On the Origin of LangOps
The evolution of the localization roadmap
by Andrew Warner
Lately, there’s been a lot of buzz in the industry about something called “LangOps” (short form for “language operations,” a term that’s admittedly a bit of a mouthful). Of course, you probably don’t need me to tell you that — if you’ve been flipping through the pages of this magazine (or scrolling through the digital version, for that matter), you’ve no doubt been following commentary on this budding concept, from experts like Arthur Wetzel and Renato Beninatto.
But few proponents of LangOps have been more vocal about it than the folks at Unbabel — particularly the company’s co-founder and chief technology officer, João Graça, who’s written about the topic quite enthusiastically in Forbes, calling it “the next frontier” for localization.
“The new paradigm we need is language operations — also known as LangOps,” Graça wrote in a 2021 piece for the widely read business magazine, where he introduced the topic to many readers who probably did not know much about localization beforehand. “While localization focuses tightly on how to translate and regionalize computer programs and websites, language operations is all about how we can efficiently operationalize this practice.”
He defines the term LangOps as a “cross-disciplinary function that helps global businesses communicate effectively with their multilingual customers and other stakeholders,” preempting this definition with the statement that the term is still evolving over time as it develops and takes hold in the industry. Graça emphasizes that LangOps isn’t just a pretentious or fancy way to describe localization. Instead, its proponents see it as a natural evolution of the practice, encompassing a broader range of language-related tasks that companies have to handle as they expand operations globally.
Ultimately, LangOps proposes a more principled and broad way of bringing technology into the localization workflow — “It’s a different way to do localization,” Graça said. His company, Unbabel, was founded on this paradigm in 2013.
If LangOps is indeed the new frontier for localization, then Graça and his team are its pioneers. In a mid-November Zoom interview with MultiLingual, Graça expanded on this notion and explained just what exactly the term LangOps means for the future of the language industry and the professionals working in it.
Before getting into all that, though, it’s worthwhile to look at what the localization industry evolved from in the first place — especially if LangOps is indeed the next evolutionary phase of the industry. Graça calls LangOps a fairly new practice, but the concept of “localization,” as we know it in our industry, hasn’t been around for particularly long either.
Though we often take the word “localization” for granted in the language industry, the term isn’t all that much older than LangOps, at least in the grand scheme of the English language. Prior to the 1990s, “translation” was the catch-all term that folks used to refer to much of the work localization professionals conduct today — if you told somebody you worked in “localization” back in the ‘50s or ‘60s, they wouldn’t have a clue what you were talking about.
“Back in the ‘70s and ‘80s, it was more true to just say ‘translation industry,’ and back then this industry was seen as a ‘cottage business,’” Michael Anobile, co-founder of the erstwhile Localization Industry Standards Association said in a 2012 interview with CSOFT International’s Ross Pfenning. “By the ‘90s, the international software companies and the multinationals, from a very pragmatic and conceptual perspective, saw this ‘cottage’ as a high tech resource group, and that’s when things started to move.”
Localization, then, evolved out of translation. It wasn’t until the dawn of the computer age that translators and programmers alike needed to come together and synthesize their skill sets, syncing up and essentially creating the field of localization. As Lionbridge’s then-content solutions architect Bert Esselink explained in the very pages of this magazine (back in 2003, when it was still called MultiLingual Computing and Technology), localization “revolves around combining language and technology to produce a product that can cross cultural and language barriers. No more, no less.”
This practice of going beyond mere translation to ensure that a given product or service also meets certain cultural standards has really only been around since the 1980s, though the word “localization” only appears to have achieved more widespread use in the English language in the ‘90s.
Esselink argued that localization as something distinct from mere translation evolved out of necessity. This was when lay people were first getting their hands on desktop computers. As technology companies like Microsoft began expanding their reach beyond the Anglosphere, they also needed to figure out how to make their products usable for speakers of languages other than English.
Of course, beyond just making them usable for consumers outside of the English-speaking world, software developers also needed to make their products and services, well, sellable. Underlying the birth of localization, then, is the notion that a product or service will sell best, not just when it resonates linguistically, but also culturally.
“Initially, software vendors dealt with this new challenge in many different ways,” Esselink wrote. “Some established in-house teams of translators and language engineers to build international support into their products. Others simply charged their international offices or distributors with the task of localizing the products. In both cases, the localization effort remained separated from the development of the original products.”
Localization has, of course, become a much more streamlined process since its early days. But as localization practices have since shifted and evolved from what they once were three or four decades ago, Graça believes it’s high time we use new terminology and practices to describe the underlying operation. According to him, LangOps takes a much broader view at language technology and its potential applications in areas beyond localization and translation — employing artificial intelligence (AI) to conduct sentiment analysis of product reviews, for instance.
It’s a natural step forward in an industry that’s constantly adopting and adapting to the latest developments in technology.
“LangOps is not just about localization — localization is just one application of LangOps,” Graça said. “Any tool that you use with language requires data to be trained. There’s a lot of them — you might use tools to generate replies, do sentiment analysis, summarization — and then there’s LangOps to manage that.”
Like many others who work on the tech side of the language services industry, Graça found his way into the business in large part due to his interest in computational linguistics. As a PhD student in computer science, Graça was particularly interested in machine translation (MT), entering MT competitions and studying machine learning methods. After completing a doctoral degree in machine learning and natural language processing (NLP), he worked toward his post-doctoral degree at the University of Pennsylvania, specializing once again in machine learning. Before he and his colleagues founded Unbabel, Graça said he spent time working on machine learning and NLP projects at various startups early on in his career — all of these experiences helped scatter the seeds from which LangOps has now sprung.
He and Vasco Pedro — CEO of Unbabel and one of the company’s other co-founders — began theorizing about some of the philosophical underpinnings of LangOps when a friend of theirs who worked at a travel startup in Lisbon brought up an interesting predicament: When customers emailed him in a language other than his native Portuguese, he wasn’t quite sure what to do. If it was a closely related language like Spanish, he’d be able to work his way through the text and “get away with writing” a response.
“If it was German though, he was completely screwed,” Graça said, noting that it was nearly impossible for him to figure out the meaning of the email without using Google Translate or some other MT tool. In turn, a simple email became a bit of a bottleneck. Graça said he and Pedro wanted to figure out how businesses could streamline communications with their customers, breaking the language barrier without having to wait a week or two for a project to be completed under the traditional localization workflow.
And in a sense, this is how LangOps was born — just as technological innovations like the desktop computer necessitated the birth of localization, LangOps was born out of the more widely available artificial intelligence (AI) tools that lay people now have access to, from Google Translate to large language models and other forms of generative AI (and perhaps the instant gratification general audiences have grown accustomed to in a world where everything and anything is accessible at the click of a button has also made speed and efficiency even more crucial today).
“We started with building something around AI and not using AI as a tool to help humans,” he said. “The principal was always to have machine translation and a crowd of people who don’t necessarily need to be translators, they just need to know the language to go in and make a few fixes here and there — and then you can work on a much bigger scale.”
To Graça, the shift from localization to LangOps parallels other efforts to operationalize certain practices in other tech-adjacent industries. In his 2021 Forbes article, Graça likened the term LangOps to similar terms in other industries, like DevOps, marketing ops, and sales ops. A year later, he expanded on this in his interview with MultiLingual, noting that new innovations in technology have allowed other industries to develop more streamlined operations while keeping humans in the loop.
As AI tools become more advanced, Graça believes this technology will have a similar effect on the localization industry to what cloud computing did for the field of IT, making the processes much more efficient and cutting down time on projects that may have taken weeks beforehand.
“If you believe that AI is the future, then it has to become an operational piece on the production pipeline,” he said. “And so LangOps was born from these two needs,” referring to the need for speed and the need to develop a more principled way of integrating AI into the production process.
Graça and Pedro have also been working alongside language industry professionals like Semantix’s Britta Aargaard and Trados founder Jochen Hummel to further flesh out this definition in the “LangOps Manifesto” — a 12-point list of qualities defining LangOps and the philosophy behind it.
Though he says the manifesto is still in development, it’s been circulating through social media in recent weeks and attendees of Unbabel’s flagship conference, LangOps Universe, this year will likely remember some of the key tenets of the LangOps Manifesto, such as the idea that LangOps is “language agnostic” and aims to help businesses understand and communicate with all customers, no matter the language they speak.
As LangOps has evidently started to catch on (just take a look at some of the commentary in this issue of MultiLingual as evidence of that), Graça believes it’s a matter of time before you begin seeing it reach a more mainstream crowd.
“Next year, I think it’s not just going to be us talking about it,” he said.
Half of all data is textual. Textual data is always multilingual. Businesses use only one or max two corporate languages. LangOps is an AI-centric approach to enable companies to scale across markets and languages.
LangOps PrinciplesLangOps Principles
1. Understand all customers
Our highest priority is to understand every customer no matter what language they use and expand our reach to the widest audience possible.
2. Support all customer facing functions
We initially focus on the most urgent needs, but promote organizational progress and architect solutions which can eventually support all customer interactions.
3. Embrace data-centric AI
Data is the key to performant AI systems for language. We collect, create, structure, maintain, and leverage textual data to deliver efficient solutions.
4. Try AI firstAI
systems are already good enough for many use cases and their performance will continue to improve. We always start by trying AI solutions to solve language problems.
5. Respect the human-in-the-loop
We use human intelligence where needed. We value the human contribution which constantly improves our machines.
6. Expect transparency, control and scalability
We create scalable systems with transparency, control to change and improve the process, and simplicity to ease operation.
7. Process data in real time
The world is moving to dynamic content creation. All content is a stream with different cadences. We build with real time at the core.
8. Build language-agnostic
We design all processes and supporting systems with ease of scaling to new languages. Users do not need language expertise, but can resort to multilingual datasets and external human contributors.
9. Promote interdisciplinary knowledge
We train LangOps in the different areas required to be efficient. We continuously enable teams to evolve and learn new skills.
10. Leverage available data and tech
We make use of available tools and data collections. We strictly account for total cost of ownership in buy or build situations.
11. Assess quality of AI
Quality is the pillar for trust in AI. Everything we create is proactively assessed through the lens of quality, so we can act before errors arise.
12. Be at the forefront
We invest time to research and study academic and commercial advances. We future proof LangOps by constantly pushing the boundaries.
Sometimes things become reality if you only put the right label on it. With that, the view on the real concept itself often becomes blurry…→ Continue Reading
Subscribe to stay updated between magazine issues.