in cents per word, these traditional rates vary from the high teens to mid twenties, with some exceptions up in the thirties and forties for high-demand, low-supply languages. MT can be cheaper, or at least its starting point tends to be cheaper, and paid, cloud-based crowdsourcing also tends to have a lower starting point.

Managing the crowd

There is an art to leveraging a vast population to produce meaningful work at machine-like speed and reduced costs.  The human element brings an added level of complexity in terms of management and quality control. In a recent panel on “The Business of Translation” at Localization World Silicon Valley on October 11, 2013, speakers discussed the importance of “reducing friction” in translation processes. This is especially true in a crowdsourcing world, where scalability, speed and usability are essential. To smooth out the areas of potential friction, crowdsourcing providers must make it as easy as possible to do business.

Systems for ordering and translating, for example, must be streamlined and intuitive. In the traditional translation environment, translators use complex computer-aided translation tools, which can be both expensive and hard to learn. By contrast, crowd speed depends on users immediately understanding what to do, and for those steps to take a minimum of time. On the customer side, scalable crowdsourcing is about having an easy way to order large volumes, most often via a straightforward, well-documented application programming interface (API). API ordering enables data to flow swiftly and seamlessly between customer and crowdsourcing provider at the click of a button. Reducing the number of clicks necessary for translators and customers to do their work is extremely important. Supporting multiple currencies and regular translator payments via international payment methods such as PayPal also plays a big role in reducing friction for both customers and translators.

Simple, transparent pricing and efficient payment systems are also crucial. Traditional translation agencies and MT providers commonly have individual word rates for each language. Individual companies can spend a lot of time and energy negotiating regularly on these many different rates. Crowdsourcing can do away with this overhead by having a single rate across languages per target quality level. For example, a standard quality level for everyday content might be as low as $0.05 per word for all languages.

Finally, any service must maintain a certain level of quality. Traditional quality assurance activities such as manual translator screening and job review simply don’t scale when you’re talking about crowdsourcing thousands of translators. Crowdsourcing models should inherently solve this issue by designing an automated quality checking process. This helps to establish a reliable level of quality and quickly identify any issues at scale. If all this automation is done correctly, crowdsourced translation providers can be cost-effective while still preserving translation quality standards by maintaining objective systems for monitoring measurement.  

In order to fuel a highly scalable crowdsourcing machine, it is essential to have a large crowd of thousands of translators. The crowd also needs to be flexible — able to grow and shrink to accommodate daily order volumes that vary greatly in size. To achieve this, crowd managers need easy-to-use tools for analyzing quality and performance statistics, anticipating future volume and understanding the elasticity of different language pairs. For example, levels of translator activity change with increases or decreases in word volume.  

A highly automated translator acquisition system is also key. This usually involves various online feeder channels as well as automated online testing. Systems like these enable new language pairs to be established in weeks or existing language pairs to be scaled up in days depending on customer needs.  

Another success factor is having a happy translation crowd. Crowd translators are often students, stay-at-home parents or people with other day jobs who are doing translation in their spare time to supplement their income, or perhaps to have fun. It’s very important for them to be able to work when they want, where they want, from a variety of devices. They are also looking for fair and supportive work environments. This includes objective quality standards, transparent payment terms, on-time payment, and tools and training to help them improve their translation careers. While crowdsourcing word rates are low compared to traditional translation, having an easy-to-use, very fast user interface enables translators to nonetheless be productive on an hourly basis. Non-monetary incentives are also important motivators.  

We’re lucky in our translation world that we now have a spectrum of tools that can be used to help the enterprise cover all its translation needs. Traditional agency translation is still going to be used for high-end work. MT is still going to be used for long technical documents. And crowdsourcing is here to stay, right in the middle of that spectrum, where high volume meets human voice. Volunteer community translation will still have its place, wherever civic-mindedness or passionate interest comes into play. But for organizations that want the benefits of the crowd and cloud, with more control, paid translation crowdsourcing will help ensure reliability, timeliness and a predictable level of quality, along with great affordability. Paid translation crowdsourcing is changing the localization landscape by bringing in more translators and making it possible to translate more content, thereby increasing access to information on a global scale.