In this paper, I argue (i) that Japanese has constructions that are almost the exact mirror images of the right-node raising constructions in English, and (ii) that the properties of those constructions, which I refer to as left-node raising constructions, can be captured straightforwardly if and only if the CONTENT values of domain objects, not those of signs, are assumed to be the principal locus of meaning assembly. In the theory proposed, it is claimed that semantic composition (including "quantifier retrieval") takes place not when some signs are syntactically combined to produce a new, larger sign but when some domain objects (which are essentially prosodic constituents) are merged (by the total or partial compaction operation) to produce a new domain object (i.e. a new, larger prosodic constituent).