We are Pangeanic. We have developed, tested and refined Deep Adaptive machine learning algorithms for autonomous publication. Deep Adaptive automatically selects the most relevant content from our 10Bn data repository to prioritize incoming material with different levels of aggressivity. This creates create state-of-the-art, ready-to-use neural MT engines that mimic the user’s style and expressions practically like a human translator.
We work closely with forward-thinking organizations to embed language AI in their processes. Deep Adaptive MT helps companies improve customer experience with more translated content, faster.
You own your content and you help improve your own language AI system, not others.
If you have a large amount of data and intend to be a heavy MT user, our team recommends training neural engines from scratch. This will allow our expert NLP personnel to clinically clean your data add it as training material at exactly the right time and with specific weights considering many factors, from sentence length to size of corpus and additional material to be used. Custom engines are typically built with Pangea stock data to cover large areas of language, modeling each engine to deeply understand the specific features of each language combination. In some cases, we can mine domain data to create a very domain specific MT engine that also covers many language areas. Our in-domain custom machine translation engines are tested by professional linguists using industry and academic human evaluation methods. Transparency in evaluation is paramount and part of an ongoing quality process and feedback. We have developed a specific platform for MT evaluation so our clients can have full control of quality evaluation, and involve internal evaluators at any point.
Clients start from any of our stock engines, built with 25-35 million bilingual segments covering most knowledge areas. Domain specific content is prioritized in 3 different levels of aggressivity (light, heavy machine learning or an aggressive training) as standard. Each level creates specific routines of machine learning and data weighing over the rest of the already learnt data. Pangeanic’s proprietary technologies will create a clone engine from our stock that will highly mimic the newly learnt vocabulary and processes. These engines are extremely fit-for-purpose and ideal when users have very specific client terminology that may be at odds with general terminology outside their sector (textile, banking and finance, law, construction, engineering, etc.) Deep Adaptive Machine Translation is very affordable for LSPs and corporate clients for peak production times or the long run. You can specialize a few MT engines for specific projects and call them once a year or every day or just create them for a specific need. No need to run glossaries!