Accenture is a global Fortune 500 company providing management consulting, technology services and outsourcing. Accenture has around 246,000 employees and serves clients in more than 120 countries around the world. Michel-Etienne Liegard is the global language translation capability lead at Accenture. He supports an industrialized process for translation management within Accenture in support of the communications between Accenture customers and their international development teams.
Thicke: Coming from the development side, how did you end up leading language translation capability in Accenture?
Liegard: I joined Accenture 22 years ago as a developer. In 2009 I moved to a team in the United Kingdom that was looking at methodology and tools for industrializing the process of applications development. Language was one of the areas we identified as having potential for improvement.
Thicke: How did this language project begin at Accenture?
Liegard: Accenture manages the externalization of IT activities for its customers. We realized a couple of years ago that translation is starting to take on more importance in this domain. More and more IT projects require language skills, and English is not necessarily a lingua franca in these types of projects where development teams located in India or the Philippines, for example, must understand IT project documentation from customers who speak Spanish, German, French or Portuguese. In 2009 we defined where we wanted to go with translation activity for projects.
The purpose of this initiative was to allow complex IT projects to be developed in Accenture delivery centers in India or the Philippines. Our internal clients may have difficulty communicating in English, so we needed the tools to facilitate this communication. Today we are using machine translation (MT) to manage internal communications of technical specifications, which need to be translated accurately into English for the developers, then sent back to clients in French or Spanish, for example.
Thicke: So Accenture is using MT to translate specifications between customers and engineering teams. What was your process for deciding which engine to use?
Liegard: Our original goal was to understand how MT could be integrated into our process and to develop this activity for our clients. Before deploying in 2009 we benchmarked a number of tools on the specific architecture and security that our clients need. We benchmarked ten to 12 tools, all the ones in the market.
Thicke: What tool did you choose and why?
Liegard: In the end we chose SYSTRAN as the tool that suits our purpose. First of all, SYSTRAN architecture is compliant with our internal security policies.
Also, one of the strengths of SYSTRAN is that it is flexible enough to allow distributed users access to easy tools for translation, as well as some advanced translation features for translators. Each and every one of our team is able to consult client profiles, which are accessible everywhere. With SYSTRAN, a specialist is able to access a client domain, and a translator or an engineer in India is able to use the same terminology.
Security of data is important to Accenture — Accenture can not reuse documents from one client for another. To reuse existing content to build the statistical language model we would need the approval of our clients. Hybrid MT (statistical and rule-based) like SYSTRAN is a good alternative to statistical tools that could be difficult to use in our context. Indeed, a pure statistical engine would require an initial set of documents for good quality translation that we don’t have when we start a new IT project.
Thicke: How do you benefit from using MT?
Liegard: Our goal is to make language translation within the development process fully transparent by making sure that there is no bottleneck — and language translation can be a bottleneck. We are limited by time when developing an application. There are different phases of delivery during the lifecycle of an application, and in each phase we make sure that the specifications for the developer of the application will be translated smoothly and sent to the right person for the development.
Our translation needs vary by phases. In some phases, a high volume of translation is needed to provide support materials to the team who is managing the development of the applications.
We do have interpreters assigned to projects — we have a pool of translators who fly to support extra activities, and the most advanced are supporting client visits internally. We also provide systems maintenance. Our clients expect us to support the maintenance of their application. So in our process we have also integrated some translators able to manage to make the translation of ad hoc incidents. What’s important is how we manage to make the process seamless for the clients within the delivery lifecycle of an application.
Thicke: How many translators do you work with in this process?
Liegard: We work with around 120-160 translators internally.
Thicke: What roles do your translators play?
Liegard: Normally the translators need to post-edit, but they also build client context through the creation of dictionaries and translation memories (TMs) that they need to revise and improve. Besides post-editing, they provide maintenance of the application, making sure that the TMs and dictionaries are up to date.
Thicke: Do you measure post-editing productivity?
Liegard: Yes. Our post-editors have been able to improve the engines to meet the target productivity. We are well above the baseline provided by external analysis, and are close to the target of 3,000-4,000 words per day.
Thicke: Do translators like this post-editing work?
Liegard: We see a very interesting profile in our translators. We are used to working with people who have an IT profile, but we realized that linguists analyze sentences just as engineers do when analyzing a technical problem.
Thicke: How important is MT to Accenture’s customers?
Liegard: There is a real need for this. Whenever we have a misunderstanding where the people don’t speak the same language, it generates significant amounts of lost time, which can be translated into elevated costs. We compare this with doing the translation quasi-immediately. There are multiplication factors; the later any action is taken, the higher the costs will be. So the time costs with MT are much less for the client. And the process is much more accurate — it is crystal clear.
Thicke: Do you have SYSTRAN integrated into a workflow process?
Liegard: We use an internal engine. We started with customized SharePoint to push the documents to the translators. We also use our customers’ processes and tools to manage codes and documents in the same process.
Thicke: Coming from the IT side, as a developer yourself, you are bringing a unique perspective to solving the language problem.
Liegard: I don’t know if it’s a unique perspective — other IT companies are currently doing the same thing. But maybe not at the same scale. Also, translation had to be integrated right into the development lifecycle so the context is different.
Thicke: How would you rate this work so far?
Liegard: We consider this language translation initiative as a success. Last year we translated 32 million words in French, German, Spanish and Portuguese. MT allows us to supply more services to our clients. Language could be considered an issue for our clients. Now it’s no longer an issue.
Thicke: What is your next challenge?
Liegard: For clients who have understood the interest of having this industrial process in place for translation, we plan to go to the next step of increasing translation productivity by proposing an industrialized model for large translation volume.