The correct use of collocation has been widely acknowledged as an essential characteristic to distinguish native English speakers from English learners. Through the analysis of CLEC, we can find that English learners often make mistakes on verb-noun collocations. In this paper, we propose a hierarchical language model that can be used to correct verb-noun collocation errors made by English learners. The language model takes the dependencies between words within a sentence into account. It parses sentences into different levels of clauses. The words within the same clause are highly correlated, and the relevance of words in different clauses is weak. The language model is more stable. Moreover, it is more accurate because collocation information is condensed. It can be used to re-rank candidates and generate classifier features. We apply this hierarchical language model to the correction of English verb-noun collocation errors. Compared with the traditional language model, the new model has better performance.