AI Is Disrupting the $40 Billion Translation Industry
The translation industry is a great example of the disrupting powers of artificial intelligence (AI). This isn’t a theoretical article. This is a personal story of how digital translation changed my life as a management consultant and triggered an exciting technological (ad)venture.
Thursday, 10:35 P.M.
It is late at night. For the past several days, a management consultant has been working on a client presentation about reducing complexity and has received a great case study as input. Unfortunately, it is in the wrong language. “I’ll email the slides to our digital assistant Brian,” one of his colleagues says. Within three minutes, the slides are back and, after a short check, good to go. “We are done! Let’s go home,” they say already grabbing their jackets.
Deep Learning Advances and Translation Quality
Back in 2017, I found in the copy room a printout of a great New York Times article describing the story of how competing teams at Google pushed the advancement of the digital translation quality. They moved from the rule-based translation to deep learning—feeding the machines with translations instead of grammar. This article changed my view of AI.
In the meantime, Google was surpassed by Germany-based DeepL in a number of languages. And you probably have an idea why; superior data for the machine to learn from. In 2019, open.ai (acquired by Microsoft) reported another large technological breakthrough. Their algorithm with the catchy name “GPT-2” generates astonishing translations with much less input. The scary thing is that the translation capabilities are just an unintended by-product of synthetic text generation. The translation technology is already amazing and keeps evolving.
Need for Huge Computing Power to Mimic Humans
The translation algorithms are incredible but require crazy performance levels—computing power that was not viable a few years ago. DeepL, for instance, has a neural translation machine performing 5,100,000,000,000,000 (5,1 petaflops) computing operations per second and is one of the 30 most powerful machines in the world. To power this beast, you need a lot of energy. A lot of cheap energy. This is why they have chosen Iceland, with its green energy. High-quality data, smart algorithms and cheap power allow translating millions of words per second at a reasonable cost.
Insane Impact on Translation Speed and Cost
While translating the traditional way needs planning, coordination and hours/days of the actual translation, the AI-powered translation of documents needs seconds. Is it appropriate to call this an ‘acceleration’? Established translation providers ask for up to €0,5 per word. Automatic translation of documents with thousands of words costs <1% of that. Is it appropriate to call this a ‘price decrease’? It is great news for global interaction and sharing knowledge between humans around the world. But at the same time, it is a huge risk for the traditional translation industry.
Of Course, AI Translation Is Not Perfect (Yet)
Yes, we have to keep it real. Sometimes even the best translation machines fail to make a correct translation. The quality depends on the provider (DeepL is one of the leading providers, followed by Google), language-pair (German to English surpasses Esperanto-Maori) and the wording (translating single words is easy but sentences aren’t always clear). The translation quality is continuously improving, but we believe you will always have to check to be perfectly sure. Machines will struggle to mimic human empathy, domain knowledge, emotion, and humor.
What About the Established Players?
The translation industry, with revenues of ~€40 bn (according to Statista) will have to adopt the newest technologies quickly highlighting the aspect of quality assurance and specific domain knowledge. They will also be urged to substantially adjust their pricing schemes and SLAs in terms of speed. The ongoing globalization and accelerating content development leads to more text to be translated… But can this effect compensate for the price erosion?
The Next Technological Advancements in the Translation Industry
We expect further technological advancements leading to maintaining more of the context and nuances in the translation. The trend of integrating translation in individual applications (as we see, for example, on Facebook) will continue and the willingness to pay for standard translation will go down to zero. Because of this, there will be more focus on custom translations. Not solely reflecting the specific domain (financial industry, healthcare, etc.) but also considering the client-specific terminology. All three trends will lead to higher digital translation acceptance and adoption in practice.
How Digital Translation Changed My Life
For more than 14 years, I have been working in management consulting to solve complex problems and shape the future of amazing companies. During this time, I have seen too much time and talent wasted on repetitive and “annoying” tasks. I believe that technological progress enables us, humans, to do more of the exciting things—in our professional and private lives. Now, at AskBrian we are building, well… Brian. He makes translations and further state-of-the-art technologies available to business professionals. Brian is powered by AI, Brian is real.
Make Your Own Opinion!
In this PDF document, you’ll find a PowerPoint slide that has been translated by Brian from English to 100 different languages, without any adjustment or human intervention.
About the Author
Pavol is founder and CEO of AskBrian—a company developing Brian and specialized enterprise digital assistants. Business professionals working in fast-paced environments ask Brian, via simple emails, for Microsoft Office files translations, PDFs, slide graphics, industry benchmarks, and further assistance. Give it a try for free!