A light way to collect comparable corpora from the Web