Our Analysts' Insights

Technology developments tend to follow a typical pattern of improvement over time, known as an S-curve. Although it is a familiar pattern, it is worth unpacking its five phases and considering how they apply to language technology and forecasts about it. Examining how they have played out with successive generations of machine translation points to a future in which other advanced natural language processing technologies have tremendous potential to deliver useful and innovative capabilities.

Last week, the Washington Post published an article about Blake Lemoine’s claim that his employer Google’s LaMDA language model/chatbot system had achieved sentience and had a “soul.” Lemoine, an engineer in the company’s responsible AI group, based his assertion on a dialogue in which LaMDA expressed human-seeming sentiments and concepts. Google placed Lemoine on leave, thereby sparking renewed discussion about what machine sentience is and what it means. What can the experience of the lan...

In recent years, there’s been a lot of buzz around “headless” systems – whether for content creation and management or for the translation workflows that feed the global customer experience. The concept being that rather than having a traditional front- and back-end (publishing and creation), these systems allow content to be magically managed, extracted, repurposed, and delivered through a myriad of end points, from mobile apps to corporate websites integrated with a partner’s own custom p...

In October 1991, Unicode 1.0 was first released. In the 30 years since that publication an entire generation of language workers have been educated and started work, never having had to know the “joys” of trying to ungarble text that had gone through multiple encodings. The introduction of Unicode has simplified life for many of us and allowed millions and millions of people to access digital resources in their own languages.

Early in my career, my wife and I lived a few houses down from a truly extraordinary woman named Lorna Call Alder. Born in 1906, Lorna was the first person to develop a now-ubiquitous approach to foreign-language education that emphasizes the deliberate and careful introduction of vocabulary in context and repetition in use within instructional materials. Almost every translator today who has learned a language through formal education owes a small debt of gratitude to this humble woman, who pas...

Our current research into continuous localization shows that the lines have begun to blur between what it means for localization teams to support a traditional Agile model versus one that is more continuous. However, the 53 people who granted in-depth interviews generally agree that the end result remains similar. Organizations must automate as much of the localization process as they can in order to deliver services for mushrooming volumes of content and code that come to them in ever smaller p...

Long gone are the days when only the biggest enterprises or language service providers had their own translation management system (TMS) hidden away in a private datacenter. The cloud has made TMSes and machine translation (MT) accessible to all. Technology costs have gone from prohibitive for smaller LSPs to within reach for everyone, including newcomers to the industry. Some commercial solutions even have innovative pricing models that make the technology affordable – or even free – rather t...

In late 2015 most developers still treated Neural Machine Translation (NMT) as a future technology that would come someday, but by the middle of 2016 several developers had released functional neural systems that seemed to outperform even the best statistical MT (SMT) engines. Since that point, momentum has shifted decisively in favor of NMT because it delivered better fluency in its output and seemed to promise the end of incomprehensible MT output. Eventually, this rate of stellar progress slo...

In 2017, CSA Research introduced the concept of “augmented translation,” a technology-centric approach to amplifying the capabilities of human translators. Although companies such as Lilt, SDL, and Unbabel had already implemented portions of this model – namely the tight integration of human and machine capabilities, we predicted that more technologies would be folded in over time to create an AI-driven platform for linguists that would make them far more efficient and capable by offloading r...

A recent examination of how computing power has changed over time calculated how much it would cost today to rent all of the computer capacity that existed at various points in the past. Up until 1968 all the worlds computers’ combined capabilities fell short of what a single server could do in 2018. It was not until 1974 that the then-current capacity (roughly equal to 0.43 teraflops) would exceed £100/month to lease in today’s costs. This exponential drop in the real cost of computing power ...