Numeral classifier constructions in Korean exhibit a correlation between linear order and possible interpretation: while the Case-medial order allows for either a specific or a nonspecific reading, the Case-final order can only be interpreted as a nonspecific indefinite. A structural account is given for the missing specific reading, which makes use of Diesing's (1992) Mapping Hypothesis, and an LF condition prohibiting representations containing unbound traces. In addition, a distinct Topic-Comment structure is proposed for sentences in which the Case marker appears to have been copied. In such cases, one instance of the 'Case' morpheme is argued to be a Topic marker. This is in line with existing proposals that the particles that mark Nominative and Accusative case can also mark discourse or information status.