Research And Implemention Of An English Collocation Error Detection And Correction System

Computer Assisted Language Learning aims to provide better help for language learners all over the world, and the collocation has been one of the difficult points, although started early, but how to extract collocation from large corpus and how to use the computer to help learners to correct collocation errors in the learning process is still a tie-in problem in this field. In traditional education, language learners usually use artificial compiled dictionaries or books to master the use of collocations, but the con-struction of these dictionaries have great demand for manpower and material resources. Therefore, this paper tries to stduy collocation library construction and collocation error correction, and implement a system that language learners can search for common collocations and can correct their collocation errors in the process of writing.At first, this paper investigates the previous study on collocation extraction and how to use Natrual Language Processing technology to correct collocation errors, by comparing some of such systems, pointing out its characteristics and shortcomings, and then from the view of practical leaner corpora, this paper proposes to use dependency grammar to extract collocations by the correspondence between dependency grammar and collocations, finally we get12class of lexical collocations, on collocation error correction, we try to rank the candidate collocation list, and combine various ranking methods, at last we achieve competitive results on six main categories of collocations. Then we establish a large collocation library and implement a collocation error correc-tion system, the system can check the collocation errors in learners’writing and correct them, By comparing with collocation dictionaries and experiments in practical learner compositions, we demonstrate the effectiveness of our collocation extraction method and collocation error correction method.In the experiments, the collocation libriry can cover more than76%of the colloca-tions in dictionaries, and in terms of collocation error detection, on CoNLL grammar checking copurs we get accuracy, recall and F-score of13.82%,24.48%,17.67%, the MRR can reach0.23on verb and noun collocations. Especially, the word embedding method to correct collocation errors is superior to the traditional WordNet, PMI and LRR method.