The active / passive relation
The lexical machinery discussed above is certainly suitable for expressing the semantic equivalence of passive and active predicates. Such pairs differ according to syntactic category, strict subcategorization (which follows from the first distinction), and selectional restriction. The lexical entry for an active/passive predicate pair appears as:
(1)
Here Mpass signifies the morphological rule which generates passive forms. No semantic derivation rule distinguishes one form from the other. Their semantic representations are identical.
Neither 16 nor 17 as presently formulated solves the problem of redundant mirror image selectional restrictions. Without a passive transformation, it seems that the selectional restrictions for V-ACTIVE and V-PASSIVE, i.e., [NPX___NPy] and [NPy____by NPX] respectively, must both be listed in the lexical entry. However, this need not be the case if a redundancy rule is used to handle mirror-image selectional restrictions. To a certain extent, a passive transformation is one way of stating such a redundancy. It is suspect, however, for the reasons discussed above. We cannot specify the redundancy by stating that a verb which has a selectional restriction [NPX NPy] entails the existence of another verb form which has a selectional restriction [NPy by NPX], since it is not true (cf. 7). Yet the redundancy is statable in the following way:
(2)
This correctly predicts that if there is a passive verb in the lexicon, then there is also an active counterpart. Moreover, 18 can be restated as a redundancy condition:
(3)
Given 3 in the lexicon, one argument for a transformational treatment of passives over a non-transformational account is neutralized. Note that 3 is not a notational variant of a transformational analysis of passives, because it does not admit exceptions (cf. n. 7). Furthermore, it should be clear that Ma and Mpass are in effect the same rule.
If 3 is taken to account for the cognitive synonymy of active passive sentence pairs, then the selectional restrictions given in 3 must be stated in terms of semantic functions rather than syntactic categories (NP’s etc.) That is, the selectional restrictions of 3 should be given as:
(4)
Here X and Y represent the semantic functions of the constituents immediately preceding and following the verb. The values of X and Y are determined by the semantic class of the verb; thus the verb send is marked [Source Theme]. Condition 19 predicts the selectional frame [Theme Source] for the passive form sent. Such specifications must be available because the semantic function of a constituent is not predictable from a syntactic tree configuration—i.e., semantic functions cannot be interpreted in terms of grammatical relations as defined in Chomsky 1965. Consider the distribution of semantic functions in subject position for verbs of motion:
(5)
There is also a subclass of motional verbs including move, push, force, remove, shove, and throw, whose subjects are interpreted as agents but not necessarily as having any other semantic function associated with motional predicates. For this reason, the linear order of semantic functions must be specified as a selectional frame.
Returning briefly to the question of semantic representation in the framework being developed here, it should be noted that every set of cognitively equivalent strings will have one abstract semantic representation. From this the various surface structures will be derivable by a process of lexicalization.16 Thus, given the abstract lexical entry for carry (V) and carried (A), we may lexicalize in one of two directions. If the adjective is chosen, a passive sentence will be derived; if the verb is chosen, an active sentence will result. Clearly there is no reason to assume that lexicalization will define mapping relations between phrase markers—i.e., transformations. (But transformations may be necessary if their output cannot be filtered by the phrase-structure rules of the base, as in the case of questions, imperatives etc.) More explicit formulations are necessary, but their value and feasibility is sufficiently clear.
From the above discussion, it should be apparent that the relationship between active and passive sentences can be handled interpretively and that the proposal presented here for such an account captures other generalizations as well. The remainder of this paper will be devoted to showing how passives may be generated by the phrase-structure rules of the base, and how certain problems for a transformational account disappear under this analysis.
