Some lexical items like and are heads but they cannot determine the grammatical category of their projection, whereas some lexical items like what can determine the grammatical category of the whole projection in a certain configuration even though they are not heads. These are referred to as labeling paradoxes. This paper explores the possibility of explaining the two types of labeling paradox by assuming that the label is not a lexical item but a bundle of features. It is proposed that the first type of labeling paradox arises from the fact that the words like and do not have grammatical category features, and the second type of paradox takes place because the internal merge can be an adjunction.