This paper aims to offer an alternative approach to some of the problems concerning the use of classifiers in Mandarin, which include the rejection of adjective modifications in front of the classifier, the detachment of classifiers from nouns, and the change of interpretation with the insertion of the particle de. Based on Adger (2025), I propose a mereological analysis that replaces merge with subjoin in structure building. The core idea is that syntactic structures are derived by connecting objects with parthood relations: each labelled syntactic object can be subjoined (modified) by at most two distinct objects as its parts, which correspond to the complement and the specifier in a classic minimalist model and are connected with the mother node in different dimensions. This means that a labelled structure is not projected upward from any head, but rather a pre-existing entity with its subparts breaking downward into two disjoint paths. I argue that numerals, classifiers, and nouns are realisations of discrete syntactic objects: an object Q for the function of quantity can take a classifier as its first part (complement) and a numeral as its second part (specifier), and although the classifier gets spelled out at Q, it can optionally have a noun as its own first part. Therefore, the existence of a specific object like Q no longer depends on an underlying lexical category like N, and being in different dimensions of an object can have consequences both in syntax and semantics. The study suggests that the relations among syntactic categories may not be established via a strictly bottom-up mechanism as in projectionist models. The universality of label sequences on the UG level may be based on parthood relations.
