Collocations: A Challenge in Computer Assisted Language Learning

The correct use of collocations is one of the most difficult tasks that the student faces when learning a second language, such that one of the goals of Computer Assisted Language Learning (CALL) is to develop programs that aim to identify collocation errors in learners’ writings and propose corrections. However, while statistical models currently used by most of these programs still manage to predict, with a reasonable probability, whether a given word combination is a valid collocation in the language in question or not, they fail to suggest corrections. At most, they offer a list of supposedly valid collocations of the base of the erroneous collocation, from which then the learner shall pick one. This is clearly unsatisfactory. We present ongoing work in which we aim to develop algorithms that do better in that they use the sentential context of the erroneous collocation to suggest a correction and in which we assess how crucial the use of Lexical Functions in the sense of the Explanatory Combinatorial Lexicology is in the context of CALL. All our work is tested on a corpus of American English learners of Spanish