That's interesting. I'd vaguely assumed the were using one of the big word use databases - I think there's a large one based on phrases used across a lot of newspapers that the natural language people are always keen on.