At a descriptive, or at least informal analytical level, it is common to say that sentences like John was attacked (by a dog) are derived by “passivizing” an active sentence like A dog attacked John. A number of things seem to happen when a verb is passivized in a canonical passive construction: an object is “promoted” to subject, the subject is “demoted” to an optional by-phrase or implicit argument, the main verb becomes a participle and a copula is inserted. However, each of these things is independent of the passive. Objects can be “promoted” in causative-anticausative pairs, by-phrases occur in a number of constructions, and verb participles can behave like adjectives. Given this, it seems highly unlikely that “passive” is primitive notion in grammar.
In this talk, I propose that we can make sense of the above properties if we properly divide the workload between syntax and semantics. (i) In the syntax, there can be a “needy” Voice head that wants to introduce a subject, but other, higher heads can select for this needy Voice head, preventing it from getting what it wants. (ii) In the semantics, the (needy) Voice head may introduce an agent role, but the agent role will only survive under two circumstances: either the Voice head got its subject in the syntax (in which case that subject will bear the agent role), or else there has to be some higher head which is capable of existentially closing over the agent role. Dividing things up in this way not only explains a number of facts about canonical passives, but also explains the morphosyntactic properties of other “passive-like” constructions, including -able adjectives and the Icelandic “modal passive -st” construction. The claim is that languages have several ways of dealing with the syntactic properties of the “needy” Voice head, and several ways of dealing with the semantically-introduced agent role.