“Autodesk makes software for people who make things,” the American design software firm headquartered in San Rafael, California says on its website.

“If you’ve ever driven a high-performance car, admired a towering skyscraper, used a smartphone, or watched a great film, chances are you’ve experienced what millions of Autodesk customers are doing with our software,” the company added.

But this company, with over 200 million users, 8,500 employees worldwide, more than USD 2bn in revenue in the fiscal year 2018, and a market cap of nearly USD 30bn, is in transition.

Since 2016, it has been transitioning its revenue model from perpetual licenses for its software products to subscription-based plans. It has also been integrating AI-enabled algorithms, machine learning, and the massive resources of cloud computing into the design process, which it calls “generative design.”

“Think humans and computers co-creating designs in ways never before imagined,” the company explained.

At the heart of this corporate journey into the future of design is no ordinary localization effort to bring the software, the brand, and the company’s design narrative across the world.

“Autodesk software is localized primarily into 13 languages (French, Italian, German, Spanish, Czech, Polish, Hungarian, Russian, Traditional Chinese, Simplified Chinese, Korean, Japanese, and Brazilian Portuguese),” Jennifer Johnson, Director of Localization Solutions at Autodesk, told Slator in an email interview.

“Autodesk.com is available in 40 locales, while marketing campaigns and STKs are localized in close to 30 languages,” she added.

Full language sets are also available in Arabic, Belgium Dutch, Belgium French, Brazilian Portuguese, Bulgarian, Catalan, Croatian, Czech, Danish, Dutch, Dutch Netherlands, English, English Australia, English CA, English UK, Estonian,  European Portuguese, Finnish, French, French Belgium, French Canadian, Georgian, German, Greek, Hebrew, Hindi, Hungarian, Indonesian, Italian, Japanese, Korean, Latin American Spanish, Latvian, Lithuanian, Malay, Mexican Spanish, Norwegian, Polish, Portuguese (Portugal), Romanian, Russian, Serbian, Simplified Chinese, Slovak, Slovenian, Spanish, Swedish, Thai, Traditional Chinese, Turkish, Ukrainian, US English, and Vietnamese.

Johnson reveals that translation at Autodesk is fully outsourced to vendors or language service providers (LSPs).

Full Outsourcing

“Autodesk does not have translators in-house. All the work is done by vendors,” she said.

Its localization team of 65 is composed of one director, 39 working in the area of globalization solutions, 18 in program and business management, four in vendor and linguistic quality management, and three on the administrative side of things. Johnson counts its process and business model innovation among the innovations the team has introduced to make localization not only effective but also cost-efficient.

Members of Autodesk’s localization team. Photo courtesy of Autodesk.

She explained that “moving from a specialist vendor model to an end-to-end model where they have three preferred vendors that provide end-to-end-services across service categories”  provides scalability, flexibility, quality and reduced cost and risk.

By entirely outsourcing production to vendors, she said that the internal teams can also focus on defining and managing programs, working on innovation and process improvement, managing and working upstream with internal stakeholders, and managing outsourced vendors.

“Autodesk does not have translators in-house”

In turn, “focusing on working with internal stakeholders on upstream improvements to reduce the number of defects and effort downstream in the localization process” leads to a reduction in engineering and quality assurance efforts by over 50%.”

Translation Volume

As expected, translation and localization volume at Autodesk is very large with around 36 millions words translated every year.

“For products, the source language is always English. For non-Prod content, infrequently, we could have a non-English source language,” Johnson said. “The top 5 target languages by volume are Japanese, Simplified Chinese, German, Korean, and French.”

With this volume, technology is essential to the process. “We use several technologies from improving process and productivity to technology to manage our budget as accurate as possible,” she said.

Technology Innovations

According to Autodesk’s localization chief, the company has a localization platform based on a microservices architecture that helps automate largely the localization process.

“This also facilitates the continuous localization process as part of the continuous integration, localization and deployment pipeline,” she said.

They also have an internal cost management tool to manage the complex set of vendors, contracts, orders, and invoices, as well as the Autodesk Linguistic Quality Management portal, which centralizes all linguistic quality management activities. This is in addition to the home-grown statistical engines that process all product localization content and as well other types of content such as knowledge base or marketing.

To ensure translation quality, Johnson said that there are four main areas that they focus on: terminology management (which involves the extraction, nomination, translation, approval, change management of a term); reference materials (which involves the creation and maintenance of style guides, speak like a person guides, and transcreation policy).

Our preferred vendors fully own language quality

“We have created our own linguistic quality framework which takes the best of LISA / TAUS – DQF. Our preferred vendors fully own language quality and are required to “self-certify” all deliverables using the linguistic quality framework,” she said.

“In addition, we have a dedicated audit vendor that does a quarterly audit of the self-certified content,” she added. Moving forward, Johnson said that linguistic quality would continue to be front and center.

“Having said that, in keeping with our model, we do not see a demand for linguists in-house. We have seen an expansion of linguistic resources at our vendors in recent years as we have worked on stabilization of language quality. Our pool of subject matter experts and marketing reviewers has as well expanded, and there are plans to increase that further to plug holes that we see currently,” she said. “Demand [for linguists] will be high, although driven by language service vendors,” she added.

NMT to Impact Cost, Productivity, and TAT

Asked how she sees the impact of machine translation on the future of localization, especially in driving down cost or improving workflows, Johnson shared that the Autodesk Localization Solutions was one of the early adopters of machine translation.

The team has dabbled with both rule-based and statistical systems, before creating its own statistical MT system based on the open source Moses engine.

“This has been in production for about 10 years, and 100% of our product content goes through our statistical MT engines. Statistical MT helped us realize significant cost savings in the range of 5% to 35% based on the language and content type. In our suite of services, we also have commercial solutions from external providers,” Johnson shared.

“We do believe that NMT will surely impact cost, productivity, turnaround time and thus the process in the time to come”

“More recently, we have been doing research on NMT and are working on internal solutions as well as exploring commercial ones. We do believe that NMT will surely impact cost, productivity, turnaround time and thus the process in the time to come,” she added.

The team’s biggest challenge at the moment, according to Johnson, is setting up a new center of excellence in Dublin, which represents nearly a third of the overall team; delivering on the promise of a fully continuous localization process as part of the continuous integration, localization and deployment pipeline; and coming up with a neural machine translation (NMT) solution to replace or complement the existing MT infrastructure.

Image: Autodesk Newsroom