Passive Syntactic Structures

ing away from terminology, these quotes straightforwardly paraphrase a syntactic passive relation between a direct object and a subject. The import of their findings is thus: the semantic role of the subject of a passive adjective is always that of the corresponding verb's deep direct object. This is exactly the conclusion that the passive sub-case of Move NP derives passive adjectives. 7.2. A non-problem concerning NP Movement with adjectival passives file:///G|/htdocs/Joe/3WEB_OK_Passives_and_Perfects_(POST).htm (12 of 21) [5/11/2003 5:08:06 PM] PASSIVE SYNTACTIC STRUCTURES Chomsky’s (1970) long accepted analysis of derived nominals, which have idiomatic and lexicalized meanings, uses transformational object proposing. The lexicalized idiomatic meanings of many passive adjectives are thus no barrier to deriving adjectival passives by NP movement. 7.3. Reconsidering the role of the Case Filter in passives Explaining section 3.7. As soon as both types of passives are derived by movement, we find that the Case Filter predicts an unstipulative and exact distinction between adjectival and verbal passives. Movement to a subject position is the only source of a DP gap after passive verbs. But the Case Filter further restricts the complement range in adjectival passives, as in (38). (38) Ann { was/ *seemed } given the letter. That letter { was/ *sounded } sent all the candidates. Who { got/ *acted } taken such unfair advantage of? How much unfair advantage { got/ *looked } taken of Bill? Peter { was/ *felt } forgiven his sins. Peter's sins { were/ *stayed } forgiven him. Those workers { were/ *seemed} allowed a lot of vacation. He { was being charged/ *felt charged } a lot of money. The examples in (38) demonstrate that a V in a verbal passive can assign case. Claims to the contrary invariably invoke ad hoc mechanisms for case in these paradigms. Structural cases should thus be assigned optionally; then if a DP doesn't receive case from V, the DP may move to get case. The government and binding motivation for what forces passive NP movement to subject position, namely the inability of an A to assign case to a DP sister, never actually worked for passive verbs without stipulations for the examples in (38). But it works perfectly for passive adjectives, as they never occur in “double object” constructions. 8. The Lexical Entry for -en as the source of the “DP gap” in passives 8.1. The Φ features on –en Nothing said so far accounts formally for the DP trace or its properties in passive structures; moreover, it remains to complete (20). file:///G|/htdocs/Joe/3WEB_OK_Passives_and_Perfects_(POST).htm (13 of 21) [5/11/2003 5:08:06 PM] PASSIVE SYNTACTIC STRUCTURES (20) Passive participles (2nd version): -en, (A), +V , .... I propose that for English, all we need say for both purposes is that passive –en also realizes Φ features (agreement): (39) Passive participles (final version): -en, (A), +V , Φ features However we develop a theory of agreement, it seems clear that a non-nominal category specified to agree with a nominal (that is what Φ features consist of) will agree with the closest ccommanding nominal, i.e. its nominal sister if it has one. Additionally, it probably can agree with no other. Several works including Emonds (2000, Ch. 4) develops a principle called Alternative Realization (“AR”) that formally characterizes inflectional morphology, including agreement of this type. More detail: Alternative Realization allows lexical entries of closed class items to exhibit features that are not canonical for the category in question, but are “borrowed” from an adjacent item. (40) Alternative Realization. If F is a cognitive syntactic feature canonically associated with a category B, F can also be spelled out in a closed class grammatical morpheme under X0, where Xk is a sister of [B, F]. A feature F in its canonical position is interpreted in LF but an alternatively realized F is not. If all the features of a morpheme are contextual frames or alternative realizations, then none are interpreted, and so such morphemes are inserted in PF by (28). Reflection on (40) shows that AR allows a dependent morpheme under a head to spell out features of a complement, which is then null. For example, an accusative Romance clitic zeroes an object DP by alternatively realizing the object's features [DEF, +FEM, +PLUR] on a morpheme under V0. The most characteristic features of a direct object are the Φ features of its D (Person, Number, Gender). Although the passive morpheme -en cannot, as an A, alternatively realize D itself, it can be alternatively realize Φ features. This allows (39) above. End of additional detail. In lexically unmarked uses of AR a canonical position [B, F] must be null whenever all of B’s features are alternatively realized (the Invisible Category Principle of Emonds, 1987). E.g., I is empty when Tense is on V. Thus morphological “doubling,” though common, is a marked option. file:///G|/htdocs/Joe/3WEB_OK_Passives_and_Perfects_(POST).htm (14 of 21) [5/11/2003 5:08:06 PM] PASSIVE SYNTACTIC STRUCTURES The AR exemplified by English and French -en appears to be unmarked in this sense. A sister DP of a passive participle is thus required to be empty by the alternatively realized Φ features specified in the lexical entry of -en. The Φ features in (39) thus explain the gap in both types of passives. This approach to passive gaps reflects McA'Nulty (1983) and Lefebvre (1988), who states: “Agreement morphology on the past participle spells out the features of the trace of the NP governed by the past participle.” 8.2. Subject-object co-indexing in passives Section 6 showed how adjectival Φ features on the adjective -en play a crucial role in ensuring that its clausal subject is “theta-free” with respect to the V host. Section 8.1 has shown that the AR source of these Φ features on -en is an empty category in object position. Moreover, unmarked adjectives, including –en, agree with their subject. By transitivity, objects and subjects of (both adjectival or verbal) passives must agree in Φ features with each other. Assuming Φ feature agreement reflects co-indexing, these two co-indexings, again by transitivity, imply that the subject and object positions of -en are also co-indexed, as in an adjectival passive: (41) If the subject DP position is theta-free, this required PF indexing can arise only by movement. This characteristic co-indexing of passive transforms is thus guaranteed by the lexical entry for -en file:///G|/htdocs/Joe/3WEB_OK_Passives_and_Perfects_(POST).htm (15 of 21) [5/11/2003 5:08:06 PM] PASSIVE SYNTACTIC STRUCTURES itself, which furnishes the agreeing category A with the phi-features of the object. That is, the lexical entry (39) minimally and elegantly captures the passive property. 8.3. Review of principles and structure of the English verbal passive (42) Typical verbal passive: Mary will get shown many letters. (43) Verbal passive structure at all levels except PF; α= [+FEM, +PLUR]i: This PF insertion option for get, be and -en, in conjunction with independently justified principles of grammar, completely explains the syntactic properties of the English verbal passive as follows: The absence of -en at syntactic levels, a choice provided by the lexical entry (39), explains why verbal passives do not tolerate adjectival prefixes (e.g. -un) or specifiers such as very, too, etc. The absence of -en at LF also explains why verbal passives have no sense of completed action. Since the lexical head of the verbal passive is V, it is natural that verbal passives of verbal idioms such as make a great deal of and take advantage of occur freely. Since V in (43) is a lexical head, the Extended Projection Principle (22) requires that it have a structural subject. But since the Headedness Corollary (36) renders SPEC(IP) inaccessible to theta role assignment by V, the required subject must be realized "less economically" by generating an extra DP inside VP, either an overt PRO or inside a case-marking by-phrase. Verbal passives are called verbal because they head phrases that have the internal structure of file:///G|/htdocs/Joe/3WEB_OK_Passives_and_Perfects_(POST).htm (16 of 21) [5/11/2003 5:08:06 PM] PASSIVE SYNTACTIC STRUCTURES VPs. An active VP is identical to a passive VP except that the latter contains one trace replacing a "passivized" DP. Other DP complements are allowed, as in (38). Simply restricting landing sites of NP movement to "theta-bar" positions provides no principled reason why seem, appear, become, etc. are not possible "auxiliaries" in the verbal passive. In the present account this follows from their not being inserted in PF. 9. Perfect Syntactic Structures A consequence of the Headedness Corollary is that the English present participle suffix [A –ing ] is unspecified in syntax for Φ agreement features. In this, it is an unusual adjective. A question arises, are other adjectival complements without Φ agreement features in syntax independently attested? In fact, Germanic and Romance languages, in which adjectives overtly agree, do have idiomatic combinations of common verbs with non-agreeing (invariant) adjective complements. In any contexts without the verbs that define the idioms, the adjectives in (44)-(45) agree normally. (44) French: avoir chaud ‘be hot’, avoir beau, ‘do in vain’, tenir bon ‘hold on’, peser lourd ‘weigh heavy’, chanter faux ‘sing off key’, etc. (45) German: recht haben ‘be (in the) right’ Thus, a Germanic and Romance lexicon may specify that the usually agreeing adjectival suffix -en (the right hand head of a participle) lacks agreement in complements of grammatical avoir, haben, etc. entirely parallel to the idioms in (44)(45). I claim this is the genesis of the composed past. Consider again for example Spanish. Its composed past combines a grammatical auxiliary haber, historically from a form meaning have, with a non-agreeing passive participle suffix V–do. I propose the same analysis for English have + V-en, whose A agreement is never overt. The reasoning here implies that any overt AR agreement of the composed past –en with an object, as in some Romance systems, is not in the syntax. Such agreement features are rather added in PF. I conclude that the structure in syntax of an active passive participle is [VP [V Ø ] [A V [A, -Φ Ø ] ] ], where have and –en are respectively inserted under the two empty nodes in PF. By Lexical Headedness (27), it correctly follows that the bold V in this structure is the head that selects and case-marks complement phrases in VP. This head is also subject to selection from outside VP. file:///G|/htdocs/Joe/3WEB_OK_Passives_and_Perfects_(POST).htm (17 of 21) [5/11/2003 5:08:06 PM] PASSIVE SYNTACTIC STRUCTURES 10. Lexical notation for cross-linguistic variation in impersonal passives Other Germanic languages have variations on a participle-based passive (Åfarli, 1989; Baker, Johnson and Roberts, 1989). In Norwegian "in situ transitive passives" no object DP moves. (46) Det vart gitt den såra soldaten ein medalje. (Norwegian) *{ It/ There } was given the wounded soldier a medal. There also exist "intransitive impersonal passives" with no object DP (German, Norwegian): (47) a. Es wurde bis spät in die Nacht getrunken. (German) it was until late in the night drunk *{ It/ There } was drunk until late in the night. b. Det vart gestikulert. (Norwegian) *{ It/ There } was gesticulated. It is of interest that English counterparts to both types are ungrammatical. (48) In all 3 systems, including English: -en alternatively realizes Φ features that occur as canonical features on an object DP. In unmarked AR, this DP must be empty, i.e., it is a trace. As shown in (46), the Norwegian counterpart of -en can also alternatively realize Φ features whose canonically realized source DP is lexical. This is a marked instance of AR. I propose to list alternatively realized features that may co-occur with (or "double") an overt source in this fashion with a lexical notation of underlining. An English noun plural (49a) is a simple example. In contrast, an English past tense, which never doubles [I, PAST], is unmarked AR (49b). (49) a. -s, N, +N___, PLUR those three boys; *those three boy b. -ed, V, +V___, PAST *could/ did cleaned; could/ did clean The interpretation of this underline is: whenever PLUR is canonically present on a D or NUM sister of some N projection, then the plural suffix must also appear on N, whether or not the node where PLUR occurs canonically is empty. More generally, this type of inflection is called agreement. (50) then expresses the fact that a Norwegian passive participle can have an overt direct object. file:///G|/htdocs/Joe/3WEB_OK_Passives_and_Perfects_(POST).htm (18 of 21) [5/11/2003 5:08:06 PM] PASSIVE SYNTACTIC STRUCTURES (50) Norwegian Passive Participle (tentative): en, (A), +V , Φ features If the source DP for the Φ features is empty, the results are as in section 9 for English; this DP must be the co-indexed trace of the subject DP. If the source DP is lexicalized, this object must still have the same index as the subject, since -en remains a predicate adjective agreeing with the subject. But when a subject that lacks a theta role is co-indexed with an overt object DP, then universal grammar must force this subject to be an expletive. This correctly yields in situ transitives as in (46). Finally, consider the intransitive impersonal passives of (47). (51) In German and Norwegian, -en can appear even when it alternatively realizes no Φ features, provided there is no object DP; the theta-bar subject can only be an expletive. To express this possibility, I propose to modify (39) by parenthesizing Φ features, which yields (52) for German and (53) for Norwegian. (52) German Passive participle: -en, (A), +V , (Φ features) (53) Norwegian Passive participle: -en, (A), +V___, (Φ features) I define parenthesized AR features to indicate that both the AR features and their canonically located source features, taken together, are optionally absent. That is, V may be intransitive. Under these lexical formalisms, German and Norwegian reproduce the English system with the addition of allowing passive intransitive verbs. The absence of an object and hence Φ features on a passive participle means that the theta-bar DP subject of a passive intransitive verb can neither be indexed with any other DP, nor can it receive a theta-role independently, as shown in section 8.1. So this DP must be an expletive and the verb impersonal, exactly as in (47). Since all verbs require a subject by the Extended Projection Principle (22), the subjects in (46) and (47) must be realized elsewhere, as a by-phrase or a PRO, just as with personal passives. The different systems of English, German, and Norwegian thus result from minimally different lexical items of absolutely standard format: (39), (52) and (53) respectively. file:///G|/htdocs/Joe/3WEB_OK_Passives_and_Perfects_(POST).htm (19 of 21) [5/11/2003 5:08:06 PM] PASSIVE SYNTACTIC STRUCTURES ReferencesÅfarli, Tor. 1989. ‘Passive in Norwegian and in English.’ Linguistic Inquiry 20. 101-108.Baker, Mark, Kyle Johnson and Ian Roberts. 1989. ‘Passive Arguments Raised.’ Linguistic Inquiry20. 219-252.Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton.---1965. Aspects of the Theory of Syntax. Cambridge: MIT Press.---1970. ‘Remarks on Nominalizations.’ In his Studies on Semantics in Generative Grammar. TheHague: Mouton.---1981. Lectures on Government and Binding. Dordrecht: Foris---1995. The Minimalist Program. Cambridge: MIT Press.Emonds, Joseph. 1985. A Unified Theory of Syntactic Categories. Dordrecht: Foris.---1987. ‘The Invisible Category Principle.’ Linguistic Inquiry 18. 613-631.---1991. ‘The Autonomy of the (Syntactic) Lexicon and Syntax.’ In Interdisciplinary Approaches toLanguage: Essays in Honor of S.-Y. Kuroda, C. Georgopoulos and R. Ishihara, eds. Dordrecht:Kluwer Academic Press.---2000. Lexicon and Grammar: the English Syntacticon. Berlin: Mouton de Gruyter.---2001. ‘The Flat Structure Economy of Semi-Lexical Heads.’ In Semi-Lexical Categories: TheFunction of Content Words and the Content of Function Words, N. Corver and H. van Riemsdijk, eds.Berlin: Mouton de Gruyter.---2002. ‘Formatting Lexical Entries: Interface Optionality and Zero.’ Theoretical and AppliedLinguistics at Kobe Shoin 5. 1-22, 2002.Fassi-Fehri, Abdulkader. 1993. Issues in the Structure of Arabic Clauses and Words. Dordrecht:Kluwer Academic Press.Lefebvre, Claire. 1988. ‘Past Participle Agreement in French: Agreement = Case.’ In Advances inRomance Linguistics, D. Birdsong and J. P. Montreuil, eds. Montreal: Publications in LanguageSciences.Harris, Zellig. 1957. ‘Cooccurrence and transformations in linguistic structure.’ Language 33. 28-340.Levin, Beth and Malka Rappaport. 1986. ‘The Formation of Adjectival Passives.’ Linguistic Inquiry17. 623-661.Lieber, Rochelle. 1980. On the Organization of the Lexicon. Doctoral dissertation. MassachusettsInstitute of Technology, Cambridge.McA'Nulty, Judith. 1983. ‘Moving Features of [ e ].’ Paper given at the Southern CaliforniaConference on Romance Linguistics. [Revised in Binding in Romance, A.-M. di Sciullo and A.Rochette, eds. Ottawa: Canadian Linguistic Association.]Newmeyer, Frederick. 1980. Linguistic Theory in America. Chicago: University of Chicago Press.Riemsdijk, Henk van and Edwin Williams. 1981. ‘NP-structure.’ The Linguistic Review 1. 171-217.Siegel, Dorothy. 1973. ‘Nonsources of Unpassives.’ In Syntax and Semantics 2, J. Kimball, ed. NewYork: Seminar Press.Sobin, Nicholas. 1985. ‘Case Assignment in Ukrainian Morphological Passive Constructions.’Linguistic Inquiry 16. 649-662.Wasow, Thomas. 1977. ‘Transformations and the Lexicon.’ In Formal Syntax, P. Culicover, T.Wasow and A. Akmajian, eds. New York: Academic Press. file:///G|/htdocs/Joe/3WEB_OK_Passives_and_Perfects_(POST).htm (20 of 21) [5/11/2003 5:08:06 PM] PASSIVE SYNTACTIC STRUCTURES [1] “Lexicalized” includes phonologically null allomorphs of lexical items, or being properly co-indexed withlexical material (e.g., under the structural conditions appropriate to traces, ellipsis, etc.).[2] More accurately, Fassi-Fehri (1990) reserves the term verbal passive for an Arabic construction more akinto Latin synthetic passives, and refers to the Arabic counterpart to an English verbal passive as an adjectival passive,because it is "internally verbal but externally adjectival."[3] Government and binding accounts have typically indicated that the passive “auxiliary” must not assign atheta role to its subject, i.e. NP movement is to a “theta-bar position.” But many verbs (seem, appear, happen, etc.)with “theta-bar” subjects still cannot be the basis of a verbal passive. In the end, such accounts simply stipulate that beand get take passive complements and other verbs do not.[4] Chomsky’s original terms are simply semantic and syntactic respectively. People often interpret theseterms as somehow implying that syntactic features don’t contribute to meaning. His original discussion makes clearthat syntactic categories and features are central to meaning.[5] The model for this is Chomsky’s (1957) analysis, which inserts uninterpreted auxiliary do aftertransformations (“do-support”), in line with (32). Such insertion is distinct from main verb do, presumably insertedlike other V prior to transformations. As Newmeyer (1980) observes, the explanatory breadth of this analysis won theday for generative grammar. We should extend rather than exclude a priori the very analysis that made the generativeenterprise successful.[6] I assume a fairly standard model of grammatical derivation (cf. Chomsky, 1995) in which underlyingphrasal domains are projected from lexical heads and then transformationally processed so as to obtain arepresentation of all aspects of grammatical meaning, called Logical Form (“LF”).At a certain derivational point called “Spell Out” (of the domain in question), the model assumes that furtheroperations affect either a structure’s pronunciation or interpretation but not both. Transformations prior to Spell Outare called “in syntax” and those after Spell Out are said to be either “in LF” (on the LF branch) or “in PF” (on the PFbranch).[7] Levin and Rappaport (sect. 4.1) assume, but in no way demonstrate, that adjectival passives are formed inthe lexicon. Moreover: "In accordance with the program initiated in Chomsky (1981), we eschew the explicit use ofsubcategorization frames in lexical entries as a representation of the complement structure of verbs." Rejecting asyntactic account based on subcategorization makes impossible a natural expression of their result, namely, thatadjectival passives lack precisely that overt DP licensed by the frame +____DP.[8] The attempt to motivate NP-movement in verbal passives by lack of case led Chomsky (1981) to assignthem to a special, "unspecified for +N" neutralized category. The latter is at bottom an equivocation: in Chomsky(1981, 50), participles that are [+V, unspecified for N] don't assign case, whereas on p. 51, it is suggested that of is [-V, unspecified for N] and does. Hence, an ad hoc "neutralized category" (unspecified for N) of highly defectivedistribution leads not to a definition of a natural class of case assigners (somehow related to [-N]) but to its opposite.[9] The object trace in a participle-based passive cannot arise independently of co-indexing of the participlewith the subject. That is, movement is not the cause of co-indexing, but vice-versa. This suggests that a landing sitefor “A-movement” must have a kind of scope marker, akin to the scope markers of “A-bar landing sites” in vanRiemsdijk and Williams’ (1981) NP Structure model.[10] This excludes the possibility of optional AR (i.e., optional inflection) in the presence of their canonicalsource features. Such an exclusion corresponds to the general empirical fact of language that inflection, if present atall, is typically obligatory. Only canonical features can be optional. file:///G|/htdocs/Joe/3WEB_OK_Passives_and_Perfects_(POST).htm (21 of 21) [5/11/2003 5:08:06 PM]