Many Natural Language Processing (NLP) systems use some form of "idiom recognition" device, that is, a list of certain idiomatic phrases that are best handled by direct lookup. The reason for listing idiomatic phrases vary somewhat with the task of the NLP system. In e.g. part- of-speech tagging, grammatically deviant phrases are of interest. In a task such as machine translation, semantical reasons also come into play.
In the thesis, I review research on cognitive as well as computational aspects of recognition of lexicalised phrases. The structure of Swedish lexicalised phrases is described in detail and a system for making use of lexicalised phrases in NLP systems is presented, based on a dictionary of many thousands of Swedish lexicalised phrases. Special problems treated are for instance semi-productivity and discontinuity of lexicalised phrases.
E-mail this message to a friend
