Fascinating essay in The New York Times by David Bellos on the limitations of Google Translate. As the article points out, because the service is based on the most probable equivalent phrases as computed by Google's analysis of an astronomically large set of paired sentences taken from human-engineered translations of documents already in the public domain, this favours common language pairings such as English and Spanish rather than less common pairings such as, say, Welsh and Czech where the corpus on the web is thin on the ground.