The University of Washington and its Professor Oren Etzoni released PanImages at the recent Machine Translation Summit in Copenhagen. Instead of searching for photos by text that may not be recognizable, the system recognizes a string of letters in about 300 different languages and looks for them in Google and Flickr. In a recent test of the process, PanImages found 57 times more results than on an average search.
The procedure works by scanning machine readable wiktionaries. As an example, Etzioni said that if you type in the Zulu word for refrigerator (“ifriji’) you would get almost 500,000 results as opposed to 2 on available search engines. While we seem to be over-saturated with images by using English, it is a comfort to know that the Zulus will be able to find appliances with less of an effort now.
Read More | Tech News Watch