The syntax and morphology of Japanese verbal nouns
The purpose of this dissertation is to argue that verbal nouns (VNs) in Japanese are verbs, contrary to the widely-held view that they are nouns. I will do this by examining the behavior of major VN-constructions and showing that all of their properties are consistent with the VN=V view, while there are some facts unexplainable under the VN=N view.
VNs have the meaning of verbs and can be used as the main predicate of clauses. This fact follows naturally from the VN=V view but cannot be explained by the VN=N view.
VN-sentences have identical syntactic structures as the corresponding regular-verb sentences, just as expected from the VN=V view.
Unlike regular verbs, however, VNs cannot support verbal affixes directly. When a VN is used in the place of a regular verb, therefore, the dummy verb su ‘do’ must be inserted between the VN and the verbal affix. I claim that this is a consequence of the fact that VNs are free morphemes whereas regular verbs are bound morphemes. This position also explains why VNs are accented like nouns rather than regular verbs.
VNs undergo productive [special characters omitted]-nominalization and form VN-nominals, whose surface form gives the impression that VNs can head NPs. It will be shown, however, that VNs can head a VP within the VN-nominal, a behavior explainable only under the VN=V view.
The final chapter examines the “Case-marking light-su construction”. It will be shown that the main predicate of the construction is an Agentive verb su, which assigns an Agent theta-role to the subject and an Accusative Case to the VN-nominal which precedes it. It will be proposed that regarding the VN-nominal as a modifier of su, with which it forms a semantic complex predicate, explains many curious properties of the construction. One of the curious properties, the “argument-raising phenomenon”, can be explained only if the VN-nominal contains a VP in which the arguments of the VN can be assigned. The behavior of this construction, therefore, also supports the VN=V view over the VN=N view.