The analysis of Passives
Within the framework of transformational grammar, it is generally accepted that phrase structure rules alone cannot account for the relationship between active sentences and their passive counterparts; therefore the derivation of passives necessarily involves a transformation henceforth referred to as PASSIVE. This position is based on the assumption that the active-passive relation is structural in nature and therefore best expressed by a transformation. In what follows I would like to suggest that both this assumption and the position built on it are debatable—(1) because they lead to certain fundamental theoretical problems, and (2) because it is possible to account for the active passive relation without a transformation, given different assumptions about the nature of that relation.
In early transformational theory, the active-passive relation was accounted for by an optional transformation which applied to an underlying structure common to both actives and passives. The relation was considered a derivational relationship—i.e., at some level of representation, actives and their corresponding passives were identical. Thus in a Syntactic structures model, a common underlying structure could be mapped onto two distinct surface structures—one passive and the other active. Figure 1 illustrates how PASSIVE accounted for the active-passive relation.
This situation changed with the publication of Katz & Postal 1964 and Chomsky 1965. In both works it was argued that active and passive sentences must be derived from different underlying structures, and that PASSIVE must be an obligatory rule. For example, in the Aspects model, active and passive sentences had the underlying representations of Figures 2 and 3 respectively.
Under this analysis, PASSIVE does not relate the derivations of an active sentence and its corresponding passive. Rather, Figures 4 and 5 apply.
The derivations for actives and passives are distinct; at no level of representation are they identical. (The same point was noted in Katz & Postal, 118, with respect to all relations between sentence types.) In fact, there is no syntactic rule which explicitly expresses the active-passive relation in the syntactic component. This is also true of other more recent proposals for a transformational analysis of passives (see R. Lakoff 1971 for details). Thus an argument that a syntactic rule PASSIVE is desirable in a grammar, because it reflects speakers’ intuitions about the active-passive relation, cannot be supported.
Given that the underlying syntactic representations of actives and passives are not identical, the active-passive relation must be established by stating an equivalence between structures at some level of representation in Figures 4 and 5. Yet no matter what level is chosen to state this equivalence, it will be the statement of equivalence, and not PASSIVE, that will express the active-passive relation. Thus we could state the equivalence in terms of deep structures—following Katz & Postal’s discussion of “similarity” (118–20), assuming that by ∆ is semantically empty. This would result in actives and passives being assigned the same semantic reading, but would still not relate them syntactically. Alternatively, the equivalence could be stated in terms of semantic readings or cognitive (i.e., truth-value) synonymy as noted in Chomsky (1965:22)— generics, quantifiers, and negation aside. In either case, the active—passive relation would be essentially semantic, and certainly not transformational. Thus a rule of interpretation would be required to account for this relation. An explicit formulation of such a rule will be discussed below.
Given the discussion above, it is an open question whether a transformation PASSIVE is necessary in an analysis which relates actives and passives by a rule of interpretation. If we accept the traditional assumption that rules of semantic interpretation operate on grammatical relations such as “subject” and “object”, which are defined on deep-structure tree configurations, then PASSIVE is a necessary rule of grammar for obvious reasons. If, however, we do not make this assumption, then the conclusion does not automatically follow. Instead, we might assume that rules of semantic interpretation operate on semantic functions which are defined independently of tree configurations, and are therefore unaffected by transformational operations which alter tree configurations in the course of a derivation. It should be possible then to determine semantic functions at a level of surface structure or near it. If, in addition, the surface structures of passives are generatable in the base—a possibility which follows from the notion of PASSIVE as a structure-preserving rule (see Emonds 1970), then a rule PASSIVE would be unnecessary.
