I don't understand how you're comparing the two.

You think the results were better in 2006, but the corpus was also different.