We’re proud to present PangeaMT, the company’s technology division and innovation hub driving Pangeanic’s language processing solutions. PangeaMT is where Pangeanic’s cutting-edge translation engines are created.
How was Pangeanic’s technology division founded?
PangeaMT leads technological development within Pangeanic. As a technology department, its work responds to specific industry needs. This division benefits from over ten years of machine translation experience and knowledge brought by CEO Manuel Herranz.
“In 2008, we collaborated with Valencia’s Polytechnic University and several other European universities to devise solutions for our customers, with a particular focus on the automotive and professional electronics sector. In those days, masses of documents accompanied machines – technical user manuals and so on, these used a “controlled language.” That’s to say, no-frills, clear language, and easy to understand terminology and expressions.” – Manuel Herranz
New collaborations for statistical language recognition patterns followed, and so a department explicitly devoted to technological development was created.
Thanks to statistical mathematics applied to language, Pangeanic became one of the few companies globally to use SMT (statistical machine translation) to produce publications faster and more successfully in languages with similar syntaxes.
From 2010 to 2015 PangeaMT struck another milestone when it formed a partnership with the Japanese Toshiba. Hybrid engines were built as part of the collaboration, aimed at solving problems in foreign languages, given statistical limitations at the time.
Since then, PangeaMT has grown and evolved, driving innovation within the industry. Keep reading to learn about use cases and the technologies that are making a difference in the industry.
How machine translation and neural networks helped information flow during the pandemic
A significant challenge facing machine translation is the acquisition of quality data for training. In order to build hyper-specialized engines, large amounts of knowledge and data are needed for the AI to analyze and learn from.
TAUS provided PangeaMT with nearly 2 million words to build healthcare-specific machine translation models. At a time when COVID-19 hit society the hardest and vaccines were underway, these customized solutions enabled medical papers to be translated from English into Spanish, German, Polish, Russian and Chinese.
Find out more about this PangeaMT project: Adaptive Machine Translation for TAUS
Deep Adaptive, the key to making machines understand human language
Manuel Herranz explains what Deep Adaptive, the technology that is revolutionizing the machine translation industry, is all about:
“Deep Adaptive is a process whereby a translation engine is perfectly adapted to a client’s specific needs, applying not only preferred terminology but also its own ‘twists’ to the output (a “seat” or a “chair” can mean different things if we are speaking about eCommerce or political seats or chairs in a Parliament, for example). Deep Adaptive Machine Translation is able to even mimic the style! Through this process, an existing engine analyzes the reference material and examines its terminology and characteristics. Then, it searches through our huge database for similar material and applies it, prioritizing the terminology and the style to create a deep adaptation. This has implications to users, as they now can manage their own AI solution in the cloud in complete privacy.”
Deep Adaptive technology allows machines to perform much deeper analysis, matching source and target language in an almost human-like way. We tend to say they understand human language, as these engines are not limited to automatic translation. Instead, they take into account parameters that until now were exclusively performed by humans. These engines can identify the stylistic use within a company or institution, even the tone and voice of a brand. They learn from data provided by humans and create a unique learning loop for each client.
With PangeaMT, this technology applies to different NLP (natural language processing) services like anonymization, summarization, web crawling, or document and data classification. It’s made up of a wide range of solutions and technological developments combined, generating a powerful database for training artificial intelligence and information management processes.
How can we use Deep Adaptive technology?
Manuel Herranz provides an example of the benefits this development offers to companies in different industries:
“One of our clients wants to monitor news that is only published on certain websites and in certain languages. Once we identify which news does or doesn’t interest these journalists, we carry out high quality translation into Spanish before categorizing articles by relevance with advanced sentiment analysis technology. Keywords are then extracted and the article is classified within a specific field. From there, our clients extrapolate the information for their journalistic analysis and conclusions. Our research is focused on making a summary or abstraction that also provides you with a lead-in to get a bird’s eye view of what the information is about.
This type of solution is particularly relevant for large organizations so they can manage knowledge.”
The value of data; the future lies in its generation and analysis
Immersed in the era of Big Data, at a time where many companies have the opportunity to generate large amounts of information to work with, Manuel makes his position clear:
Organizations that are not aware that they have a data problem, have a big data problem.
Collecting data is no longer enough, the real value is in its analysis, in the deep knowledge that it can provide to an organization. Information about its users, customers and community; better analyzing what is happening with and around your community, proposing better solutions that meet the needs of your audience and building customer loyalty.
The capacity to monitor multiple language perspectives and extract truly relevant information, truly is deep analysis. How we communicate regardless of language, what opinions a product or service receives, when and how people are mentioned in texts all form part of it. Technology must work to extract all these parameters for further human analysis, so we can make better decisions.
“By applying advanced mathematics, we can “predict” the odds of something happening, and that’s really valuable. We can predict how a sentence is said in another language while also adding a spin on it if desired. We can predict the likelihood of negative comments following, say, a product launch, and also extract the worst results – helping you figure out what exactly went wrong or what your users don’t like.
We can classify information at a speed at which humans cannot function.
This generates new and valuable jobs for society too, as we rely ever more on information to make us better humans.”
The key, as Manuel concludes, is in the powerful combination of artificial and human intelligence.