No. 111/2020

Photo: Humboldt Foundation/Axel Martens More than 80 languages are spoken in Ethiopia. That makes it very difficult and expensive for courts to find experts to transcribe oral witness statements in the various languages. Self-learning computer systems could offer a remedy. Solomon Teferra Abate and his wife, Martha Yifiru Tachbelie, are developing processes for automatically converting spoken language into written text. “Many people in Ethiopia can’t read or write; not least because written Ethiopian is far more difficult than English, for example,” Solomon Abate explains. An app on your smartphone that translates the spoken word into text could be a big help in many respects. The structure of the languages poses a challenge. It is extremely difficult for the computer to recognise the various manifestations of a word because grammatical distinctions are essentially achieved by additions to individual words. Moreover, there are no comprehen- sive linguistic dictionaries for the Ethiopian languages that could be used as a data base. So, the couple employs a trick: Even though Ethiopian languages use a system of symbols that is fundamentally different from our alphabet, a considerable proportion of the basic sounds are identical to those in German or English. “That’s why we train the model with acoustic data that is available for other lan- guages – including German.” Text RALF GRÖTKER DR SOLOMON TEFERRA ABATE and his wife, DR MARTHA YIFIRU TACHBELIE , from Addis Ababa University in Ethiopia are Georg Forster Research Fellows at the University of Bremen. HOW DOES A LANGUAGE APP HELP IN COURT, MR ABATE AND MS TACHBELIE? 11 HUMBOLDT KOSMOS 111/2020