Some problems with passive
Aside from the major problem discussed above, there are two fundamental weaknesses in the transformational analysis of passives. One involves the derivation and interpretation of truncated passives (i.e., without by-phrases); the other, the specification of a lexical feature on a predicate which distinguishes verbs that undergo passivization from those that do not.
Truncated passives can be generated transformationally either by an ellipsis rule which deletes a wh-phrase, or by a rule which obligatorily preposes the underlying object—thus passivizing the verb—into underlying subject position when the latter is lexically and semantically empty (see Emonds for a detailed discussion of this analysis).
The ellipsis analysis derives truncated passives in 1 from the full passives in 2 by means of optional rules which delete the phrases by someone and by something:
The deletion of by someone or by something would not violate the recoverability condition on deletions, since the deleted elements can be considered pronominal representatives of the general categories HUMAN and NON-HUMAN noun (cf. Chomsky 1964:71). The ellipsis analysis predicts that both 1 and 2 are well-formed. Yet it does not account for the more plausible sources of 3a-f, namely 4a—f (3a—c are from Emonds):
Figure 6
Here 3a–f cannot be derived from 4a–f because this would violate the recoverability condition on deletions: the by-phrases in 4 do not contain pronominal representatives of general categories of nouns. Nor is it possible to derive 3a–f from full passives containing by-phrases with a pronoun someone or something, since the full passives, if they are not interpreted as anomalous, receive different interpretations from the truncated versions. Thus 3e is not synonymous with? Jane was elected president of the club by someone. Hence such examples argue against the ellipsis analysis of truncated passives.
Alternatively, truncated passives could be derived from underlying structures like Figure 1 which contain empty subject nodes (henceforth the empty-node analysis). The empty node in Figure 1 is considered to be lexically empty and therefore not subject to semantic interpretation (cf. Emonds 1970). The empty subject triggers obligatory preposing of the object into subject position and passivization of the verb. Given this analysis, the problem of 3a—f for the ellipsis analysis never arises. However, an objection can be made concerning the use of empty nodes to account for deleted elements. The problem may be posed in the following way. Even though we can account for the absence of underlying subjects in truncated passives by employing empty nodes, it is still necessary to use transformations to handle other deletion processes. For example, a rule of Complement Subject Deletion is necessary in order to distinguish the derivations of 5–6:
(5) Harry expects to be nominated.
(6) Harry is expected to be nominated.
Figure 2
If sentences like 5–6 were to be handled entirely in terms of empty nodes, we would have to suppose that the underlying structure of both was as shown in Figure 2.
Sentence 6 would be derived by NP-Preposing Harry twice, once into the subject position of S2 and again into the subject position of S1. The derivation of 5 is somewhat more problematic, since it would involve an unmotivated preposing rule which did not result in passivization of the matrix verb. In any case, such hypothetical analyses are impossible, since 5–6 are not synonymous, and thus should not have the same underlying structure. So it appears that the use of empty nodes to handle deletions is somewhat limited. In the case under discussion, they are motivated mainly by the hypothesis that all sentences have underlying SVO order. In other words, the use of the empty node in the derivation of truncated passives is not motivated empirically but rather is a way of retaining a transformational analysis of passives.
Besides the difficulties involved with the underlying structures of passives, a rule PASSIVE also creates problems with respect to the lexical entries of verbs. Most transformationalists agree that the ability to passivize is a lexical property of verbs—disregarding for the moment the question of derived nominals (but see Chomsky 1970). Yet there is serious disagreement as to what sort of lexical feature marks this ability. The basic problem revolves around the method for excluding verbs like resemble, fit, involve, and weigh from the operation of PASSIVE, even though they take what might loosely be described as NP objects:
(7)
Two solutions have been proposed: one employs strict-subcategorization features (Chomsky 1965); the other, rule features (G. Lakoff 1970).
The strict-subcategorization solution is based on the observation that verbs like those in 7 do not take manner adverbials freely whereas verbs that allow passivization do. In this way the strict-subcategorization feature [+____ NP ADVMANNER] uniquely marks verbs which undergo PASSIVE. The node ADV-MANNER would be directly dominated by VP, in keeping with the constraint that strictsubcategorization rules be limited to strictly local transformations.
Under this analysis, one expansion of the ADV-MANNER node would be the string by- ∆. Yet if ADV-MANNER occurs only once as the rightmost daughter of VP, then the analysis does not account for passives containing manner adverbials other than by- ∆; e.g.,
In fact, the manner-adverbial analysis in its present form predicts that manner adverbials do not occur in passive sentences. As this is clearly false, it would be necessary to allow ADV-MANNER at least two consecutive expansions. This would necessarily entail the rather ad-hoc condition that if both expansions are chosen, then one and only one must be the string by- ∆. The condition rules out the possibility of sentences with more than one lexical manner adverb per VP. Yet it also suggests that the string by-∆should not be considered a manner adverb—to the extent that this string, unlike other manner adverbs, can co-occur with another manner adverb within the same VP. Even though an agentive by-phrase or a manner adverbial does not co-occur with verbs like those in 7, it remains to be demonstrated that both manifest similar syntactic or semantic behavior. As they are not in syntactic complementary distribution, there is no clear syntactic motivation for assuming they are both generated under the same deep-structure node (see also n. 6).
Contrary to the manner-adverbial analysis, some evidence suggests that the wh-phrase cannot be generated only under ADV-MANNER because verbs like know, believe, think, see, consider, and hear can passivize although they cannot take manner adverbials freely (this observation is from G. Lakoff):
As Lakoff notes, in order to handle 9 and still treat by-∆ as an adverbial under VP, it would be necessary to find another node which could expand as the wh-phrase. The problem with this is that the verbs in both 7 and 9 are stative. To make the analysis work, we must distinguish between the two sets of stative verbs in terms of some adverbial which cannot freely co-occur with the former. However, even if this can be accomplished, the analysis will involve the above-mentioned ad-hoc condition on the double expansion of this second adverbial node under VP. In any case, this analysis is undesirable because of its lack of generality.
The rule-feature solution is also undesirable. According to this proposal, the lexical entries of the verbs in 7 will be marked with a rule feature [- PASSIVE]. If PASSIVE applies to a phrase marker which contains one of these verbs, the derivational history of the string will conflict with the feature in the lexical entry, and the derivation will be rejected. The weakness of this solution is that rule features are little more than ad-hoc devices for blocking ungrammatical strings which the transformational component would otherwise generate. The rule-feature analysis is no more revealing than a list of verbs which are exceptions to the passive rule. Neither explains why such verbs are exceptions.
Given the unacceptability of both the rule-feature and strict-subcategorization proposals, there seems to be no non-ad-hoc way for PASSIVE to filter out the ungrammatical strings in 7. Assuming the filtering function of transformations (see Chomsky 1965), PASSIVE, in its various formulations, cannot be considered a totally satisfactory rule because it is not an adequate filter. This is not to suggest, however, that because a rule does not adequately filter out all ungrammatical strings, it is in principle not viable. For example, there is no justification for discarding PASSIVE because it allows the ungrammatical string John was behaved by himself, since the non-well formedness of this string can be accounted for by an independently motivated constraint on pronominalization. Yet when a transformation yields ungrammatical output which cannot be filtered by independently motivated constraints, and when other alternatives which do not include such a transformation are possible, it seems reasonable to work on the assumption that the transformation may be unnecessary.
