Adobe is not only changing the world through digital experience, the company is also changing the language industry through community engagement and continual innovation. Ranked by Fortune magazine as the most admired tech company in computer software, Adobe’s award-winning technologies have redefined business, entertainment and personal communications. Christine Duran is the senior manager of Translation Technology at Adobe, which has over 9,000 employees worldwide and is headquartered in San Jose, California.
Thicke: Adobe is by definition innovative. What are you doing that’s especially innovative?
Duran: Adobe is changing our entire way of doing localization, from who participates in the translation process down to the infrastructure and systems that support localization. We’re embracing the fact that it is a multilingual world and language can’t be a barrier to a great customer experience.
Thicke: What is your main role?
Duran: I have been working for the last couple of years on the mandate “translate anything, anytime, for anybody.” We wanted to come up with a platform to enable almost anyone, anywhere, anytime, to translate almost any type of content and publish it in multiple modes on multiple devices.
Thicke: That’s a big mandate. Tell me more.
Duran: Traditionally, localization has been fairly straightforward with delineated silos of content — software, user interface (UI), help files, website and so on — published on predictable schedules. We used to have systems that were highly optimized around each silo. The one uniting factor involved centrally managed translation memory and globalization management systems (GMS). Now all those distinctions are blowing up. You find marketing content inside of software, software embedded in web content and social media everywhere. Who gets to publish content has changed as well. Content comes from inside and outside of Adobe. User-generated content is more important than ever. Marketing content, representing your voice to the customer, is where most companies still want to maintain control, but even that is seeing more collaboration between companies and outside producers — think of amateur Super Bowl ads. All this collaboration is great for enhancing the customer experience, but it is a challenge to localize content from so many sources, much of it unstructured and unscheduled.
Thicke: What’s the biggest challenge?
Duran: Keeping up with the constant changes in who creates content, when and how. We can easily localize half a billion words per year of traditional content. The question is how we are going to react to this new world, where content is exploding, different contributors are involved, and content is being published in different media and places. We are adapting our localization processes and infrastructure to handle both the short-term changes, but also to be flexible enough to meet future demands.
Thicke: How does translation technology fit in this ecosystem?
Duran: From a translation technology point of view, we have to enable this process. Traditional computer-aided translation (CAT) tools and even GMS were innovative in their time, and many companies use these tools extensively, but they are focused on files, translators and ISPs. These tools are overly complicated and overly simplistic at the same time. The CAT tools are too complicated for casual translators. GMS are optimized for file-based translation workflows. The localization process is still mostly driven by documents and assets, but that’s not all we localize anymore. We have to be able to go from a simple tweet to a complex multimedia file. Also, GMS have no visibility into other parts of the localization life cycle — such as testing, engineering, publishing, finance and procurement. You can keep adding capability to the GMS by connecting it to other technologies such as machine translation (MT) engines and content management systems, and there are some technology providers who are trying to supply that whole technology stack. A large, diverse company such as Adobe can’t dictate one set of technology for all its businesses. Other providers will say “We have connectors to anything you need,” and that’s true. But you can go too far with that model as well. You can create a situation where every enabling technology and client have a direct connector to the GMS. If the GMS goes down or you want to change providers, then everything else is affected. It’s a lot of overhead to maintain all those connectors. So we decided, let’s stop trying to make the GMS the center of the universe and let’s make it one part of the content delivery ecosystem where we may not know what the next piece of needed technology may be. We want the GMS focused on what it’s good at: professional translation and leveraging.
Thicke: What will you replace it with?
Duran: We’re working on a globalization tools platform that features a localization service bus. This lets us abstract linguistic functions and other functions into services. For example, we can offer leveraging, terminology and MT as standalone services as well as part of a GMS workflow. We can abstract the file and data filtering as a standalone service and write filters once instead of replicating them in many different systems. We can create flexible workflows across services — meta-workflows — and include steps such as updating the finance system. So, I can send a request for a dynamic MT translation or a complex human translation and review workflow that will take a week. The service bus concept should also allow us to trigger localization workflows from events in other systems. Instead of the GMS being the center of the universe, it becomes one of many systems involved in localization. We can change the GMS or MT engines or finance systems without affecting the other pieces of the ecosystem. We can support clients and consuming services of all sorts — inside and outside the firewall, desktop, mobile or web-based, and broker them through the service bus.
Thicke: Who are your stakeholders?
Duran: Traditionally, my stakeholders are program managers, localization vendors, internationalization engineers and content writers, but I also have a new set of stakeholders that includes Adobe product teams, Geo reviewers, channel partners, third-party publishers, user communities and crowds. By “crowds” I mean people who are not professional translators, but have a passion for Adobe products and want to contribute to expanding localized content.
Thicke: What kind of content are you handling?
Duran: Our systems currently handle the product UI (desktop or SaaS-based) and documentation, including traditional help, knowledge bases, tutorials and elearning. We also localize all of Adobeâ€¨.com and basically anything published on the web. There’s more and more multimedia from places I never would have imagined.
Thicke: So anything and everything!
Duran: I’m trying to design systems that are accessible to a community user, a traditional localization vendor or a partner. I want to be able to support professional, heavy clients for complex translation tasks as well as lighter, simpler interfaces for community translation or feedback. I’d like to be able to instantly translate a web page and edit the results in place as well as send a 100-page document off to a more complex workflow.
Thicke: How do you handle community translation? How do you motivate the community at Adobe?
Duran: There are a number of different community efforts at Adobe. There has been an existing set of user groups and various levels of group participation for different products, with more or less Adobe involvement depending on the type of group. We’ve supported user forums for a long time; some of that is replicated internationally. Of course, there has always been the crowd. Some types of products lend themselves to communities, while others cater to a crowd with a common interest in the product but who don’t necessarily meet together or have strong personal associations.
Thicke: So for you, the crowd and the community both have strong interests in a product, but the crowd doesn’t work together as a community will?
Duran: More like the crowd is geographically disbursed and probably not meeting together regularly. For example, I’m interested in Photoshop Express, but I may not go to a user group meeting and hang out with other users. But as a Flex developer, I probably want to be part of a focused community. The crowd has a common interest, a casual interest. A complex product, on the other hand, might have a focused community that wants to work closely together. At Adobe we’re trying to reach people from all across that spectrum. Certain products have dedicated communities. People want to write articles, provide sample code and jump in to help.
Thicke: Do you direct your communities of users?
Duran: As I said, there are different types of user groups with more or less direction from Adobe. But in terms of translation, we take a hands-off approach. If they want to translate something, we want to help them get the information they need. If there is something they want translated within their communities that we are not going to translate, we can facilitate that. We encourage them to organize translation tasks in a way that works for them. They can choose to motivate their members with recognition or competitions — whatever they like.
Thicke: Why would the community go to the trouble of translating Adobe content?
Duran: To get content that they feel is valuable. In some communities, the articles they pick for translation aren’t the articles that we would have chosen. They could be digests of blogs that we might not think to translate. It could be content coming from Adobe, but it could also be content about Adobe products coming from other sources that they want to publish on their own sites. As long as the content is licensed under Creative Commons, we’re happy to facilitate community translation.
We’ve also just rolled out community translation on Adobe TV. This is content that Adobe has created. It could be training videos, evangelist talks or elearning modules. Adobe has partnered with a third party, dotSub, to enable the community to translate the subtitles. They might be resellers who want to see the training in local languages such as Indonesian or maybe enthusiasts. They apply to be a translator, and then we qualify them. dotSub provides the underlying technology. Adobe TV is more of a crowd translation paradigm because individuals choose what’s interesting to them. And the subtitling is fun; it’s something they can do in an hour or a couple of hours. It works because it’s a finite task.
In my opinion, the idea that the crowd is going to translate long documents is unrealistic. They don’t have that kind of staying power. And they have day jobs. So, they tend to translate excerpts of texts. LSPs are not going to lose this work to communities or crowdsourcing because it’s not the kind of work communities want to do.
Thicke: Do you have a hierarchy for what content the crowd can translate?
Duran: We just started the community translation efforts, so there are no hard and fast rules in place at this point. We evaluate each case and work closely with product teams to make decisions about what can be crowdsourced and what can’t.
Thicke: What about MT?
Duran: What I see is that companies have more interest in MT as they have more content to get out there. Across the board, for anything that needs to be translated quickly or where the quality expectation is “good enough,” MT is being used. Many companies machine translate transient content such as service notes and knowledgebase articles because it is more important to have a useful, timely translation than a perfect translation or no translation at all. However, when Adobe first deployed MT, we went straight into product documentation. We first looked at our Action Script developer documents — millions of words, class names and descriptions, and a rapid publishing cycle. In a traditional process it would have taken two months and seven translators for a new language; with MT, it took a few days for the first cut. Even with post-editing, it was done in half the amount of time.
Thicke: What about quality?
Duran: Adobe draws from an extensive pool of professional translators to ensure all critical materials are precise and of the highest quality. We’ve found, however, that there are many other materials — such as blogs, developer articles or subtitles for a video — which the community finds of value and chooses to translate voluntarily and for their benefit. This is usually material Adobe would not translate, and we try to empower the community by providing tools. Most community translations are moderated, and there are voting mechanisms for community members to indicate which translations are best.
Thicke: What innovation are you most proud of?
Duran: This whole ecosystem of enabling localization in different paradigms, for different people and different types of content. We have given other people the ability to participate in the process, and the process is becoming more flexible, timely and continuous.