A Corpus - based Collocation Assistant for Swedish Text

A collocation is a recurrent combination of words, such as commit a crime, whose meaning is fairly transparent but which can not be changed arbitrarily, even if the rules of grammar are adhered to and the individual meanings of the words are preserved. If a writer tries to use do a crime, the job of a collocation assistant is to suggest a better alternative such as commit a crime just like a spell checker would do for spelling errors. While spell checkers and grammar checkers have become everyday tools, collocation assistants are still in the prototype stage. In the last couple of years, a series of articles have been published on the development of several di erent tools aimed at learners of English. This report describes the design, implementation and evaluation of the rst collocation assistant for written Swedish. Novel features include the possibility to use a Random Indexing vector space model for measuring semantic similarity, and the use of a full dependency parser. Ett korpusbaserat kollokationsverktyg