This paper discusses the function COMP with data from Cantonese. We first point out some possible problems for previous proposals for the elimination of COMP. We next take up the concept of mixed language using Cantonese data. Initial results show that Cantonese is indeed a mixed language. We do, however, propose that the concept of mixed languages should be extended to include degrees or extents to which a language can be regarded as a mixed language. Our analysis in this paper supports a finer-grained categorization of grammatical functions in linguistic theory.