Collocation and Trillocation

In this paper we proposed that the neglected three words collocations (trillocation) should be emphasized in collocation study. From the point of view of colligations, more useful collocations could be covered by adding a third category. For a specific third word, it will help avoid the unnaturalness of a two words collocation. A statistic based automatic trillocation extracting system is proposed and achieves 45% correctness of trillocations. The system first extract two words collocations based on a statistic measure together with grammatical information and then extract the collocations (the third words) of the acquired two words collocations. Human experts also judged the ambiguity and naturalness of the collocations and the correspondent trillocations. The results show that the extracted trillocations could be utilized to help disambiguate word senses and generate natural language output.