37
Passive Author(s): Osvaldo A. Jaeggli Source: Linguistic Inquiry, Vol. 17, No. 4 (Autumn, 1986), pp. 587-622 Published by: The MIT Press Stable URL: http://www.jstor.org/stable/4178510 . Accessed: 10/05/2011 04:55 Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at . http://www.jstor.org/action/showPublisher?publisherCode=mitpress. . Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission. JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact [email protected]. The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry. http://www.jstor.org

Jaeggli 1986 Passive

  • Upload
    ka6ka

  • View
    235

  • Download
    16

Embed Size (px)

DESCRIPTION

passive

Citation preview

PassiveAuthor(s): Osvaldo A. JaeggliSource: Linguistic Inquiry, Vol. 17, No. 4 (Autumn, 1986), pp. 587-622Published by: The MIT PressStable URL: http://www.jstor.org/stable/4178510 .Accessed: 10/05/2011 04:55

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unlessyou have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and youmay use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .http://www.jstor.org/action/showPublisher?publisherCode=mitpress. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printedpage of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range ofcontent in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new formsof scholarship. For more information about JSTOR, please contact [email protected].

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to Linguistic Inquiry.

http://www.jstor.org

Osvaldo A. Jaeggli Passive

One of the most important recent innovations in syntactic theory concerns the shift from language-particular, construction-specific rules to analyses in terms of general principles from interacting modules of grammar. Perhaps the paradigmatic example of this shift involves the analysis of the passive construction. This article provides a fully explicit, modular analysis of this construction.

I present a detailed account of the passive construction, based largely on data from English but also taking into consideration facts from other languages. I examine what the crucial properties of this construction are and how they are manifested in the grammar of English. My approach makes it essential to pay close attention to the interaction of syntactic processes and morphological processes. Under this approach, there is no single rule of Passive. Passive constructions are simply the result of the interaction of certain morphological and syntactic operations. Only these operations have any theoretical validity.

I argue that a passive verb involves the "absorption" of the external 0-role of the verb and that it is prevented from assigning objective Case. I give a restrictive char- acterization of the process of "absorption" that has interesting empirical consequences for issues related to compositional 0-role assignment. I provide an explicit account of the interpretation of by-phrases in passive sentences and passive nominals and point out the differences and similarities between these two constructions. I suggest a constraint, the Affectedness Constraint, to account for certain restrictions found in passive nomi- nals. I argue that so-called implicit arguments are not mapped onto an NP argument position, and I consider the phenomenon of "control" into purpose clauses by implicit arguments.

According to Chomsky (1981, 124), a passive sentence such as (1) exhibits the two crucial properties stated in (2):

(1) The rat was killed by the cat. (2) a. [NP,S] does not receive a 0-role.

b. [NP,VP] does not receive Case within VP.

The argument in [NP,S] position bears the 0-role that the verb kill assigns to [NP,VP]

I would like to thank Adriana Belletti, Hagit Borer, Noam Chomsky, Bernard Comrie, Murvet Eng, Isabelle Haik, Ken Hale, Nina Hyams, Kyle Johnson, Jay Keyser, Ian Roberts, Tom Roeper, Tim Stowell, and two anonymous reviewers for LI for very useful discussion of the contents of this article. They are clearly not responsible for any of the errors that persisted past their comments.

Linguistic Inquiry, Volume 17, Number 4, Fall 1986 587-622 ? 1986 by The Massachusetts Institute of Technology 587

588 OSVALDO A. JAEGGLI

position. That is, the surface subject of a passive sentence corresponds to the "logical" object of the verb. A lexical direct object is not allowed in (1). These two facts are adequately captured by the statements in (2). In order to see what they follow from, we need to consider the theory of lexical structures.

1. Lexical Entries

The lexicon can be considered as a set of lexical entries, each consisting of a distinctive feature matrix and a set of features of various sorts, as in Chomsky (1965, 164). These features specify syntactic and semantic information associated with the lexical item in question. Minimally, we may assume that this information includes syntactic subcate- gorization features as well as 0-roles to be assigned to the arguments of the predicate (compare the "thematic grids" of Stowell (1981)). Subcategorization features include elements strictly within the government domain of the subcategorizing element. How- ever, a predicate may take an argument that does not fall within that domain. This argument is often called the "external argument" of a predicate (Williams (1981)).' As an argument of a predicate, it also bears a 0-role (at least partially) specified in the lexical entry of the predicate. Thus, the list of 0-roles associated with a predicate is not restricted to those 0-roles that are assigned within the government domain of the head.

I will make the further assumption, essentially following Chomsky (1965), that 0- roles and subcategorization features are "linked" wherever possible. Zubizarreta (1985) points out that this type of theory is motivated by the existence of certain lexical op- erations that crucially require this linking to be specified. She argues that this feature of the "classical" theory of subcategorization makes it superior to the view presented in Williams (1981). By "linking" is understood that 0-roles are to be associated with a subcategorization feature. If subcategorization features are taken to be "positional in- dexes" for argument positions, 0-roles can be said to be assigned positional indexes in the lexical entry of a predicate. Thus, the lexical entries of hit and put will be as follows:2

(3) a. hit: Os Od=[ NP]

b. put: Os Od= NP] 01= P NP]

l The external argument is in fact an argument of the entire predicate phrase, not just an argument of the verb that heads the predicate phrase. Evidence for this is given by sentences such as John broke his arm. See Chomsky (1981) and below for discussion.

2 Given the indeterminacy surrounding the nature of the particular thematic role assigned to any particular argument, I will use the following symbols to represent the 0-roles without naming them: Os for the 0-role assigned to the subject of a predicate (for hit, put this would be Agent); Od for the 0-role assigned to the direct object of a predicate, typically Theme; 61 for Locative; etc. No theoretical significance should be attributed to these symbols, just as it is unclear what theoretical significance, if any, should be attached to the labels Agent, Theme, Goal.

PASSIVE 589

Since one argument of each verb does not fall within its domain of subcategorization, its 0-role remains unlinked, since linking to subcategorization features is enforced only wherever possible and it is clearly impossible to enforce linking if there is no subcate- gorized position for that argument. Using Williams's terminology, a 0-role that is unlinked in the lexical entry of a predicate will be called an external 0-role, and all 0-roles that are linked in the lexical entry of a predicate will be called internal 0-roles.3

The linking of 0-roles and subcategorization features is to be interpreted as follows: a 0-role Ox linked to a syntactic position in the lexical entry of a predicate must be assigned to an NP occupying that position (or to an NP in a chain that ends in that position). Thus, the 0-role Od of hit (e.g. Theme) must be assigned either to a referential expression occupying the position [NP,VP] or to a referential expression occupying another position but in a chain that ends in that position. The assignment is made oblig- atory by the 0-Criterion and also (crucially) by the requirement that if a predicate X subcategorizes the position Y, then X 0-marks Y (see Chomsky (1981, 37)). The second requirement is motivated by consideration of cases such as John ate it/there. These sentences are impossible with expletive it/there interpretation, although the lexical items it/there can surely count as expletives in other sentences. That is, John ate it/there cannot mean 'John ate'. As possible expletive elements, the 0-Criterion should not force the assignment of a 0-role to it/there. Nevertheless, the sentence is impossible in this in- terpretation. Hence the need for the requirement that all subcategorized elements be assigned 0-roles. If all subcategorized positions must be 0-marked, these facts are im- mediately accounted for.

Consider next what happens with constituents that are optionally subcategorized by a predicate. If such a constituent contains a referential expression, it will require a 0-role. Thus, optionally subcategorized positions must also be linked to 0-roles listed in the lexical entry of a predicate. These 0-roles themselves, however, are also to be con- sidered optional. That is to say, if an optionally subcategorized position is not present in a deep structure, we will assume that the 0-role that would have been assigned to that position is also missing from the lexical entry of the predicate. In other words, "op- tionality" extends to the 0-role itself. Thus, the verb seem, which optionally subcate- gorizes a goal NP, will be listed as follows:

(4) seem: Od = [ S'] (Og = [ to NP])

Assignment of Og will be subject to the 0-Criterion. This way of encoding optionally subcategorized constituents has as a consequence that linked 0-roles will always be as- signed to an argument position, modulo the 0-Criterion. The consequences of this fact will be pointed out below.

3There may be important regularities concerning these conventions. That is, it may be true that in the unmarked case a particular 0-role, say Theme, is assigned to [NP,VP], that Goal or Source is assigned to the objects of certain prepositions, and so forth. These subregularities can be encoded as lexical redundancy rules in the lexicon.

590 OSVALDO A. JAEGGLI

Given this view of the lexicon, the terms argument, internal argument, and external argument may be defined as follows:

(5) X = NP is an argument of Y iff X is assigned a 0-role listed in the lexical entry of Y by Y or by a projection of Y.

(6) X is an internal argument of Y iff X is an argument of Y and X occupies a position mentioned in the lexical entry of Y.

(7) X is an external argument of Y iff X is an argument of Y and X is not an internal argument of Y.

Now let us return to examining the basic facts of the passive construction.

2. 0-Role Absorption

First let us consider the property of passive stated in (2a): [NP,S] does not receive a 0- role. This property in fact can be observed directly in passive sentences where preposing has not occurred, as in (8):

(8) It was believed that the conclusion was false.

[NP,S] position here is occupied by an expletive element, it; that is, [NP,S] position does not receive a 0-role. Assuming that this property is descriptively correct, the fol- lowing question arises: what mechanism prevents the assignment of the external 0-role of a verb to [NP,S] position in a passive sentence?

The answer involves the role played by the passive suffix -en. Basically, I claim that this suffix functions as the recipient of the external 0-role of the predicate. Once it is assigned to this suffix, it can no longer be assigned to [NP,S] position. Hence, [NP,S] position is not assigned a 0-role in a passive construction. The fact that this position is not assigned a 0-role allows movement into that position. This is the familiar NP- preposing found in passive sentences. (We will see below what forces this movement in certain instances.) This property of the passive suffix has occasionally been referred to as 0-role absorption.

0-role absorption raises two questions. First, why is it the external 0-role of a predi- cate that is absorbed? Second, what does it mean to say that a 0-role is "absorbed"?

The answer to the first question is given by the theory of lexical structure presented above. The external 0-role of a predicate is the only unlinked 0-role in the lexical entry of a predicate. Thus, it is free to be "associated" with any element whatsoever. There is no requirement that it be associated with a particular NP in a particular structural

4 The term argument is used ambiguously in Chomsky (1981). In one sense, it is coextensive with the definition given here. That is, it is understood as a relational or functional notion, like direct-object-of or subject-of. (It might be better to use the term argument-of in this sense.) In the other, it is used in an absolute sense, and it is closer in meaning to 'referential expression' as in the original Pisa lectures. The sense intended in the statement of the 0-Criterion in Chomsky (1981) is clearly the latter.

(i) 0-Criterion (Chomsky (1981, 36)) Each argument bears one and only one 0-role, and each 0-role is assigned to one and only one argument.

PASSIVE 591

configuration. Since this is, the only 0-role that has this freedom, this is the only 0-role that can be absorbed by a suffix without violating the linking conventions. All other 0- roles are linked in the lexical entry of a predicate. This linking is satisfied only if that 0-role is assigned to the NP position that the 0-role is linked to. Absorbing it would not satisfy the linking requirement. Thus, internal 0-roles are unavailable for "absorption." Pursuing this issue further, we may ask, Why should the external 0-role be associated with anything at all? That is, why is sentence (9a) never interpreted as a paraphrase of (9b) (with expletive it):

(9) a. It kills the rat. b. The rat was killed.

(A similar question applies to There kills the rat, but other issues bearing on the distri- bution of there may be involved in this case.) Such impossible interpretations are in effect excluded by a subpart of the 0-Criterion that makes 0-role assignment obligatory. For the sake of concreteness, we can state such a principle as follows:

(10) All 0-roles must be uniquely assigned.

We will say that in (9b) the external 0-role has been assigned to the passive suffix. Since the verb lacks this suffix in (9a), the external 0-role must be assigned to [NP,S] position, barring an expletive interpretation for it. Another consequence of (10) is that the 0-role assigned to the passive suffix cannot also be assigned by the predicate to [NP,S] position because 0-roles must be assigned uniquely; that is, they cannot be assigned by a predicate to more than one element.

The answer to the second question (What does it mean to say that a 0-role is "ab- sorbed" by a suffix ?) involves explicating the notion of "absorption." This notion has been left rather vague in recent studies. It plays an important role in the analysis of several intriguing phenomena, such as the behavior of clitics, restructuring, and re- analysis. (I exclude from consideration here the use of the term absorption in connection with "complex" quantifiers, as in Higginbotham and May (1981).) It would be desirable to reduce the meaning of absorption to that of other independently needed grammatical mechanisms. In particular, I claim that it is identical to what is typically called feature assignment-in other words, that the passive suffix "absorbs" the external 0-role of a predicate simply by being assigned that 0-role. Nothing more is involved. In order for this simplification to be possible, however, we must explain how it is that a 0-role can be assigned to a suffix. We have already seen that this is possible only with the external 0-role of a predicate, since this is the only unlinked 0-role. But why should we allow this 0-role to be assigned to a bound morpheme? 0-roles are typically assigned only to NPs (and perhaps some PPs). Let us assume, however, that this is uniquely a function of the linking of 0-roles with subcategorization frames in the lexicon, as discussed earlier. Then nothing prevents an unlinked 0-role from being assigned to a suffix. Following Borer (1984), I assume that this assignment process is allowed to occur as part of the mapping from D-Structure to S-Structure (henceforth, "in the syntax") if it does not

592 OSVALDO A. JAEGGLI

violate the Projection Principle. Given this possibility, let us define "absorption" simply as assignment to a bound morpheme. This entails that features may be assigned to affixes in the syntax as long as principles such as the Projection Principle are satisfied. If the lexical specifications of a lexical item are not satisfied by a particular rule, that rule is not allowed to apply in the syntax.

This view of absorption entails that if an element X is to absorb a feature from an element Y, X must be within the assignment domain for that feature of Y. In the cases that concern us this domain is determined by the notion of "government." The suffix -en can absorb the external 0-role of a verb because it is governed by the verb. This view of absorption, then, restricts the scope of this process to instances of government in this case. This view also entails that certain affixes, in this case the passive -en, may have what I will call syntactic visibility; that is, they are visible to other syntactic pro- cesses. A passive participle, then, has the following syntactic structure:

(11) [+V,-IN]

t+V, -N] en I

~~~Os

kill Od= [ NP]

Below I will give several arguments to show that indeed this is the case.' I have presented a theory of lexical structure and absorption that allows property

(2a) to be met in passive structures. Next the fact that this property is obligatorily met (in other words, the fact that 0-role absorption is obligatory) must be accounted for. I see no nonstipulative way of doing this. I will simply assume that the passive suffix -en requires the assignment of the external 0-role. This is its defining morphological/syntactic property. In fact, it may be said more generally that this is the defining characteristic of the passive construction. In English this property may be stated in the lexicon as an idiosyncratic property of the passive morpheme. As far as I know, no other theory has a nonstipulative account of this fact. In a sense, the issue is terminological. If "passive" is crucially defined as a process that involves the absorption of the external 0-role, then the passive morpheme is simply that morpheme which has as its defining characteristic this particular requirement.6

5 See also Fabb (1984) for relevant discussion of syntactic affixes and the structure of passive participles. 6 "Active" past participles in English are also suffixed with -en, but in these cases the affix does not

absorb the external 0-role of the verbal base. An anonymous reviewer points out that the fact that active past participles of transitive verbs are capable of assigning Case might then be a corollary of the lack of 0-role absorption. See section 3 for a discussion of the Case properties of passive participles. The reviewer further suggests that passive participle agreement in some languages may also be derived from the fact that the passive suffix absorbs a 0-role and Case. In many languages the participle carries agreement features in the passive construction, but in general not elsewhere. Suppose that an element must be N-like to carry Case and a 0- role. Then we might say that the passive suffix in these languages can be considered N-like only if it receives certain features for number and gender (N-like features) under agreement. The suggestion seems plausible and is quite compatible with the analysis presented in the text.

PASSIVE 593

The stipulation mentioned in the preceding paragraph is empirically motivated by the fact that passive morphology can appear (in English and many other languages) only with verbs that assign (or participate in assigning) an external 0-role.7 Thus, a raising verb, such as seem, does not passivize (*John was seemed to have left). This fact extends to languages that passivize intransitives, like German or Dutch. (See sentences (22), (23) for examples of passives of intransitives.) In such languages only those intransitives that assign an external 0-role are allowed to passivize. Ergatives-that is, intransitives that assign only an internal 0-role-fail to passivize. Consider the following Dutch examples, from Perlmutter (1978), which illustrate this point (similar examples can be constructed in German):

(12) a. In dit weeshuis groien de kinderen erg snel. in this orphanage grow the children very fast

b. *In dit weeshuis wordt er door de kinderen erg snel gegroeid. in this orphanage is it by the children very fast grown

3. Case

Let us now consider the second crucial characteristic of passive mentioned in (2b): [NP,VP] does not receive Case within VP. In fact, as stated this property is true only of English. Consideration of the facts in Spanish and/or Italian makes it immediately clear that this property should be stated in a slightly different way. Consider the following sentences:

(13) Le fue entregado un libro a Maria por Pedro. to + her was handed a book to Maria by Pedro 'A book was handed to Maria by Pedro.'

(14) Ne furono riconosciuti molti. of+ them were recognized many 'Many of them were recognized.'

The sentences contain a lexical NP following the passive participle. These NPs are arguably in [NP,VP] position. In Spanish the NP un libro 'a book' precedes the indirect object and the passive by-phrase, neither of which is particularly heavy. It cannot have been inverted into that position from [NP,S] position. Subjects cannot normally be in- verted in between a verb, an indirect object, and a prepositional phrase. Inversion from subject position places the inverted subject in VP-final position. Complements that are internal to the VP can appear to the right of inverted subjects only if they are heavy or focused. Since this is not the case in (13), an inversion analysis for the subject is untenable

7 Nonpassivizable idioms, such as *The bucket was kicked by Bill, can be described as expressions that do not permit absorption of their external 0-roles. It becomes unnecessary to mention movement, a welcome result insofar as movement in fact can be shown to be irrelevant. Consider in this respect John seems t to have kicked the bucket, Who kicked the bucket, and the faire-par construction in French (and other Romance languages), which does not allow nonpassivizable idioms but which does not involve movement, as in *Jean a fait casser la croute par Pierre (see Kayne (1975)).

594 OSVALDO A. JAEGGLI

here. In the Italian sentence ne cliticization provides direct evidence that the NP molti is in [NP,VP] position. Ne cliticization is not possible from the inverted subject position (see Burzio (1981) and Belletti and Rizzi (1981)).

These NPs, however, do receive Case (otherwise, the Case Filter would rule (13), (14) ungrammatical). In fact, they are assigned nominative Case. This is possible because, as is well known, in these languages an NP inside the VP can be assigned nominative Case. Consider the following active sentences:

(15) Llego Juan. arrived Juan 'Juan arrived.'

(16) Le gustan las manzanas a Juan. to + him please the apples to Juan 'Juan likes apples.'

(17) Ne sono arrivati molti. of+ them are arrived many 'Many of them have arrived.'

In Spanish, animate specific NPs in [NP,VP] position give direct evidence that the Case assigned to this position in passive sentences is not objective. In transitive structures, animate specific NPs that are assigned objective Case are always preceded by the marker a:

(18) Vimos *(a) Juan. we + saw

This marker is obligatorily absent in passive structures:

(19) En la fiesta fue presentada (*a) Maria por su padre. at the party was introduced Maria by her father 'At the party Maria was introduced by her father.'

Given these facts, (2b) should be restated to read: [NP,VP] does not receive a verbal Case. Objective Case is a verbal Case, nominative Case is not. Why is this position not assigned a verbal Case? Before considering this question, I will examine the most im- portant effect of verbal Case absorption in the passive construction: NP-preposing.

In English, verbal Case absorption has the same effect as Case absorption in general, since English does not have the option of assigning nominative Case within the VP. In the D-Structure representation (20), underlying John was killed, the subject position is assigned no 0-role, and the object is assigned no Case:

(20) [e] was killed John

If killed cannot assign Case to John in this structure, the only way for (20) to surface as a grammatical S-Structure construction is for the direct object NP, John, to be moved into subject position where it can be assigned nominative Case. This movement is pos-

PASSIVE 595

sible since [NP,S] position is not a 0-position, the external 0-role of kill having been absorbed by the passive suffix -en. NP-preposing, then, is forced by Case theory. The Romance examples mentioned above confirm this point. If a Case feature can be assigned to the postparticipial position, movement is not forced (although it is still possible). Assuming that clausal complements do not require Case in English, this theory also correctly predicts that movement is not required in sentences such as (21), as pointed out in Chomsky (1981):

(21) It was believed/held/reasoned/ . .. that the conclusion was false.

Case theory, then, forces movement in precisely those instances where movement is obligatory, while allowing optionality in those instances where the facts require such optionality.

Case absorption can be interpreted as analogous to 0-role absorption. We might assume that the passive suffix -en is assigned (and requires the assignment of) objective Case. Once this Case is assigned to the suffix, it cannot be assigned further to an NP in object position.

This interpretation of Case absorption, however, predicts that only verbs that assign objective Case may passivize. Under the standard assumption that intransitive verbs do not assign Case, then, this theory predicts that intransitive verbs should systematically fail to passivize. Although this is true in English, it is certainly not true universally. As mentioned above, some languages passivize certain intransitive verbs. Consider the fol- lowing German and Dutch examples:

(22) a. Es wurde getanzt. it was danced 'There was dancing.'

b. Es wurde bis spat in die Nacht getrunken. it was till late in the night drunk 'Drinking went on till late at night.'

(23) Er wordt gefloten. it was whistled 'There was whistling.'

Under the assumption that German and Dutch intransitive verbs do not assign objective Case, it cannot be the case that the passive suffix in these languages also requires the assignment of objective Case, since this requirement would not be met with the verbs in (22) and (23). Such facts appear to indicate that in some languages the passive suffix does not need to absorb verbal Case.

However, a problem immediately arises. The data mentioned above appear to in- dicate that in languages such as German and Dutch the passive suffix does not need to absorb objective Case. However, even in these languages, in those instances where the passive suffix can absorb objective Case, it must do so obligatorily. Thus, the following German sentence is ungrammatical:

IBM
Hervorheben

596 OSVALDO A. JAEGGLI

(24) *Es wird diesen Roman von vielen Studenten gelesen. it is this-OBJ novel-OBJ by many students read

If we conclude, based on the sentences in (22), that the passive suffix in German does not need to absorb objective Case, then (24) should be an available option. In order to rule this sentence out, we would have to claim that accusative Case absorption is oblig- atory in German whenever possible. This seems odd, at best.

There is an alternative analysis that does not run into this problem and might there- fore be considered preferable. Instead of assuming Case absorption-that is, Case as- signment to the passive suffix-we might interpret the apparent intransitivity of passive participles as an effect of categorial neutralization brought about by the passive suffix- ation rule. Rouveret and Vergnaud (1980, 121) argue that passive participles are not [+ V, - N] elements. Rather, they claim that these elements belong to a neutralized category [ + V]. This categorial change could be considered an effect of the morphological rule that affixes the passive suffix. Then we can say that this neutralized category cannot assign accusative Case. This effect would follow from some (presumably) universal prin- ciple of Case theory (at least for unmarked cases). Under this view, no absorption need have occurred. The resulting category is simply unable to assign Case, not because the Case feature has been assigned to something else (the suffix), but rather because the resulting category has lost its capacity to assign Case altogether.

However, the idea that categorial neutralization is responsible for the "intransitive" nature of passives is also not without problems. It forces us to assume that in (25) the NP a book is not assigned Case by the passive participle:

(25) John was given a book by Bill.

Rather, it must have inherent Case, assigned to it in some other way that is not sensitive to the categorial status of the governing verbal element. Although such an account may be viable for English, it does not work so well for other languages. Inherent Case is typically tied to a particular thematic relation. In standard English only the theme ar- gument of a double object verb can be assigned inherent Case in a passive structure as in (25). Note the ungrammaticality of A book was given John in most dialects of English (although some dialects do accept it, especially if the goal argument is pronominal, as in ?The job was offered me last week). In other languages, however, passive participles appear to be capable of assigning Case to NPs that are not necessarily thematically restricted in any way. Consider the following data from Kinyarwanda, cited in Siewierska (1984, 60, 62):

(26) a. Ibaruwa yohererejwe Maria na Yohani. letter sent + PASS Maria by Yohani 'The letter was sent (to) Maria by Yohani.'

b. Maria yohererejwe ibaruwa na Yohani. Maria sent + PASS letter by Yohani 'Maria was sent the letter by Yohani.'

IBM
Hervorheben

PASSIVE 597

c. Ikaramu yandikishijwe ibaruwa na Yohani. pen write + with + PASS letter by Yohani 'The pen was written a letter with by Yohani.'

The passive verb in all of these sentences takes a lexical complement: in (26a) the goal argument, in (26b) the theme argument, and in (26c) a subject that corresponds to the instrumental argument. Thus, a passive verb in Kinyarwanda has the ability to take a lexical complement without paying particular attention to its thematic status. This sug- gests that these NPs are assigned structural rather than inherent Case, under the as- sumption mentioned above that inherent Case is tied closely to 0-roles, whereas struc- tural Case is completely free. Siewierska (1984, 61-62) demonstrates that the NP ibaruwa in (26b) cliticizes and relativizes like a direct object, providing additional evidence that it is in fact a direct object. If these processes are sensitive to verbal Case, as cliticization clearly seems to be, then we must assume that these NPs are assigned structural Case. (The reader is referred to Siewierska (1984) for additional evidence that passive parti- ciples must be considered to be transitive in many other languages.)

Scandinavian languages also allow passives like (26a,b):8

(27) a. Jens ble gitt en bok. Jens was given a book 'Jens was given a book.'

b. En bok ble gitt Jens. a book was given Jens 'A book was given Jens.'

Once again, the lexical complement of the passive participle is not thematically restricted. This makes an inherent Case proposal appear doubtful. Instead, it is more natural to assume that here too the passive participle is still allowed to assign structural Case in these instances.

The Kinyarwanda and Scandinavian facts (and many of the other cases illustrated in detail in Siewierska (1984)) call into question the assumption that passive participles are never capable of assigning structural Case. These facts cast serious doubts on the cross-linguistic validity of the neutralization proposal, since such a proposal entails that passive verbs should never assign structural Case. Consider, then, the following alter- native account. Suppose first that passive morphology does indeed absorb Case, in the sense that it requires that structural Case be assigned to it. Suppose next that in lan- guages that display the range of facts found in Kinyarwanda or Scandinavian, the relevant verbs are capable of assigning two structural Cases. One of these Cases is assigned to the passive morpheme; the other one is free to be assigned to a lexical complement of the verb. Since the Case in question is structural, it is not tied to a particular thematic

8 This fact was pointed out by an anonymous reviewer, who suggests that passives should retain their ability to assign Case in some instances. The analysis developed in the text follows some of the suggestions of this reviewer.

598 OSVALDO A. JAEGGLI

relation. The thematic status of the lexicai complement is irrelevant. The passive mor- pheme in these languages, then, absorbs one of the (n) structural Cases assigned by the bare Verb. This will account for the facts presented above.

The German and Dutch facts can be made consistent with this analysis under similar assumptions. Note, however, that we must assume that in these languages intransitive verbs also assign structural Case. This will allow for (22) and (23). Such verbs would still be prevented from taking an internal argument by the 0-Criterion, since they assign only an external 0-role. To account for the ungrammaticality of (24), we assume that in these languages verbs uniformly assign one and only one structural Case. If this Case is assigned to the passive morpheme, it can no longer be assigned to the direct object. If it is assigned to the direct object, it can no longer be assigned to the passive morpheme. In either instance a violation results.

Finally, English would differ from Kinyarwanda, Scandinavian, German, and Dutch as follows. In English, as in German and Dutch, verbs uniformly assign one and only one structural Case. If the passive morpheme also absorbs structural Case in English, then the NP a book in (25) is assigned inherent Case. As pointed out above, this seems reasonable, since such NPs are thematically restricted. English differs from German and Dutch in that intransitive verbs in English are not structural Case assigners. This would then rule out passives of all intransitives in English, as desired.9

To summarize, I have claimed that the crucial properties of the passive construction are to be accounted for in the following way. [NP,S] position is not assigned a 0-role because the external 0-role of the bare verb is absorbed by the passive morpheme. This is the defining characteristic of passive. The Case properties of passives have been analyzed in an analogous way as the result of Case absorption.10 Case absorption may be parametrized both with regard to the number of structural Cases assigned by a bare

9 Roberts (1985) suggests a different alternative worth considering. He points out that in German and Dutch it is possible to assign nominative Case VP-internally in certain instances. This raises the possibility that the passive morpheme in these languages may be assigned nominative Case, thus satisfying its Case requirement. Such an account would not require the assumption that intransitive verbs like those in (23) and (24) assign structural Case in German and Dutch. Although such an alternative may be viable for these lan- guages, it is not viable for Scandinavian or Kinyarwanda, where there is no evidence that nominative Case is assigned VP-internally. Thus, for those languages we must opt for the analysis in the text.

10 This view of the Case properties of passive participles also has important consequences for the analysis of pseudopassives, as in (i):

(i) This bed was slept in by George Washington. This sentence must also involve no assignment of Case for an NP in VP, in this instance the object of the preposition in. The analysis proposed here forces us to assume that these instances of Case assignment also originate from the verb. If the preposition assigned Case independently of the verb, the analysis would predict that (ii) should be grammatical, whereas clearly it is not:

(ii) *There was slept in this bed by George Washington. Let us assume, then, that in these instances the object of the preposition is in fact assigned structural Case (indirectly) by the verb. Once the verb loses its capacity to assign structural Case via affixation of -en, it cannot assign Case to the object of in.

These assumptions interact in an interesting way with the Uniformity Condition on Case Assignment of Chomsky (1986): if x is an inherent Case assigner, then x Case-marks p iff x 0-marks the p-chain. Together, the Uniformity Condition and the analysis of pseudopassives sketched above predict that in (ii) the NP in [NP,S] position is not 0-marked by the preposition of which it is a D-Structure object. This is the desired

IBM
Hervorheben
IBM
Hervorheben
IBM
Hervorheben

PASSIVE 599

verb and with regard to the types of verbs that assign structural Case in a language. I assume that the unmarked situation is for a verb to assign one and only one structural Case to an internal argument. But there appear to be languages where a verb may assign more than one structural Case (for example, Scandinavian and Kinyarwanda), as well as languages where verbs that have no internal arguments assign structural Case (for example, German and Dutch). Thus, the particular properties of the Case behavior of passive verbs admit of several language-particular options.

4. The Interpretation of By-Phrases

The analysis presented so far gives an account of the two properties of passive con- structions stated in (2). It does not, however, provide an account of a third property of passive constructions. Consider the following very familiar fact: the NP in a passive by- phrase is interpreted as bearing the external 0-role of the passivized predicate. This is true regardless of the particular nature of that 0-role. It is incorrect to claim that an NP in a passive by-phrase is always interpreted as an Agent. In fact, it is interpreted as an Agent only when the external 0-role of the passivized predicate is Agent. When the external 0-role is Source,1' the NP in the by-phrase is interpreted as a Source. When the external 0-role is Goal, the NP in the by-phrase is interpreted as Goal. When the external 0-role is Experiencer, the NP in the by-phrase is interpreted as Experiencer. For example:

(28) a. Bill was killed by Mary. (Agent) b. The package was sent by John. (Source) c. The letter was received by Bill. (Goal) d. That professor is feared by all students. (Experiencer)

This point, which has been noted (in slightly different ways) since the first studies in generative grammar, must receive an account within any theory of the passive construc- tion. A related fact to be accounted for is that passive by-phrases are restricted to passive sentences. Thus, active sentences may contain by-phrases only if such PPs are inter- preted as locatives or instrumentals. For example:

(29) a. John is killing Mary by Bill. b. John is traveling through Europe by train-

Below I suggest a unified analysis for these two facts.

empirical result. The NP the bed in (ii) must be interpreted as theme (or perhaps patient), not as a locative. Compare (iii), where England can hardly be interpreted as a themelpatient:

(iii) *England was slept in by George Washington. Thus, this analysis accounts for a peculiar property of pseudopassives: the NP in [NP,S] position in a pseudopassive is never interpreted as a pure thematic dependent of the preposition of which it is an object at D-Structure.

" Source and Goal external arguments are often interpreted as Agent Source and Agent Goal. These are not instances of two 0-roles assigned to an argument. Rather, they might be interpreted as composite 0-roles. I have nothing new to say about this. For relevant discussion, see Chomsky (1981, 139, fn. 14).

600 OSVALDO A. JAEGGLI

The basic idea I wish to pursue is quite simple. My claim is that the passive suffix -en is crucially involved in transferring the external 0-role onto the NP in the by-phrase in a passive sentence. It acts as a 0-role assigner to that NP. If it fails to transfer the external 0-role, the NP in the passive by-phrase fails to meet the 0-Criterion. The prep- osition by on its own is only capable of assigning locative or instrumental 0-roles. This will yield a grammatical surface structure for (30),

(30) John was killed by Bill.

but the interpretation will be 'Someone killed John in the vicinity of/next to Bill'. Cru- cially, Bill is interpreted as the agent of the killing only if the external 0-role of kill is transferred onto it by the passive suffix. Similarly, (29a) is ungrammatical with an agen- tive interpretation for Bill. Since the sentence does not contain a passive suffix, the external 0-role of kill cannot be assigned to Bill. The only possible reading will be one on which Bill is interpreted as a locative. These are the desired results.

Let us see how this basic idea might be implemented. I begin by considering the structure of a passive participle. Following Lieber (1980; 1983) and the previous analysis of 0-role absorption, I assume that passive participles have the following structure:

(31) [+V,-N]

[+V,-N] ed I Os

kill Od

I will interpret 0-role transmission simply as 0-role assignment from the passive suffix to the by-phrase. This is possible since, as in Lieber (1983), I assume that the argument structures of both the verbal head and the passive suffix percolate to the branching node dominating them. The external 0-role is then assigned to the passive by-phrase. Assuming that it is assigned to the PP, it percolates to the head of the PP, the preposition by, and from there is assigned to the object of by.

It follows from this analysis of 0-role assignment to by-phrases that an NP in a by- phrase is assigned the "pure" external 0-role of the passive predicate. An NP in [NP,S] position in an active sentence is assigned a compositional 0-role, which includes, but is not identical with, the "pure" external 0-role of the head of the predicate (see Chomsky (1981, 104-105)). The compositional nature of the 0-role assigned to [NP,S] position in active sentences is best observed in constructions like John lost his way or John raised his hand. John is an Agent of the action expressed by the verb in these sentences, but he is also the patient of (or affected by) the action expressed by the verb. Now note that this interpretation is lost in the corresponding passive sentences: His way was lost by John, His hand was raised by John. These facts are predicted by the proposed an- alysis, given the difference between a compositional 0-role assigned to [NP,S] position

PASSIVE 601

in active sentences and the "pure" external 0-role assigned to the NP in by-phrases in passives. 12

At this point we might wonder what forces this assignment process. In particular, why is (32) totally impossible, with expletive itlthere, meaning 'The man was killed'?

(32) *The man was killed by it/there.

Intuitively this impossibility is felt to be related to the impossibility of having an expletive in a subcategorized position. The answer I will propose involves the subcategorization properties of the passive suffix. My claim is that the passive suffix optionally subca- tegorizes for a [by NP] prepositional phrase. This subcategorization property of the passive suffix is listed in its lexical entry. If the by-phrase is subcategorized by the passive suffix, the passive suffix will have to assign the by-phrase a 0-role, given the condition that if X subcategorizes for Y, then X assigns a 0-role to Y. Thus, if the passive by- phrase is present, it must receive a 0-role from the passive suffix. 0-role assignment in this instance, then, is forced by the usual mechanisms that force 0-role assignment to subcategorized complements. Nothing special needs to be added.

Consider next what happens when the by-phrase is missing, as in The man was killed. Must we say that in such instances the 0-role is not present at all? This is how the proposed analysis treats other optionally subcategorized complements. But in the case of a missing by-phrase this would not be a desirable consequence. For there is considerable evidence that even when the by-phrase is missing, the external 0-role of a predicate is present in a passive construction. I will review this evidence shortly, when I discuss so-called implicit arguments. For the time being, let us simply assume that the correct representation of a passive sentence without a by-phrase is as in (33a), and not as in (33b) (where the external 0-role is not represented at all):

(33) a. [The man] was [kill + ed] t Os Od

b. [The man] was [kill+ ed] t Od

Why should the external 0-role be allowed to stay on the passive suffix? Recall that in discussing optionally subcategorized complements, we assumed that optionality extends to the 0-role as well. This would entail the representation in (33b). But this is not the case here.

The crucial difference between passive by-phrases and other optionally subcate- gorized constituents concerns the linking conventions in the lexicon. Other optionally subcategorized constituents are typically linked to the 0-roles to be assigned to them in the lexical entry of the predicate that optionally subcategorizes for them. Such 0-roles are listed in the lexical entries of those predicates because they are lexical properties

12 Compare in this respect the analysis in Marantz (1984). There it is assumed that the NP in the by-phrase receives the 0-role assigned by VP to [NP,S] position in an active sentence. Such an analysis would not give an account of the facts mentioned in the text.

602 OSVALDO A. JAEGGLI

of those predicates. But the passive suffix -en is not listed in the lexicon as having a 0- role to assign. It receives such a 0-role only after it has been suffixed to a verbal stem. Since it does not by its own right have such a 0-role, its subcategorized complement, the passive by-phrase, is not linked to a 0-role in its lexical entry. Hence, it can have an optionally subcategorized complement without this optionality also extending to the 0-role that it ultimately would assign to this complement. Simply put, since it does not have a 0-role in its lexical entry, the optionality of the subcategorized complement cannot extend to the 0-role. This is what allows a representation like (33a). A representation like (33b) is ruled out by the stipulation discussed above that the passive suffix must be assigned the external 0-role of the predicate to which it is attached.

The idea that by-phrases are optionally subcategorized complements of the passive suffix has two interesting empirical consequences. The first concerns variation among languages concerning the availability of by-phrases; the second concerns the "argument" status of a by-phrase.

Some languages do not appear to allow by-phrases. Comrie (1977) points out that by-phrases are not allowed at all in Latvian and in the more traditional styles of Persian and Classical Arabic. Siewierska (1984, 35) gives the following list of languages where "passive clauses may not appear with an overt agent": Latvian, Urdu, Kupia, Classical Arabic, Amharic, Igbo, Tera, Sonrai, Fijian, Atjnjamathanha, Cupenlo, Cora, Huichol, Cahuilla, Shoshoni, and Pepecano. The analysis I have proposed offers a very straight- forward way of describing this situation across languages. All that needs to be said is that in these languages the passive suffix does not subcategorize for a by-phrase. The presence of a by-phrase will then always result in a violation of subcategorization and the 0-Criterion. The NP in the by-phrase will fail to meet the 0-Criterion, since no 0- role transmission would occur.

Yet in other languages it appears that passive clauses must always contain an agent phrase. Siewierska (1984, 35-39) states that in the Dravidian language Kota (Subbiah (1972)) and the Austronesian languages Palauan (Josephs (1975) and Wilson (1972)) and Indonesian (Chung (1976a,b)), there are passive clauses that must always occur with an agent phrase. In these languages, however, the passive verb agrees in number and gender with the NP in the agent phrase. Perhaps the requirement that the verb must agree with this NP forces agent retention. The passive morpheme, then, might be assumed to ob- ligatorily require the agent phrase in order to satisfy its agreement properties, just as Agr might be supposed to require the presence of a subject NP in English or Romance. This would account for the obligatoriness of the agent phrase.'3

'3 Even in English, where by-phrases are generally completely optional, it appears to be the case that in certain structures an agent phrase is obligatory. Consider the following examples from Mihailovic (1966, 123- 124):

(i) a. On his deathbed he was succeeded by his daughter. b. *On his deathbed he was succeeded.

(ii) a. His first insult was followed by an even worse one. b. *His first insult/one was followed.

PASSIVE 603

Some confirmation for the theory presented above comes from an impersonal pas- sive construction in Polish. This construction does not allow by-phrases."4 Consider the following sentences from Comrie (1977):

(34) a. Zapukano w drzwi (*przez s4siada). was-knocked at door by neighbor

b. Idzie sig szybko (*przez uczniow). was-walked quickly by schoolboys

c. Zabito Jana (*przez Marysie). was-killed Jan by Mary

This construction is formed with a "special impersonal form, distinct from all other forms with a subject" (Comrie (1977, 49)). The theory defended here has a straightfor- ward way of describing the impossibility of by-phrases in these constructions: it must be a property of this special form that it does not subcategorize for a by-phrase.

Next, what is the argument status of NPs in by-phrases? Given the definition of argument in (5), a by-phrase is not an argument of the verb. The NP in question does not receive its 0-role from the predicate or from a projection of the predicate; rather, it

(iii) a. He was brought up by his parents. b. *He was brought up.

(iv) a. The part of the mother was played by Miss Perkinson. b. *The part of the mother was played.

However, although (iiib) and (ivb) are unacceptable as given, they become acceptable if a PP or adverbial is added to the VP:

(v) a. He was brought up in Cambridge. b. The part of the mother was played well.

The verb follow may be passivized without an agent phrase if its subject is animate: (vi) The Russian spy was surely followed.

These facts remain a mystery within the analysis suggested in the text. See Siewierska (1984) for further discussion.

14 A by-phrase is often not allowed in impersonal passive constructions. Siewierska (1984, 100) reports that "crosslinguistic studies reveal that agentless impersonal passives tend to predominate" and lists the following languages as having impersonal passive constructions that cannot take an overt agent: Ute, Nez- Perce, Mojave, Kannada, Kolami, Bengali, Arabic, Turkish, Maasai, Spanish, Italian. This tendency may be related to several properties of the particular way in which impersonal passives are formed. Within the analysis adopted here, we might claim that the passive morpheme cannot subcategorize for a complement by-phrase in case it fails to absorb Case, as happens in an impersonal passive construction. This gives a straightforward description of the inability to take an overt agent. In turn, the failure of subcategorization may be related to the fact that if the passive morpheme has no Case, it cannot assign a 0-role to the agentive PP, since the passive morpheme itself will not be assigned the external 0-role of the predicate, merely preventing the assignment of that 0-role by the VP.

At any rate, this situation is merely a tendency. There are languages with impersonal passive constructions that do allow an agent phrase, among them German and Dutch. Consider the following examples. ((i)-(ii) are from Comrie (1977), (iii) is from Kirsner (1976)):

(i) Es wurde gestern von uns getanzt. it was yesterday by us danced

(ii) Es wurde dem Schuler vom Lehrer geholfen. it was to + the student by + the teacher helped

(iii) Er wordt door de jongens gefloten. it was by the boys whistled

604 OSVALDO A. JAEGGLI

receives it from the preposition by. If anything, the NP is an argument of by. The whole PP might be considered an argument of -en; but once again, it is not an argument of the verbal predicate. The issue is relevant because Zubizarreta (1985) presents arguments that passive by-phrases should not be considered arguments of the verbal predicate. She points out that Lieber (1983) shows that if an argument-taking lexical item is a member of a compound, either its argument structure is satisfied entirely within the compound or it is satisfied entirely outside the compound. This explains the ungrammaticality of the examples in (35):

(35) a. *candy-handing to children b. *book-putting on the shelves

This generalization does not extend to passive by-phrases:

(36) a. cigarette-smoking by children b. revenue-sharing by local communities

The acceptability of (36a-b) appears to show that by-phrases are not arguments of the head noun. It should be pointed out, however, that the examples in (36) do not involve instances of the passive suffix. The by-phrases in these examples must be interpreted along different lines, as in passive nominals such as the destruction of Rome by the barbarians. 15 (But below I present an analysis of passive nominals that also leads to the conclusion that by-phrases are not arguments of the head noun.) It is not altogether clear that the examples in (36) are relevant for evaluating whether by-phrases in passive sen- tences are arguments of the verb or not.

There are some facts, however, that suggest that NPs in by-phrases do not have the same argument status as other NPs within the VP. Consider the following sentences:

(37) a. His mother likes John. b. His mother talks to John often. c. His mother talked to Sally about John. d. *His mother was killed by John.

Whereas it is quite possible for the possessive pronoun in the NP in subject position to be understood as coreferential with the direct object NP in (37a), the indirect object in (37b), and the object of the preposition about in (37c), it is impossible-or at best highly marginal-for it to be coreferential with the object in a passive by-phrase (hence the * on (37d)). Note that if the by-phrase is interpreted as a locative, coreferentiality is once again possible. This suggests that the lack of coreferentiality in (37d) is directly connected

15 Roberts (1985, chap. 2) points out that -ing compounds based on head nouns that do not allow nominal passives, such as *danger's sense by John, *Bill's fear by Marty, *Mary's respect by John, are also ungram- matical with a by-phrase: *vibration-sensing by psychics, *spider-fearing by arachnophobes, *intellectual- respecting by gullibles. This supports the view that these -ing compounds pattern more closely with nominals than with clauses with respect to their ability to take a by-phrase. Compare Danger was sensed by John, Bill is feared by Marty, Mary was respected by John.

PASSIVE 605

to its thematic status with respect to the predicate. If NPs in by-phrases are not arguments of the predicate, these facts might be accounted for by assuming that only true arguments enter into the theory of argument binding. The lack of argument status for John in (37d) would then explain the failure of coreferentiality.

Before considering certain crucial properties of sentences like (34a), I will digress to consider so-called passive nominals.

5. Passive Nominals

Certain derived nominals exhibit passive-like structures (Chomsky (1970)):

(38) a. the barbarians' destruction of Rome b. Euler's proof of the theorem c. Rembrandt's portrait of Aristotle

(39) a. the destruction of Rome by the barbarians b. the proof of the theorem by Euler c. the portrait of Aristotle by Rembrandt

(40) a. Rome's destruction by the barbarians b. ?the theorem's proof by Euler c. Aristotle's portrait by Rembrandt

Linguistic theory should make precise what relation these structures bear to passive sentences. Although an in-depth investigation of this issue falls beyond the scope of this section, I will consider it with an eye toward ascertaining certain fundamental simi- larities and differences.

I assume that a derived nominal like destruction, proof, portrait is a "predicate" in the sense that it takes arguments to which it assigns 0-roles. These 0-roles are listed in its lexical entry. This is not a property of all nouns. On the contrary, the great majority of nouns do not take arguments to which they assign 0-roles. Undoubtedly, the ones under consideration do because they are derived from verbs. Whatever lexical process derives these nouns from verbs, then, involves the conservation of the thematic prop- erties of the verbal base. Note that this is not a necessary requirement of lexical pro- cesses. I assume with Williams (1981) that lexical rules allow a variety of operations on 0-roles, including deletion, addition, internalization, and externalization.

The conservation of the thematic properties of verbal bases in derived nominals is quite straightforward, with one exception. It is a general property of nominals that they do not require subjects. That is, the external argument of a verb is in general made optional by the deverbalizing process. This is a property of the derived category, Noun (Chomsky (1981, 40)). A nominal like destruction, then, will have the following lexical structure:

(41) destruction: (Os) Od=[ NP]

606 OSVALDO A. JAEGGLI

Movement into subject position of the nominal, as in (42),

(42) Rome's destruction

is possible given the optionality of the external 0-role, and only when the external 0- role is not present. When the external 0-role is part of the entry of the head noun in a particular structure, it will be assigned to the Specifier-of-NP position. Movement of an NP bearing another 0-role into that position will result in a 0-Criterion violation.

This property of nominals is consistent with the following obvious fact: in English, passive nominals do not contain passive suffixes. The lack of a passive suffix makes absorption unavailable to nominals. This means that the structure associated with (42) does not contain Os. This fact will become relevant in the next section when we consider so-called implicit arguments.

Movement is never forced in nominals. This follows straightforwardly given the general availability of Of Insertion to assign Case to the object of a nominal. (See Stowell (1981) for a discussion of this rule.) Nothing special need be said.

The most difficult questions concerning passive nominals involve the interpretation of the by-phrase. Since passive nominals do not involve a passive suffix, 0-role trans- mission cannot be effected as it is in sentences. What mechanism, then, interprets sen- tences like those in (39), (40)? Though I have no fully satisfactory answer to this question, I would like to point out factors that I believe are certain to be involved in any correct solution.

By-phrases in passive nominals exhibit interesting differences from by-phrases in sentences. Whereas by-phrases are fully productive in sentences, this is not the case with nominals. By-phrases in nominals are best when the NPs they contain are interpreted as Agents. Any other 0-role is either marginal or impossible. The examples in (43)-(46) are cited by Jackendoff (1977, 92, 93), who attributes them to Hornstein (1975):

(43) a. The package was received by John. b. ?*the receipt of the package by John

(44) a. Harry was feared by John. b. *the fear of Harry by John

(45) a. Danger was sensed by John. b. *the sense of danger by John

(46) a. Mary was respected by John. b. *the respect for Mary by John

These facts can be accounted for as follows. 0-role transmission is not available in passive nominals since they lack the passive suffix. Acceptable by-phrases are interpreted, then, with one unique thematic relation, Agent, assigned by default (perhaps at LF). As a matter of execution, let us assume that the thematic relation of the NP in the by-phrase (always Agent) is assigned by default to that NP only if it is nondistinct from the external 0-role of the head Noun. In the ungrammatical examples in (44)-(46) this thematic re- lation clashes with the thematic properties of the head nouns fear, sense, respect. This

PASSIVE 607

produces ungrammaticality. The first example, (43b), is slightly better because the verbal base assigns an Agent Goal role to its subject (see footnote 5). This role is more com- patible with the default role Agent. In fact, (45b) is quite acceptable if John is interpreted fully agentively.

Another difference between passive nominals and passive sentences pointed out in Anderson (1977) and Fiengo (1979) concerns the interpretation of certain structures in- volving NP-preposing. Consider the following sentences, which illustrate this peculiarity of passive nominals (from Fiengo (1979)):

(47) a. *great relief's expression by John b. the expression of great relief by John c. Great relief was expressed by John.

(48) a. *some money's gift to the library by John b. the gift of some money to the library by John c. Some money was given to the library by John.

(49) a. *irregularities' acknowledgment by the senators b. the acknowledgment of irregularities by the senators c. Irregularities were acknowledged by the senators.

These should be compared to the more familiar cases like (50a-c), where preposing is allowed:

(50) a. Rome's destruction by the barbarians b. the product's distribution by the owners c. the movement's infiltration by spies

Fiengo, who attributes the observation to Anderson (1977), points out that movement is disallowed in nominals where the object is unaffected. In all the nominals in (47)-(49) the object of the nominal is not affected by the action expressed by the nominal. In the nominals in (50), on the other hand, the object is affected by the action expressed by the nominal.

Given the ideas outlined above concerning passive nominals, the following analysis comes to mind. Let us assume that those derived nouns belonging to the class that does not allow NP-preposing are special in the following sense: their external 0-roles are not marked as being optional. These nouns always assign a 0-role to the specifier position. Movement to that position leads to ungrammaticality as a result of the 0-Criterion. If no movement occurs, the structure is grammatical. This accounts for the lack of move- ment. These ideas can be made more precise by stating the following principle:

Affectedness Constraint

If a complement of X is unaffected, it is impossible to eliminate the external 0-role of X.

Predicates that take unaffected objects, then, must always express their external 0-role. This is not a problem in passive sentences, since the external 0-role is carried by the

608 OSVALDO A. JAEGGLI

passive suffix. But in passive nominals, a problem arises if NP-movement occurs. In those instances the external 0-role cannot be present, since there is no way for it to be absorbed. This violates the Affectedness Constraint.

The Affectedness Constraint also accounts for the impossibility of *John's belief to be immortal. Passive nominals necessarily involve the elimination of the external 0-role of the nominal. However, since the infinitival complement of belief is an unaffected object, this violates the Affectedness Constraint. Hence, passive nominals of belief are not allowed. As for the impossibility of *the belief of John to be immortal, I assume it is related to the impossibility of Of Insertion (or, more generally, genitive Case assign- ment) before NPs that are not thematically related to the head nominal.

But why should the Affectedness Constraint hold at all? One way of making some sense of such a condition is to consider the thematic interpretation of the external and internal arguments of predicates in instances where the objects are affected, and compare it to what happens when the objects are unaffected. When an object of a predicate is affected by the action expressed by that predicate, its thematic interpretation is well defined. Thus, in a sentence such as John hit Bill, Bill is patient; in John built a house, a house is the result or outcome of the action expressed by the predicate. These thematic relations appear to be quite independent of the thematic relation holding between the external argument and the predicate. But in sentences involving unaffected objects, this is not the case. In The senators acknowledged great irregularities, no well-defined the- matic relation holds between great irregularities and the predicate of which it is an object. Rather, the thematic interpretation of the object appears to be thematically dependent on the thematic interpretation of the subject. Let us say that in these instances the 0- role associated with the internal argument of the predicate is a function of the external 0-role. More precisely, let us express this idea by saying that in these cases, Od = f (Os). But if the external 0-role is missing in a particular structure, it would then become impossible to compute the internal 0-role. Thus, such a structure would be illegitimate because the internal thematic relation would not be well defined. It is only well defined when the external 0-role is also present. If this line of reasoning is correct, the Affect- edness Constraint would simply be an empirical generalization following from a more articulated theory of thematic roles. It would not have the status of an independent principle of grammar. Although I believe this is the correct way to proceed, no com- prehensive theory of thematic structure now exists from which the Affectedness Con- straint can be derived. But the direction of research appears quite clear and promising.

Returning now to sentences (44b), (45b), and (46b), the NP in the by-phrase of these structures must have been assigned the 0-role Agent by default, since this 0-role is nondistinct from the 0-role that has been assigned to the Specifier-of-NP position by the head noun (but see below for further comments). In these structures, then, two positions have been marked with the Agent 0-role: the specifier position via the familiar 0-marking by a head, the NP in the by-phrase via 0-marking by default. This situation obviously does not result in a 0-Criterion violation, since the structures are grammatical. The 0-

PASSIVE 609

Criterion, then, must be properly interpreted as a condition on 0-role assignment and not on the presence or absence of NPs with a specific 0-role interpretation.

If the head of a derived nominal contains Os in its lexical structure, this 0-role must be uniquely assigned, given (10). What is it assigned to? One possibility is that it is assigned by N to the specifier position of the nominal. This assumption is unproblematic when the specifier position is occupied by a referential expression, as in (38), for instance, the barbarians' destruction of Rome. If the head does not contain a referential expres- sion, I will make the auxiliary hypothesis that the 0-role is assigned nonetheless. Since it does not make any sense to assign a 0-role to a determiner such as the in the destruction of Rome to prove a point, I will assume that the determiner is inserted into specifier position after S-Structure and that such nominals contain a PRO in subject position that carries the 0-role. Such a PRO is allowed under the assumption that the nominal head does not govern the specifier position of NP. This requires a more restricted notion of government than the one offered by Aoun and Sportiche (1983), one that requires strict c-command, or perhaps one that takes into account "directionality." Another possibility would be to assume that the PRO in specifier position is somehow protected from gov- ernment. For example, it might be maintained that the PRO in specifier position is in fact contained in a bigger NP as the sister of an abstract counterpart of 's, as in [[[PRO] 0]i destruction]. The empty 's does not govern, and the NP dominating PRO -0 protects the PRO from government by the head noun. Though I will not choose between the alternatives, I will tentatively adopt such a result, pending further investigations in this area.

This proposal leaves the way open for NPs of the form [PRO book], which do not appear to be possible (Aoun and Sportiche (1983)). But these NPs are excluded by the familiar requirement that PRO must be in a 0-chain: compare *PRO to seem that John is crazy would be terrible for his political career with For it to seem that John is crazy would be terrible for his political career. Nouns like book do not assign 0-roles. Hence, a PRO in the specifier position of such nominals would fail to be in a 0-chain. Whatever mechanism enforces this requirement for PRO will rule such structures ungrammatical. PRO must always be in a 0-chain. This may be due to the fact that PRO, as an empty pronominal anaphor, must meet certain identification requirements, and these require- ments include such a stipulation. Note, however, that a referential NP may appear in specifier position of an NP headed by books, as in John's books. If books does not assign a 0-role to its specifier position, how does John satisfy the 0-Criterion in such structures? In such cases John is interpreted as the "possessor" of the head N, where the "pos- sessor" relation is in fact very vaguely defined. It may involve ownership, possession, authorship, the books that John has been talking about incessantly, or many other imaginable connections between John and the books. In other words, in these cases there is no clear thematic relation between the head noun and its specifier. Let us assume that the "possession" relation is assigned by default simply in order to satisfy the re- quirement that every referential expression should be fully interpreted in a given thematic

610 OSVALDO A. JAEGGLI

domain. This default 0-role would allow John to pass the 0-Criterion in structures such as John's books. However, this possession relation is not the product of a 0-chain, since it is not assigned to the specifier by the head N. It is not sufficient to meet the requirement on PRO; hence the impossibility of *[PRO books].

The preceding discussion concerning PRO in the specifier position of NPs involves an implicit assumption that should be made explicit. I assumed earlier, in (10), that all 0-roles must be uniquely assigned. That is, under this view, 0-role assignment is obliga- tory. Such a principle rules out It/There kills the rat with expletive itlthere, with the interpretation 'the rat was killed'. Given such a principle, we are forced to conclude that certain NPs contain PRO in specifier position, as noted above. This follows if we assume that the principle in (10) holds indiscriminately within all thematic domains. However, it is not inconceivable that the requirement that an external 0-role be obli- gatorily assigned is not uniform across different syntactic categories. Thus, it may be the case that the external 0-role must always be assigned in a clause, but not in an NP. This situation, in turn, may follow from the well-known fact that external 0-roles are optional in NPs, but not so in clauses. In NPs, then, the assignment of the external 0- role may be optional, even if the 0-role is present in the structure in the head N. This view would then not require that we posit a PRO in specifier position in structures such as the destruction of the city. Note, however, that the optionality of assignment must be limited to instances where specifier position is occupied by a nonreferential expres- sion, like a determiner, and must never occur in instances where that position is occupied by a referential expression. Thus, we maintain the result that movement into specifier position of a nominal is only possible if the structure does not contain an external 0- role. This result will be further motivated in the later discussion of purpose clauses in derived nominals (section 7)16

These remarks are rather speculative, and they require making several nontrivial theoretical assumptions (for example, concerning the proper notion of "government"). Nevertheless, I hope to show that several rather interesting facts follow from an analysis along the general lines indicated above. This in turn should provide some empirical motivation for these decisions.

Finally, I would like to point out a well-known fact concerning passive nominals in Italian that underscores the difference between passive nominals and passive sentences.

16 Safir (1984, 624, fn. 24) points out that the linked reading characteristic of constructions that contain PRO subjects, as in (i) (see Lebeaux (1984) for the original observation concerning these sentences), is also found with certain derived nominals, as in (ii):

(i) To raise the rent is to irritate the tenants. (ii) Devotion to her country is devotion to his flag.

Safir concludes that "if the phenomenon observed by Lebeaux in [(i)] is to be described in terms of the presence of a PRO subject of the infinitives, then the phenomenon in [(ii)] suggests that these nominals also have PRO subjects." Note, however, that these nominals lack a determiner. Interestingly, Lebeaux (1984) points out that -ing nominals with a determiner do not exhibit a linked reading. For example:

(iii) The winning of games requires the losing of games. Thus, the presence or absence of a determiner appears to be crucial in these cases.

PASSIVE 611

In passive nominals by-phrases are formed with a special prepositional form: da parte di. This form is reserved for passive nominals and does not occur in passive sentences. The preposition da in Italian must be reserved for those instances where 0-role trans- mission occurs. It is incapable of assigning 0-roles by default. Only the prepositional form da parte di can accomplish this. Interesting confirmation for this position comes from a consideration of middle constructions. Belletti (1982) points out that (as in other languages) middles do not allow by-phrases in Italian. But they do allow da parte di phrases. Consider the following sentences from Belletti (1982):

(51) a. I dolci al cioccolato si mangiano in questa pasticceria (*da Mario). the sweets of chocolate eat-pl in this bakery by Mario

b. Le matterie letterarie si studiano in questa universita (*da molti the courses literary study-pl in this university by many studenti). students

c. Si dicono cose impensabili da parte di tutti ultimamente. say-pl things unthinkable on part of everyone lately

We can now turn to the analysis of constructions such as (33a), The man was killed. Such constructions are said to contain an "implicit argument." The external 0-role in such constructions is not realized on a lexical NP in an argument position. I argue that it is assigned to the passive suffix. In a sense, however, the term implicit argument is misleading. Given the analysis presented below, the external 0-role of the verb is ex- pressed explicitly on the passive suffix. It simply is not expressed on an NP position. Compare with this the situation that obtains in so-called passive nominals that lack a "passive" suffix. Nominals can truly be said to contain implicit arguments. Neverthe- less, I will arbitrarily continue to use the term implicit argument to refer to cases such as (33a), following much current usage.

6. Implicit Arguments

Chomsky (1981, chap. 2, fns. 60 and 101) points out that there is reason to believe that passives without a by-phrase are nevertheless "agentive." He cites the following ob- servation from Manzini (1983): a passive sentence allows a purpose clause, whereas a "middle" construction does not. For example:

(52) a. They decreased the price [to help the poor]. b. The price was decreased [to help the poor]. c. *The price decreased [to help the poor].

A similar phenomenon can be observed with adverbs that Jackendoff (1972) calls agent- oriented adverbs:

(53) a. They decreased the price willingly. b. The price was decreased willingly. c. *The price decreased willingly.

612 OSVALDO A. JAEGGLI

These facts provide strong motivation for distinguishing between middles and passives without a by-phrase. (For ease of exposition, I will refer to passives without a by-phrase as agentless passives, though this in no way entails that I believe either that these passives do not have an agent or that all passive by-phrases contain NPs that bear the Agent 0-role.) A natural way of distinguishing between them has been to claim that an Agent 0-role is present in passives even in the absence of a by-phrase, whereas this is not the case in middles. The lexical rule that forms middle verbs like decrease must involve the deletion of the external 0-role of the verb and the externalization of the Theme 0-role, all of these operations being possible lexical operations. In passives the external 0-role is not deleted. Rather, it is assigned to the suffix -en. But if this much is clear, the precise way to implement this idea has often been left rather vague. What does it mean to say that an Agent 0-role is present in an agentless passive? Where is it located? What properties does it have, and how do these properties interact with purpose clauses and subject-oriented adverbials? In this section I will sketch an analysis that provides answers to these questions.

The first point to be made is that it is incompatible with most recent versions of the theory of grammar to assume that implicit arguments are actually carried by an empty category in a VP-internal argument position. That is, I want to argue against a structure like (55) for (54):

(54) The man was killed. (55) [the man] was [killed t [EC]

Od Os

In order for (55) to be a well-formed syntactic representation, it must be possible to identify the empty category as one of the recognized independently motivated empty categories. Let us assume that the inventory of empty categories consists of the following elements: anaphors, variables, PRO (which can be either controlled or arbitrary in in- terpretation), and pronominals (of which there are two kinds: expletive and referential).

We can quickly dismiss the first possibility. If the empty category were an anaphor, it would violate principle A of the binding theory, which requires that anaphors be A- bound within their governing categories, since there is no A-binder to bind it.

At first sight, the second possibility (that the empty category is a variable) appears far-fetched. There is no overt operator to bind the empty category, and variables must be operator bound. However, one might assume that the structure contains an empty operator that binds the variable. This possibility, which would have to be motivated further, might appear feasible. (54) would then have the structure shown in (56):

(56) [Operatori] [[the manb was killed tj [EC]i]

But this proposal runs into severe problems. One problem concerns the general availability of the mechanism of empty operators

in English. Although empty operators do exist in English, they are prohibited in struc- tures such as (56). If we make allowances to license an empty operator in such a structure,

PASSIVE 613

we will have to explain why it does not also make the unacceptable sentences in (57a- d) possible, with the structures given in (57e-h):

(57) a. *John hit. b. *John put on the table. c. *John put the book. d. *Gave the book to Susan. e. [Operatori] [John hit ECJ] f. [Operatori] [John put ECi on the table] g. [Operatori] [John put the book ECJ] h. [Operatori] [ECi gave the book to Susan]

Another problem concerns instances of control. It is well known that a variable can act as a controller in a structure of control, as in the following sentences:

(58) a. John persuaded Bill [PRO to go to the store]. b. Who did John persuade t [PRO to go to the store]? c. John promised Bill [PRO to go to the store]. d. Who did John say t promised Bill [PRO to go to the store]?

The variables in (58b,d) function perfectly well as controllers of the PRO subject of the infinitival embedded under persuade, promise. This behavior in fact extends to those empty operators that are allowed in English, as in (59):

(59) a. the man [[Operator] Bill persuaded t [PRO to go] b. the man [[Operator] Bill thinks t promised Mary [PRO to go to the store]

However, the "variable" in (56) does not behave in this way. Consider:

(60) *[Operator]j Billi was promised ti ECj [PRO to go to Disneyland]

Thus, if (54) contains a variable, as in (56), it must be a special type of variable, with properties quite different from those normally associated with variables. (The accept- ability of John was promised [PRO to be allowed to go] is irrelevant to this argument, since in this sentence PRO is controlled by John, not the implicit argument of promise.)

Consider next the third choice for the identification of the empty category in (55). The empty category cannot be a pronominal anaphor. It is in a governed position, and PRO is not allowed in governed positions. Furthermore, Chomsky (1986) points out that implicit arguments cannot be controlled. Consider:

(61) a. They expected [PRO to give damaging testimony]. b. They expected [damaging testimony to be given].

In (61b) the giver(s) of testimony cannot be they. This evidence suggests strongly, then, that the empty category in (55) cannot be PRO.

Finally, since English does not allow null pronominal nonanaphors, the possibility that the empty category in (55) is a null pronominal (for instance, pro) can be dismissed.

614 OSVALDO A. JAEGGLI

In fact, it is not clear whether this possibility should be available. If it is, sentence (61b) could have the excluded interpretation mentioned above, with pro instead of PRO. Since null pronominals need not be ungoverned, if the empty category in (57) could be a pro, it would allow an impossible reading. Note that a full pronoun in a parallel structural position does allow such a reading: They expected an office to be given to them, where they, them are coreferential.'7

As this exhausts the inventory of possible empty categories, I conclude that there is no empty category bearing the Agent 0-role in (54). Instead, I will assume that the Agent 0-role in (54) is realized on the passive suffix -en. The problems mentioned above do not arise under this analysis. As there is no anaphor, or pronominal nonanaphor, the problems associated with these options do not arise. As there is no NP argument bearing the Agent 0-role, the unavailability of a controller in (60) is straightforwardly accounted for. For the same reason, the unavailability of control of the implicit argument, as in (61b), is also accounted for.

Chomsky (1986) identifies another property of the implicit argument, which concerns predication. Whereas PRO enters into predication, this is not possible with implicit arguments:

(62) a. They expected [PRO to leave the room sad]. b. *The room was left sad.

The theory presented here easily accounts for this. Following Williams (1980), I will assume that predication involves mutual c-command of a subject position and a predicate. This condition is satisfied in (62a). But since (62b) does not contain an NP position bearing the Agent 0-role, predication cannot obtain. If it obtains at all, it yields the interpretation that it was the room that was sad, as in Bill was left crying all alone.

Assuming then that implicit arguments are not mapped onto NP argument positions, let us continue to explore some of the properties of constructions such as (54). In par- ticular, let us consider once again the contrast between (52b) and (52c). Examples such as (52b) have led to the opinion that an implicit argument functions as the controller of

17 Epstein (1984) presents interesting arguments claiming that pro exists in English and that it is interpreted as a universal quantifier. He argues that pro exists in sentences such as (ia), with the LF structure given in (ib):

(i) a. It is fun to play baseball. b. (x) [it is fun (for) x [x to play baseball]]

The empty category complement of fun would be a pro. This pro, however, functions as a pronominal with respect to coreference:

(ii) They expected it to be fun to play baseball. This is in contrast to the facts concerning implicit arguments of passives. Also, the unexpressed argument of fun that Epstein claims to be a pro can control into passive infinitivals:

(iii) It is fun to be tickled by elephants. Once again, this is in marked contrast to implicit arguments of passives, as will be shown below. These differences between the unexpressed argument investigated by Epstein and implicit arguments of passives suggest that they should be treated differently.

PASSIVE 615

the PRO subject of the infinitival purpose clause. (For an extensive discussion of implicit arguments under this assumption, see Roeper (1983).) Control is claimed to be involved for two closely related reasons: (a) In (52b) the subject of the infinitival purpose clause is interpreted as "coreferential with" the implicit argument. In fact, the implicit argument is interpreted as "arbitrary" in reference and so is the subject of the infinitival purpose clause. (b) "Arbitrary" reference is a notion that comes under the general theory of control. I will argue below that "control" by the implicit argument in sentences such as (52b) is radically different from the familiar notion of control. I will show that the phenomenon obeys different generalizations, and that it would be a mistake to analyze it under the general theory of control. To distinguish it from standard cases of control, I will label it thematic control instead, and refer to the familiar notion of control as argument control.

One difference between thematic control and argument control has already been mentioned. Whereas implicit arguments cannot participate in argument control, they do participate in thematic control. Compare in this respect (60) with (52b), repeated below:

(52) b. The price was decreased to help the poor. (60) *Bill was promised to go to Disneyland.

Conflating the two notions would only confuse the issue. It cannot be claimed, for ex- ample, that control is impossible in (60) because of peculiarities of the verb promise. It is only argument control that is impossible in (60). Thematic control is quite possible in similar structures, as illustrated by the following sentences:

(63) a. John was promised that he would go to Disneyland [to get him to stop crying].

b. John was promised that he would be the winner [to make him feel a little bit better].

The purpose clauses in these sentences (in brackets) are controlled by the implicit ar- gument of promise. Thus, the implicit argument of promise is quite capable of thematic control; it is simply incapable of argument control. This follows naturally if we assume that argument control involves control from argument positions, whereas thematic con- trol does not. Promise is a subject (argument) control verb. It designates as its controller the argument that carries its external 0-role. Since its external 0-role is not carried by an argument in (59) and (62), but rather is carried by the passive suffix, it cannot par- ticipate in argument control. But it can participate in thematic control since thematic control does not involve control from argument positions.

Another difference between argument control and thematic control is that argument control of sentences in object position requires c-command,18 whereas thematic control does not. Thus, even though the external 0-role of promise is carried by an NP in (64),

18 It is well known that the c-command condition is relaxed in certain cases of object control, as in I said to Mary to go to the store. These cases always involve indirect objects, though. I will assume that, these cases aside, the c-command condition holds in general.

616 OSVALDO A. JAEGGLI

PRO is still not controlled by Bill because Bill does not c-command PRO:

(64) *John was promised by Bill [PRO to win].

But thematic control is not sensitive to this c-command restriction:

(65) The price was decreased by the government [PRO to help the poor].

The PRO subject of the infinitival purpose clause is thematically controlled by the government.

Yet another difference between thematic control and argument control is that ar- gument control is possible at a distance, but thematic control is not. For example:

(66) They thought I had suggested that [PRO feeding each other] would be difficult. (from Chomsky (1981, 78))

Consider next what happens in instances of thematic control:

(67) John was told [PRO to clean the house] [in order PRO to impress the guests].

Sentence (67) is ambiguous. The subject of the purpose clause can be interpreted as coreferential either with John or with the implicit argument of told. That is, the sentence is ambiguous between the following two paraphrases: John was told to clean the house in order for John to impress the guests, or John was told to clean the house in order for whoever told John to clean the house to impress the guests. This ambiguity can be analyzed as an instance of structural ambiguity. If the purpose clause is outside the infinitival complement of told, its subject will be thematically controlled by the implicit argument of told. If the purpose clause is within the infinitival complement, it will be controlled by the subject of clean, which in this case is argument controlled by John. Now note that this ambiguity disappears in the following example, which involves one more level of embedding:

(68) John was told that [PRO to clean the house [in order to impress the guests]] is foolish.

In (68) the subject of the purpose clause can be controlled only by the subject of clean, either John or an arbitrary referent. It cannot be interpreted as thematically controlled by the implicit argument of told. Long-distance thematic control is impossible. This is yet another difference between argument control and thematic control.

Finally, argument control is quite possible into passive infinitivals,19 as the following sentences show:

(69) a. John wants [PRO to be loved by everyone]. b. Bill tried [PRO to be introduced to Mary]. c. John persuaded Bill [PRO to be arrested by the KGB].

'" This fact is a clear indication that argument control cannot possibly be a D-Structure phenomenon, as claimed in Roeper (1983). It may be the case that thematic control is a deep structure phenomenon, though.

PASSIVE 617

But thematic control is not possible into passive infinitivals. Compare:

(70) a. The gifts were brought [PRO to impress the Indians]. b. *The gifts were brought [PRO to be admired by the Indians].

(71) a. The report was carefully prepared [PRO to impress the board of directors]. b. *The report was carefully prepared [PRO to be congratulated by the board

of directors]. (72) a. The structure of DNA was investigated [PRO to advance our knowledge

of molecular biology]. b. *The structure of DNA was investigated [PRO to be awarded the Nobel

Prize].

Argument control is a process that takes into account properties of derived structures, as the sentences in (69) show. The sentences in (70)-(72), on the other hand, lead us to consider thematic control to be a D-Structure phenomenon. Perhaps the differences between thematic control and argument control are all a function of the level of rep- resentation at which each control process occurs. Thematic control would occur at D- Structure, whereas argument control would occur at S-Structure. If this is so, the dif- ferences need not be interpreted as involving different rules or rule systems. Instead, they may involve the same rule applying at different levels of representation. If this is true, it would provide further motivation for the notion of distinct linguistic levels of representation; and the phenomenon I have been calling thematic control would be evi- dence for an independent level of D-Structure.

To summarize, the facts presented above strongly suggest that thematic control should be distinguished from argument control.20 I have suggested that the distinction

20 Lasnik (1984) also argues for distinguishing what I am calling thematic control from standard argument control. He suggests that in (i)

(i) The ship was sunk [PRO to prove a point]. it is not the passive affix that is controlling the PRO in the purpose clause; rather, the matrix clause The ship was sunk is the controller. That is, the event itself is interpreted as intended to prove a point. Lasnik (1984) calls this alternative analysis event control. Williams (1984) also discusses such a possibility, which he labels S-control.

According to Lasnik, the ungrammaticality of (ii) (ii) *The ship was sunk [PRO to be promoted].

could then be explained by the theory of event control as follows: "Event is not a possible subject of the passive VP in the purpose clause in [(ii)]." That is, (ii) is ruled out presumably for the same reason as (iiia- b):

(iii) a. *The sinking of the ship was promoted. b. *The event was promoted.

Note, however, that this account does not carry over to the sentences in (71) and (72), as should be clear from the sentences in (iv):

(iv) a. The careful preparation of the report was praised by the board of directors. b. The discovery of the structure of DNA was considered brilliant.

In these cases the event in question is a possible subject of the passive VP in the purpose clause. But still the sentences in (71) and (72) are unacceptable. Sentences like (v) might also constitute a problem for the theory of event control:

618 OSVALDO A. JAEGGLI

may be a matter of the level of representation at which the control rule applies. With this idea in mind, I would like to investigate other instances that might be analyzed as instances of control by an implicit argument.

7. Purpose Clauses in Derived Nominals

Purpose clauses are also found in derived nominals. But not all derived nominals allow them. Roeper (1983) points out the following contrasts:

(73) a. the destruction of the city by the enemy to make a point b. *the destroyer of the city to make a point c. the destruction of the city to make a point

Let us begin by considering (73b). Its ungrammaticality suggests that the crucial property required for the correct interpretation of the purpose clause-namely, the presence of an Agent 0-role-is missing in this case. At first glance this might seem surprising, since the suffix -er is typically referred to as an "agentive nominalizer" (see, for example, Lieber (1983, 267)). But this does not mean that the derived nominal itself contains an Agent 0-role that is syntactically accessible to the purpose clause. In fact, this would be rather surprising. Destroyer denotes an individual, and individuals typically do not have Agents or Causes. Events have Agents or Causes. Thus, it is natural to assume that destruction-an Event-contains an Agent 0-role, but not destroyer. In fact, de- struction freely allows purpose clauses, as can be seen in (73a,c). Suffixation of -er, on the other hand, must involve the deletion of the Agent 0-role of the verb destroy. This would account for the ungrammaticality of (73b).

Consider next the following contrast, also from Roeper (1983):

(74) a. the destruction of the city to prove a point b. *the city's destruction to prove a point

This contrast follows in a straightforward way from the theory of passive nominals sketched in section 5. Recall that in order for NP-movement to take place in a nominal, it must be the case that the external 0-role of the head of the nominal is not present in the lexical structure of the head of the nominal. If it were, there would be no way to "deflect" its assignment to the specifier position, since nominals do not carry passive

(v) *The house was struck by lightning [PRO to start a fire]. In this case the event is quite capable of being the subject of the VP in the purpose clause, as (vi) demonstrates:

(vi) The lightning striking the house started a fire. An event control analysis would then predict that (v) should be acceptable. The problem with this approach is obvious. Intentionality must be involved in order for sentences that involve thematic control to be acceptable. In (v) lightning cannot be attributed the intention of starting a fire; neither can it be interpreted as an instrument (unless one interprets the agent of strike to be Jupiter, in which case the sentence becomes acceptable). Events do not have intentions; individuals bearing a particular thematic relation to a predicate do. An event control theory must capture this fact in order to avoid the problems mentioned above.

PASSIVE 619

suffixes and 0-role assignment to specifier position is obligatory in nominals if that po- sition occupies a referential expression. Assignment of the external 0-role to the specifier position in (74b) will result in a 0-Criterion violation, since the city is in a chain with [NP,VP] position, which is assigned the Od of destruction. But if the agentive 0-role of the nominal is missing, the ungrammaticality of (74b) reduces to the ungrammaticality of (73b) or of the "middle" construction in (5 1c). There is no problem, then, in accounting for this contrast.

Note that control of purpose clauses in derived nominals does not have the properties of what I have been calling thematic control. Rather, it behaves like an instance of argument control. It requires c-command, as shown by the fact that even if a by-phrase is added to (74b), it is still ungrammatical:

(75) *the city's destruction by the army to prove a point

It is possible at a distance, just like the case of argument control mentioned above. For example:

(76) Their statement that I had suggested that feeding each other would be difficult is completely false.

And it is possible to control into passive infinitivals:

(77) a. The attempt to be introduced to the king failed. b. *The secretary was called to be introduced to the king.

(77b) is ungrammatical on the reading that it is the person who called the secretary who wanted to be introduced to the king. These criteria all point to the conclusion that control in nominals is always argument control. This follows under the analysis presented here, since in effect nominals do not have passive suffixes that can absorb a 0-role. That is, thematic control appears not to be possible in nominals.

Finally, consider the following contrast also pointed out in Roeper (1983).

(78) a. *The fact was unmentioned/undetected to prove a point. b. The fact was mentioned/detected to prove a point.

(78a) is an instance of an "adjectival passive." I will assume that these structures do not involve NP-preposing. Rather, they are base-generated essentially as such (see Siegel (1973) and Wasow (1977)). The participle appears prefixed with negative un-, a prefix that attaches only to adjectives. The ungrammaticality of (78a) suggests once again that these sentences do not contain an Agent 0-role that is accessible for thematic control, whereas the sentences in (78b) do. Further evidence on behalf of this analysis comes from the fact that subject-oriented adverbs are also excluded in this construction, as has been noted by Siegel (1973). For example:

(79) a. Nixon's gaffe wasn't enthusiastically publicized. b. Nixon's gaffe was (*enthusiastically) unpublicized.

620 OSVALDO A. JAEGGLI

Subject-oriented adverbs also require the presence of an external argument in the struc- ture in which they occur.

Furthermore, these facts correlate with the peculiarities related to the presence or absence of by-phrases in adjectival passives. Siegel (1973, 301, fn. 2) points out that many speakers find adjectival passives with by-phrases ungrammatical. Those who do find such sentences grammatical generally report that they vary in acceptability ac- cording to the particular content of the by-phrase. Thus, some speakers find a contrast between the following sentences (from Siegel (1973)):

(80) a. Antarctica is uninhabited by man. b. *Antarctica is uninhabited by the guy next door who plays Olympic

dominoes.

In these and similar cases by-phrases are best only when they contain generic NPs. No similar restriction occurs with true passives. I assume that whatever grammatical mech- anism is at work to interpret by-phrases in adjectival passives, it is different from the one that occurs with other passives, and crucially it is not one that involves accessibility to the external 0-role of the verbal base of the adjectives found in these structures.

Rdferences

Anderson, M. (1977) "Transformations in Noun Phrases," ms., University of Connecticut, Storrs. Aoun, J. and D. Sportiche (1983) "On the Formal Theory of Government," The Linguistic Review

2, 211-236. Belletti, A. (1982) "'Morphological' Passive and Pro-Drop: The Impersonal Construction in

Italian," Journal of Linguistic Research 2, 1-34. Belletti, A. and L. Rizzi (1981) "The Syntax of ne: Some Theoretical Implications," The Linguistic

Review 1, 117-154. Borer, H. (1984) "The Projection Principle and Rules of Morphology," in C. Jones and P. Sells,

eds., Proceedings of the Fourteenth Annual Meeting of NELS, GLSA, University of Mas- sachusetts, Amherst.

Burzio, L. (1981) Intransitive Verbs and Italian Auxiliaries, Doctoral dissertation, MIT, Cam- bridge, Massachusetts.

Chomsky, N. (1965) Aspects of the Theory of Syntax, MIT Press, Cambridge, Massachusetts. Chomsky, N. (1970) "Remarks on Nominalization," in Studies on Semantics in Generative Gram-

mar, Mouton, The Hague. Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht. Chomsky, N. (1986) Knowledge of Language: Its Nature, Origin, and Use, Praeger, New York. Chung, S. (1976a) "On the Subject of Two Passives in Indonesian," in C. Li, ed., Subject and

Topic, Academic Press, New York. Chung, S. (1976b) "An Object-Creating Rule in Bahasa Indonesia," Linguistic Inquiry 7, 41-87. Comrie, B. (1977) "In Defense of Spontaneous Demotion," in P. Cole and J. Sadock, eds., Gram-

matical Relations, Syntax and Semantics 8, Academic Press, New York. Epstein, S. D. (1984) "Quantifier PRO and the LF Representation of PROarb," Linguistic Inquiry

15, 499-505. Fabb, N. (1984) Syntactic Affixation, Doctoral dissertation, MIT, Cambridge, Massachusetts.

PASSIVE 621

Fiengo, R. (1979) Surface Structure: The Interface of Autonomous Components, Harvard Uni- versity Press, Cambridge, Massachusetts.

Higginbotham, J. and R. May (1981) "Questions, Quantifiers, and Crossing," The Linguistic Review 1, 41-79.

Hornstein, N. (1975) "S and the X' Notation," Montreal Working Papers in Linguistics 4, De- partment of Linguistics, McGill University, Montreal.

Jackendoff, R. (1972) Semantic Interpretation in Generative Grammar, MIT Press, Cambridge, Massachusetts.

Jackendoff, R. (1977) X-Syntax: A Study of Phrase Structure, MIT Press, Cambridge, Mas- sachusetts.

Josephs, L. (1975) Palauan Reference Grammar, University of Hawaii Press, Honolulu. Kayne, R. S. (1975) French Syntax, MIT Press, Cambridge, Massachusetts. Kirsner, R. S. (1976) "On the Subject of Pseudo-passives in Standard Dutch and the Semantics

of Background Agents," in C. Li, ed., Subject and Topic, Academic Press, New York. Lasnik, H. (1984) "Random Thoughts on Implicit Arguments," ms., University of Connecticut,

Storrs, and MIT, Cambridge, Massachusetts. Lebeaux, D. (1984) "Anaphoric Binding and the Definition of PRO," in C. Jones and P. Sells,

eds., Proceedings of the Fourteenth Annual Meeting of NELS, GLSA, University of Mas- sachusetts, Amherst.

Lieber, R. (1980) On the Organization of the Lexicon, Doctoral dissertation, MIT, Cambridge, Massachusetts.

Lieber, R. (1983) "Argument Linking and Compounds in English," Linguistic Inquiry 14, 251- 285.

Manzini, M. R. (1983) "On Control and Control Theory," Linguistic Inquiry 14, 421-467. Marantz, A. (1984) On the Nature of Grammatical Relations, MIT Press, Cambridge,

Massachusetts. Mihailovic, L. (1966) "The Agent in the English Passive," English Language Teaching 21, 123-

126. Perlmutter, D. (1978) "Impersonal Passives and the Unaccusative Hypothesis," in J. Jaeger et

al., eds., Proceedings of the Fourth Annual Meeting of the Berkeley Linguistics Society, Berkeley, California.

Roberts, I. (1985) The Representation of Implicit and Dethematized Subjects, Doctoral disser- tation, University of Southern California, Los Angeles.

Roeper, T. (1983) "Implicit Thematic Roles in the Lexicon and Syntax," ms., University of Massachusetts, Amherst.

Rouveret, A. and J.-R. Vergnaud (1980) "Specifying Reference to the Subject," Linguistic Inquiry 11, 97-202.

Safir, K. (1984) "Multiple Variable Binding," Linguistic Inquiry 15, 603-638. Siegel, D. (1973) "Nonsources of Unpassives," in J. P. Kimball, ed., Syntax and Semantics, vol.

2, Academic Press, New York. Siewierska, A. (1984) The Passive: A Comparative Linguistic Analysis, Croom Helm, London. Stowell, T. (1981) Origins of Phrase Structure, Doctoral dissertation, MIT, Cambridge, Mas-

sachusetts. Subbiah, G. (1972) "A Note on Agreement in Kota," in S. Agesthialingom and S. V. Shanmugan,

eds., Third Seminar on Dravidian Linguistics, Annamalainagar, Annamalai University. Wasow, T. (1977) "Transformations and the Lexicon," in P. W. Culicover, T. Wasow, and

A. Akmajian, eds., Formal Syntax, Academic Press, New York. Williams, E. (1980) "Predication," Linguistic Inquiry 11, 203-238.

622 OSVALDO A. JAEGGLI

Williams, E. (1981) "Argument Structure and Morphology," The Linguistic Review 1, 81-114. Williams, E. (1984) "PRO and Subject of NP," ms., University of Massachusetts, Amherst, and

MIT, Cambridge, Massachusetts. Wilson, H. (1972) "The Phonology and Syntax of the Palauan Verb Affixes," Working Papers in

Linguistics 4.3, University of Hawaii, Honolulu. Zubizarreta, M. L. (1985) "The Relation between Morphophonology and Morphosyntax: The Case

of Romance Causatives," Linguistic Inquiry 16, 247-289.

Department of Linguistics University of Southern California Los Angeles, California 90089-1693