yes, this must be kept in mind all the time.以下是引用 xiaoz 在 2005-11-11 20:52:37 的发言：
Corpora do not provide negative evidence. If the pattern does not exist in a corpus, you cannot say that it does not exist or is "wrong" in a language. If a pattern occurs a few times in a corpus, you can not conclude that it is "correct". The corpus-based approach focus on centrality and typicality based on statistics. Collocations are essentially defined statistically.
Here are the collocation statistics of the words mentioned in the above postings (based on the BNC world, -3-+3, min. frequency=5):
[本贴已被 作者 于 2005年11月11日 21时42分47秒 编辑过]