Supertext makes its mark on Silicon Valley – with second place in LocWorld35’s Innovation Challenge

Supertext CTO Rémy Blättler presented a solution on the topic of AI, translation memories and terminology onboarding that put even global players like IBM in the shade.

Supertext was full of excitement as it traveled to Silicon Valley in November for LocWorld, the leading localization industry technology conference. Rémy Blättler, co-founder and CTO of Supertext, brought with him his own software, designed to create termbases and translation memories for new customers.

Six months in 15 minutes

As part of the Process Innovation Challenge, Blättler gave two short presentations pitching the innovative solution against developers from well-known industry and technology giants such as Moravia and IBM – and achieved a podium finish with second place in the challenge.

The principle that the shorter the presentation, the more intensive the preparation held true here: the Supertext IT team had been working on the implementation for six months, assisted by specialists from the ZHAW School of Engineering and Spinning Bytes. The release is planned for summer 2018.

TMs and termbases for everyone. Entirely automatic and entirely free.

The software solves a widespread problem in the translation industry: when a customer changes agencies, they are rarely able to take their translation memories or termbases with them. These databases enable massive cost savings, but are extremely time-consuming to build. In comparison with existing providers, new agencies therefore often find it difficult to achieve the necessary efficiency.

Supertext’s software provides a web-based remedy – and all it requires is the URL of the company’s website. The tool crawls the website and automatically compiles a translation memory using the available multilingual content. Alignment is then used to directly create a supplementary termbase. The result: tmx and tbx files that can then be imported into CAT tools such as SDL Trados.

The concept is simple, but it’s powered by an interplay of diverse technologies, as well as plenty of artificial intelligence: the term extraction is based on GENSIM word2vec, a shallow neural network, and neural translation concepts such as the BLEU, Gale-Church and Levenshtein algorithms are applied during alignment. The technical details can be found in Rémy Blättler’s presentation.

Title image via Twitter: Rémy Blättler (left) with presenter Jeff Kiser and winner Tomas Franc.

Related posts

Leave a Reply

Your email address will not be published. Required fields are marked *