Georg Ell, Phrase CEO
Setting the Pace in Localization Technology
St John’s College, Cambridge
NUMBER OF EMPLOYEES
FAVORITE PLACE VISITED
I took part in a round-the-world sailing race — against the prevailing winds — in 2004
When it comes to the modern language technology business, keeping your eye on the ball is essential. It’s a complicated, ever-changing world, which won’t come as a novel observation to anyone who has been paying attention to the past few years. Georg Ell, CEO of Phrase, is one of those careful observers. Since his appointment to the position in April 2022, he’s shepherded in a transition from the company’s former identity, Memsource, to the Phrase banner it flies today. But in the midst of that transition, Ell was careful to not lose focus on what makes Phrase the successful business it is: delivering for its customers. What’s the company’s secret? Well, according to Ell, it all comes down to being deliberate about focus, teamwork, and collaboration. This means setting values such as “Teams Built on Respect” and an operational cadence that ensures prioritization and cross-functional cooperation. And to Phrase, that means alignment around three key pillars: artificial intelligence (AI), workflow, and scale.
You might say the language industry stands on the cusp of a new future defined by rapidly evolving technologies. What is Phrase’s position on grappling with these changes while staying true to its values?
It’s an incredibly exciting time to be in the industry. For a long time, much of the language industry has been viewed as an afterthought, relegated to the end of the development lifecycle in both product and marketing. Those of us in the industry have been frustrated with the missed opportunities associated with this.
Since ChatGPT burst onto the scene with version 3.5 in November, and even more with version 4 in March, business leaders around the world, in every company, are trying to understand how they can incorporate language AI in their company strategy. We are at the forefront now, in the eye of the storm of this evolving landscape. So we think it’s the most exciting time ever to be in the industry.
We have a point of view on the future that in the next five years, all types of content on the internet could be streamed in real-time. There may be no canonical version of a website, with all manner of content being generated for each visit of individual users, meaning the volumes of language text floating around the internet will be astronomically larger than they are today. This explosion of generated content will result in increased complexity, and will require the effective use of AI — hence we are prioritizing three pillars AI, workflow, and scale to address this vision.
There will be no dominant general language AI, large language model (LLM), hyper-scaler, or company that comes from nowhere to be the single, best-for-all-use-cases ML model. Instead, there will be a plethora of models, machine learning models, LLMs, or machine translation (MT) engines, including custom ones built in-house by companies for particular purposes from focussed datasets. We want to play a neutral role and help customers to navigate all of those different models and pick the right ones for them to use. We believe that the dynamic quality estimation technology required to effectively pick and choose the best combination for the job at hand is vital to the value we can add.
Even as businesses achieve enormous scale and unlock new opportunities with all of these different models, they are still going to need a degree of humans-in-the-loop. The percentage of human involvement may drop, but as volumes grow, then the absolute quantum of human-in-the-loop may well remain constant. In any case, as we’ve said, these factors inform the three major development areas for us now: AI, workflow, and scale.
This journey is a continuation of exactly what we’ve been doing for 12+ years — building the best enterprise-class language technology in the world. We are clear that we are a software company — not services. Our customers tell us time and again that they don’t want to be tied into a ‘turnkey’ solution which locks them in. They want composable, configurable, and scalable technology that caters to their complex needs, guides them in selecting the best AI for their challenges, and brings the best analytics available to manage their translation pipelines with a vendor neutral approach to help manage the various language service providers they need. We’ll continue to drive cutting-edge research and development into our core areas of focus, so that we can deliver absolutely the best enterprise-grade language technology suite for our growing customer base.
What do you envision the future of localization to be in light of that?
As I’ve said, the future will be built around harnessing new levels of AI, workflow, and scale, to succeed and thrive in a world of exploding content volumes. We think volumes may grow a billion-fold as GenAI takes hold — and even if we are wrong by several orders of magnitude, we believe we can still be sufficiently ‘right’ to cater for the future which will involve more machine-powered language technology. Localization has, by definition for a long time, relied extensively on human services. The manual process of content localization simply can’t meet the existing latent demand, let alone the increasing volumes we see ahead with Gen AI. The majority of enterprise content is still not localized because until now it simply hasn’t been economic or fast enough to do so. That starts to change now, with our recent releases, Phrase Language AI and Phrase Custom AI in particular.
A future-ready company strategy needs to consider how to prepare for the AI driven acceleration of content and how best to use that to meet customer expectations and stay competitive. Localization teams that harness technology in the right way will become even more strategically important to their organizations as they expand the value they deliver to new areas.
Front Row Left to Right: David Čaněk – Founder & Board Member, Simone Bohnenberger-Rich, PhD – CPO, Georg Ell – CEO, Dr. Alon Lavie – VP AI Research.
Top Row Left to Right: Bharat Siyani – VP People, Jason Hemingway – CMO, Stephen Lumenta – CTO, Andrea Tabacchi – CCO, Ian Woolley – CRO, Martin Konop – CFO.
AI, LLMs, MT, and more are all changing by the month, and any current shortcomings may well be ironed out as it progresses. What new or expanded capabilities do you envision down the road, and how will it intersect with language work?
We need to trust the quality of machine production and evaluate what will need to be human reviewed rather than by machine. As well as how to adapt these technologies over time as things change. From the outside, individual company needs for localization can look quite similar, in reality the implementation and requirements are different almost every time. This means integrations, flexibility, configuration, and composability are key to the next-generation of localization software.
Advances in custom model building, MT engine aggregation, quality estimation, and workflow automation are all new and expanded capabilities that will come into play, making processes more efficient and reducing the need for manual oversight. In essence, as these technologies advance, they will continuously redefine the interplay between machines and humans in the realm of language work. Adaptation will be at the core, ensuring that while we leverage the power of AI, we also retain the nuanced understanding and expertise that only humans can provide.
The landscape is experiencing a paradigm shift. Cutting-edge advancements in GenAI, MT, and QE are dismantling long-standing barriers. These technological leaps promise to elevate automation rates. We aim to reduce the percentage of human touch from the level that it is at today and that some things will go from heavy human touch to medium, medium to light, and light to none — to move things down that scale so that more and more volumes are economically viable to be localized at scale, with high quality, and the easy, fast turnaround that a business needs. Until now, only a fraction of an organization’s content could be effectively localized — today, everything is addressable.
It’s clear that the capabilities down the road have to provide value to all parts of the business, adding new levels of accessibility, quality estimation and workflow, as well as the central analytics to show results and manage costs. This balanced and democratic model, where technology meets human expertise, has deeply resonated with large enterprises, underscoring our commitment to understanding and fulfilling their diverse needs, speeding up the time to market while maintaining the integrity of translated and localized content.
In our recent September “Cadence” product releases we announced new capabilities for the Phrase Localization Suite that demonstrate how these developments are coming together. With Phrase Language AI we’re helping our customers scale their MT offering to all employees, via an API today and a completely new portal that’s coming soon. This means high quality MT, including custom MT models trained by our new Phrase Custom AI, can be made available for every employee in a company-approved, secure fashion. We’re already seeing this open up new opportunities for affordable, high-quality solutions to previously non-localized content, without over burdening the existing localization team, who can now provide an even more value-added service to their internal stakeholders.
What is Phrase’s vision for embracing this developing future? And what’s the roadmap for achieving this vision?
This rapidly developing future is rooted in the transformative promise of generative AI and workflow at scale. I have often wondered why our online experiences on major websites remain so static and universal. The potential of generative AI is such that it can revolutionize the internet as we know it, creating a hyper-personalized experience.
Instead of visiting a website and seeing a standard interface, imagine a world where the content and every interaction dynamically shifts based on our interests, recent activities, or preferences. It’s not just about a tailored shopping experience; it’s about an entire online experience delivered in real-time, around each individual across every interaction they have.
Such a future paints a picture where content production increases exponentially, in magnitudes beyond our current comprehension — potentially in the 10^9 order of magnitude. And with such colossal content generation, the traditional content value pyramid — where high-value, low-volume content is human-curated and low-value, high-volume content is machine-generated — will invert. The bulk of our content, even the high-value sections, will be machine-generated, simply because humans cannot produce content at the scale machines can.
Phrase is positioning itself for this imminent transformation. We will enable our customers to facilitate hyper-automation by moving to increasingly lower touch localization, while giving them full control over every step in the process according to their needs to balance quality, speed, and volume. This will allow our customers to: reach new levels of scalability with dynamic and hyper-personalized content and achieve radical unit cost savings by automating even “hard to solve localization problems”.
For over a decade, we have been laying the groundwork with:
- An API-first approach: For the past ten years we’ve been building with an API-first mindset. This ensures flexibility and seamless integration with other technologies and platforms our customers use.
- An emphasis on machine learning: Over the past five years, we’ve deeply integrated machine learning into our product.
- Sophisticated workflow automation: For the past two years we’ve been building a sophisticated workflow automation capability with Phrase Orchestrator, that was launched earlier this year.
- Dynamic evaluation: We’re moving towards dynamically evaluating machine learning models. As content flows in, it will be assessed based on various customer constraints, from time to budget. Instead of banking on a single AI model or MT provider, our strategy is to offer quality visibility across all available tools. We aim to pick the right tool for the right job at the right time, ensuring optimal application of machine learning technology.
- Human-AI collaboration: Despite the surge in content volume, we believe human intervention remains invaluable. Even if a small fraction of massive content volumes requires human review, we’re prepared for it. The complexity is undeniable, but our focus on AI, workflow, and scale ensures we’re well-equipped.
We’re not in the race to create the winning AI model; we’re here to ensure businesses can make sense of the evolving AI landscape, adopt it with optimal quality, and integrate it seamlessly into their operations. Our vision is to guide businesses, to help them navigate a real-time, dynamically generated, content-rich digital future.
Could you tell us about the process of building your leadership team?
To deliver against our vision of the future, we’ve built our teams around the world in the US, EMEA and APAC. When doing this, we’ve been very clear on our values:
- Fully embrace diversity
- Teams built on respect
- Show up to make a difference
We’ve recruited leading talent at multiple levels from C-Suite, VP, and Director level, and throughout the company, finding people who bring world-class expertise and align with our values. We have been — and continue to — work hard on building a culture that attracts talent from both around the industry and to join from outside. You can see the combination of priority focus areas and emphasis on values, for example, in the appointments of Alon Lavie as VP, AI Research, who joined with an experienced team, and with our Chief Product Officer, Simone Bohnenberger-Rich. And as you see in our team photo here, we also benefit from the continued advice of David Čaněk, original Memsource founder, who sits on our Board. This combination of continuity of experience, world-renowned talent, and fresh thinking, is a potent mixture driving our high ambition level for what is possible at Phrase.
Likewise, the company itself has gone through its own evolution, transitioning from Memsource to the Phrase branding we know today. Can you reflect on that journey and some of the lessons learned along the way?
The transformation from Memsource to Phrase, which took place exactly a year ago, was guided by a strong sense of purpose and determination. This might come as a surprise, but the catalyst for this name change was David Čaněk, Memsource’s founder. When the acquisition of Phrase by Memsource took place, the imperative was to unite under one distinct name, ensuring a seamless integration. “Phrase” was David’s choice. We spent months meticulously refining our brand identity before launching in September last year. We were sure to be highly intentional about it and extremely clear.
For example, months before the launch, we made a concerted effort to live the brand internally. We integrated the new brand identity into our weekly company meetings, internal slide decks, and out internal software, using the Phrase logo and name wherever we could without alerting the general public or customers and partners to it. And we talked specifically about the kinds of language that it was OK to use or not to use in relation to our brand, so that we coalesce on a common vocabulary which is really important when you’re trying to get a new identity across.
Furthermore, we built a new way of operating within the company based on this concept called “the cadence”. It’s a quarterly organizing principle that we have found has helped us to be much more disciplined and coordinate more effectively across different departments in the company. While it’s not perfect, it’s improving all the time, and it also improves our ability to forecast our roadmap and be clear about what’s coming and when and to make commitments on that basis. When we say something is in a cadence, it has a more than 98% chance of hitting that cadence, which is good for a product roadmap.
Could you speak to the work going into recent releases?
At the heart of our recent releases is our focus on leveraging the power of AI to define the next-generation localization software. We’ve introduced several innovative AI capabilities to the Phrase Localization Suite to improve translation delivery times, enhance the quality, and manage costs more effectively.
One of our significant innovations this month is the introduction of Phrase Language AI, which was formerly known as Phrase Translate. This is not just a rebrand, but an extension of what we offer. Phrase Language AI incorporates a combination of MT and LLM capabilities. Its unique feature is that it selects the optimal MT engine for each case, offering better translation quality than relying on a single vendor. As I’ve mentioned earlier, in this release we also made Phrase Language AI available via API, which opens out access to every employee in a business to a secure, company-approved, MT capability. In the near future we’ll also be adding a new portal capability to make access even easier. By doing so, we’re providing companies with the means to localize a broad spectrum of content, from learning & development material, or legal contracts, to internal communications and sales and marketing material, at an accelerated pace without compromising on linguistic quality.
Taking our focus on quality even further, we’ve introduced Phrase Custom AI. This allows businesses to develop machine learning models tailored for specific needs, ensuring high-quality translation, even for complex content. Historically this would have taken weeks or months to do, requiring specialist skills and a lot of time and cost making custom engines out of reach for almost every company. We’ve ensured that this customization process is easy and fast, without the need for specialist skills or external professional services, putting them in reach of almost every company in the world, that’s a fundamental inversion of the status quo. What previously took weeks or months and specialist skills can now be done in hours for a fraction of the cost with Phrase Custom AI.
These custom models, based on our proprietary Phrase NextMT engine, can be accessed via Phrase Language AI, democratizing access to high-quality translation across an entire enterprise, so every employee in an organization can access these custom models via API, should the customer choose. Furthermore, to improve user experience, we’ve enhanced Phrase Orchestrator, our sophisticated localization workflow automation tool, with a new library function pre-populated with templates so users will find it even more intuitive. And with enhanced file management it’s even more adaptable to new applications including language pivoting. We’ve also expanded our language capabilities with new language pairs for Phrase NextMT (English–Swedish and English–Japanese) and introduced several other enhancements to our Suite.
Our vision at Phrase is clear. We aim to continuously define the next-generation of localization technology with our innovative approach, providing customers with what they should and can expect: real solutions that add immediate business value. We believe that by integrating AI, workflow, and scale effectively, with a keen emphasis on Dynamic Quality Evaluation, we’re paving the way for the future of localization technology.
What are you focused on right now?
At the moment, we’re diving deep into refining our pricing model to better cater to the evolving needs of our customers. Instead of licensing our various products separately, we’re making a significant shift. We’re launching a single subscription for Phrase. With this, subscribers will instantly get access to all of our comprehensive capabilities: from Phrase TMS, Phrase Strings, Phrase Language AI, and Phrase Custom AI, to Phrase Orchestrator, Phrase Analytics, and Phrase NextMT.
The beauty of this model is its inclusivity. Be it a startup, a mid-sized firm, or a giant enterprise, everyone starts with the primary platform license. Customers can then buy additional capacity, but the foundation is that every customer immediately has access to all our capabilities. Essentially, it’s like having a complete language suite at your fingertips. You decide what your business needs, whether it’s software/app localization, AI/MT, analytics, workflow automation, or more, and tap into those specific capability sets. But that’s not where the excitement ends. We’re also exploring avenues to seamlessly integrate with third-party systems, whether via API or other interfaces, enabling our customers to plug in the capability sets, in a manner that aligns with their business operations.
And that’s where our Suite really stands out. Already today, we believe we are the superset of at least 6 other types of companies — providing 90% of capabilities offered by other TMS providers, MT aggregators, workflow companies, website translation, and machine language quality estimation companies. We’re not just another provider in the mix; we are the amalgamation of all these capabilities. What we offer our customers is an affordable pathway to access this immense capability pool, saving them from the hassle and costs of sourcing these services from half a dozen other providers. I think a key message here is that we are opening “new possibilities” for localization teams to scale their efforts to every department in a business using a combination of human and MT that is tailored to what they need, removing the constraints of today’s localization model.
If you’ve been trying to wrap your head around the concepts of blockchain, Web3.0, or the metaverse, look no further. Nimdzi’s Nadežda Jakubková has you…→ Continue Reading
Subscribe to stay updated between magazine issues.