From ebiquity.umbc.edu:
Yesterday Google announced a very interesting resource with 175M short, unique text strings that were used to refer to one of 7.6M Wikipedia articles. This should be very useful for research on information extraction from text.