Reference in language generally involves two main ingredients: the grammatical description and a way to link that description and the intended referent. Sometimes, this link is overtly given by gestural or linguistic content. At other times, the intended referent can be so salient that neither element needs to be overtly produced. This paper investigates definite expressions in the nominal domain that make use of these elements in different combinations. Focusing on definite descriptions and demonstrative descriptions, I argue that the denotational difference between these expressions lies in whether the linking argument is present or not. Demonstratives are argued to realize a binary structure that requires both the description and the linker, with crosslinguistic evidence suggesting that this linker is inherently deictic. Definite descriptions, on the other hand, are analyzed as realizing the unary structure that only contains the description and no linker. The main implications of this analysis are as follows. First, the anti-uniqueness inference that is often associated with demonstratives is derived from general pragmatic principles. Second, a general ban against sentence-internal semantic composition of different modalities of languages such as description and deixis is proposed. Finally, demonstratives are highlighted for their unique and universal function of taking deixis and composing it with the rest of the linguistic content.