Please activate JavaScript!
Please install Adobe Flash Player, click here for download

oec_04_ePaper

Oec. Dezember 2015 27 Accessing the global brain The young startup company Kunendo did what nobody has ever done before: Translating a book from English into German in four hours. The secret lies in a new programming language, which cleverly combines computational power with human brainpower. Priska Feichter TAKE-OFF At the same time, the system is designed to reduce or even eliminate human errors. Prof. Abraham Bernstein, who initiated the project and acts as an adviser to the University spin-off today, was surprised by the result. “The reduction in costs and the increase in speed compared to a traditional translation were amazing.” Translating the John Grisham bestseller in the pilot project took roughly four hours and cost less than 70 US dollars. “The translation wasn’t perfect. But professio- nal translators confirmed that it fulfilled 90% of a traditional translation”, underlines Prof. Bernstein. “Since the original research study, we have improved the quality to be comparable to that of profes- sional translations.” Asked about his motivation to support a startup, he says: “I knew from my time in the US that some opportunities need to be seized now or never. After careful consideration we decided to offer our service commercially.” Kunendo has big plans for the future: By mid- 2016, the company aims to work with a crowd of 1000 freelancers for at least ten national and international clients. At Technopark in vibrant Zurich West, two former informatics students of the University of Zurich are working on an ambitious project. With their programming language Crowd- Lang, originally developed in the context of research at UZH, Patrick Minder and Damian Schärli connect to what they call the “global brain”. They access the cognitive skills of peo- ple from around the world and combine them with the number crunching capabilities of computer systems, profiting from the fact that the internet reduced the cost of collaborating with anybody anywhere the world to almost zero. The example of a book translation illustrates how the interaction works. First, the software divides the text into single sentences, which are then translated by a translation software. Next, participants who were specifically selected by CrowdLang check and improve the translation; other participants vote on the best version. The program then puts the text together again. Text flow and grammar are checked in a similar combination of machine and human intelligence resulting in a final translation. Kunendo’s technology clearly differentiates itself from other crowd services. The program doesn’t distribute the tasks to anyone, but takes into account the specific skills different people have and the incentives they best res- pond to. This ensures higher quality results. Oec. Dezember 201527

Seitenübersicht