XTM International at JIAMCATT: The magic of Inter-language Vector Space

XTM International at JIAMCATT

XTM International at JIAMCATT: The magic of Inter-language Vector Space

Human brains are hardwired to make assumptions about words and their meaning.

“Sir Anthony Hopkins als bester Schauspieler gewann bei den Oscars 2021.”

On seeing or hearing this German language statement, your brain is performing hundreds of complex operations to make out its meaning in English.

“Sir Anthony Hopkins has won the Best Actor award at the 2021 Oscars.”

This example is particularly poignant, of course, since Hopkins won his Oscar this week for a moving portrayal of a man suffering from dementia. This reminder that the human brain, for all its wonders, is a vulnerable thing, comes at a time when artificial intelligence is offering us greater and greater support.

The 2021 International Annual Meeting on Computer-Assisted Translation and Terminology (JIAMCATT) offered us timely insights. From April 19th to 23rd, industry experts discussed the impact of AI, with an emphasis on the game-changing advances in NLP Artificial Intelligence, Inter-language Vector Space (ILVS).

Word alignment – how the magic happens

As Dr. Rafał Jaworski eloquently explained, ILVS mirrors the linguistic problem-solving of the human brain on a macro-level. Just like your brain, ILVS is quick to sort out similarities between words across 250 languages. That makes a whopping 31,125 language pairs in total!

JIAMCATT XTM International
Inter-language Vector Space: word-level alignment

While the processes underpinning ILVS may seem complex, the mechanism itself is simple and elegant. The magic lies in the amount of data used and the speed at which these alignments happen; more than 200 terabytes of textual data are used. That’s the size of the Bible multiplied 40 trillion times, and the alignments are completed in well under one second!

The XTM AI team took the research of Google and Facebook one step further – they harnessed bilingual dictionaries and performed the alignment of vector spaces. In the ILVS framework, each word has its unique fingerprint, a mathematical vector to describe the word within the corpus and its relationship with other words.

The impact of  Inter-language Vector Space

ILVS was first introduced in XTM Cloud 12.4 (released in July 2020), and immediately became a go-to source of intelligent automation for linguists and localization managers. “Do more, faster, with ease” is the philosophy behind this NLP technology. Enhancing productivity, freeing users from tedious, repetitive localization tasks.

Here are just a few of Inter-language Vector Space’s many benefits:

    • Automating over 90% of bilingual term extraction. Project managers can create accurate glossaries during TM alignment. XTM Cloud version 12.7 is further allowing for bilingual term extraction on files such as XLIFF or TMX during the TM import. It takes 85% less time to create glossaries, and users see up to 90% accuracy for extracted terms.
    • Up to 20% improvement in translator productivity. This increases up to 80% for MT post-editing. Thanks to the auto-placement of inlines, linguists can translate text with inline elements such as HTML tags with speed. The algorithms automatically transfer inline elements from source to target segments, saving the linguist’s time and boosting their productivity.
    • Up to 90%+ effort and costs benefit for new translation projects where no previous TM exists. The application of auto-alignment helps create translation memory quickly from zero.

And we’re only scratching the surface; more enhancements will follow. Among them is Computer Aided Review that enables the highlighting of translation errors during the QA phase.

XTM AI delivers intelligent automation and enhances your linguistic productivity. It’s that simple.

JIAMCATT exists to bring people together in pursuit of more efficient, more secure and more people-focused use of translation technology. This year’s conference, coming after a year of unprecedented disruption and uncertainty, was both reassuring and inspiring. In the same way that we’ve seen private and public sector leaders unite against the common enemy of coronavirus, JIAMCATT saw leaders unite to discuss and celebrate our common friend and ally, language service technology. Helping us communicate, helping us trade. Not replacing human ingenuity, giving it a platform to soar. Thank you to everyone involved for an excellent week!