Modern Grammars of Case anderson phần 9 ppt

there being ‘tangling’ of dependency arcs. That is, with ‘raising-to object’, both linearity requirements are met, in the position dictated by John’s simplex modiWer relation to read. In many word-order types such a compatible position cannot be provided: it depends on the shared argument in question being able to occur between and adjacent (barring circumstantials) to the two verb forms (here expected and to read). This accounts at least in part for the restricted distribution of ‘raising-to-object’. Horn (1985) observes that this condition of juxtaposition of the shared element to both heads cannot be met in VSO structures, for instance—which, of course, is not to say that ‘raising-to-object’ cannot occur in languages where the unmarked order is VSO). The capacity to avoid ‘tangling’ avoids the need for reference to the ectopicity condition, and thus for interpretation of raisees as simplex mod- iWers; in (16b) John behaves as any other dependent, in this case of expected. Status of raisees as a simplex modiWer is contingent on ectopicity. The restriction of unalloyed speciWer status for free absolutives to those dependent on {P} raises interesting questions, questions that, however, lead us away from our ongoing story. I therefore mention only one. The free absolu- tive is (derived) speciWer to a functional category, {P}. Both heads in (8b, c) are also functional, respectively functor and comparator. On the other hand, it is not obvious what might be the identity of the speciWers to be associated with the lexical categories nouns and verbs (a continuing point of disagree- ment). Of the lexical categories only adjective commonly has a speciWer attributed to it, illustrated by (8a). Could it be that the positive of gradient adjectives, as well as the comparative, involves a comparator, in the former case absorbed rather than independent (more)asin(8c)? This comparator, like the overt one in (8c), is absent with non-gradient adjectives. In that case, we could associate speciWcation uniquely with functional categories. What is relevant to us at this point, however, is the recognition that it is the absolutive that may be free, or unsubcategorized-for, and thus the ‘host’ for raising, rather than (only) subjects. This eliminates, by virtue of the ‘sharing’ property of the free absolutive, the need for recourse to ‘exceptional case- marking’ and the like in accounting for the syntax of such ‘raising-to-object’ sentences. Nor, of course, are we led to a proliferation of ‘functional categor- ies’ parasitic upon {P} (or ‘InX’). 11.2.5 The category of the infinitive The representations in (14) and (16) lack any indication of the category status of the inWnitival word to: the categorization given is that for the following verb. This now requires our attention, even though it too is not central to our 344 Modern Grammars of Case main concerns here. I present one analysis that is compatible with the framework we have been adopting. Anderson (2001a) again follows Anderson (1997)—and ultimately Pullum (1982)—in regarding this to as a verb which, like seem, takes another verb as its only argument but diVers from other verbs in being an exception to Wniteness formation: it is always non-Wnite, and by virtue of its categoriza- tion is transparent to the requirements of a governing predicator. It is thus a kind of necessarily non-Wnite ‘auxiliary’. Both the to and the non-Wnite auxiliary are transparent to the demands of the upper (‘control’) verb in Roger persuaded Bertram to be patient. In particular, the expectation that the non-auxiliary verb subordinate to persuade involves some kind of ‘agency’ is not interfered with by the intervening superordinate to (as well as the be). Anderson (2001a) assumes that to is introduced by a ‘syntactic redundancy’ that applies above all {P;N} that depend on another (non-modal, but possibly non-verbal) predicator—i.e. on P, any combination with P (except simple {P}). Let us refer to this as ‘inWnitivization’: (17) Infinitivization P P | {P;N/{P;N}}* | {P;N } {P;N} (where ‘{P;N/{P;N}} à ’ ¼ inWnitival to, with the asterisk indicating its incap- acity for Wniteness formation). This is, according to Anderson (2001a), the unmarked redundancy involving dependent {P;N} in such circumstances: participles, for instance, involve a more complex speciWcation, as discussed in, for exmple, Anderson (1992: chs. 5, 7). In present terms, however, we can characterize what is going on here more straightforwardly, and less particularistically, in terms of the mechanism of ‘periphrasis’ discussed in, for instance, §8.2. Recall the role of the be auxiliary in allowing the progressive form to satisfy a valency and thus gain access to Wnite predications (among other things). The inWnitival to can be seen as performing a similar function. In most circumstances in English the ‘bare’ form of the verb has to be accompanied by an element, realized as to, that allows it to be part of the valency of an upper verb. Suppose we do indeed characterize what I shall refer to as the ‘bare inWnitive’ as in (18a), i.e. as simply having the lexical categorization of non- operative verbs: Argument-Sharing I: Raising 345 (18) a. {P;N} b. InWnitive complementation {P;N/{P;N}} ,{P;N/{P;N/{P;N}}} c. To ¼ {P;N/{P;N}} Say ‘inWnitive-taking’ verbs (if we ignore other predicator types, for simplicity of exposition), are subcategorized in the lexicon as ‘/{P;N}’, as on the left of (18b). However, these normally undergo (18b) to give a derived representation on the left. Their valency will thus not be satisWed by the ‘bare inWnitive’ of any verb, which is simply ‘{P;N}’. To is the ‘go-between’ here: its categorization and subcategorization as in (18c) means that it, triv ially, satisWes the valency given on the right of (18b), but also that it has a valency itself that can be satisWed by the bare inWnitive. The construction as a whole is a ‘periphrastic inWnitive’. If this characterization is appropriate, the representation in (16), for in- stance, should be expanded as in (19)—where the speciWcation of expect shows the application of (18b), inWnitive complementation: (19) {P} | {{abs}} {P;N/{erg,loc}{P;N/{P;N}}} : {{erg,loc}} {{abs}} {P;N/{P;N}} | : {N} {{abs}} {P;N/{abs}{erg}} : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : {{erg}} {{abs}} | | {N} {N} : : Kate expected John to read Rasselas John is ‘raised’ twice, so that read shares its ergative argument with the free absolutives of to and the {P;N} component of expected, while the latter shares the argument of its {{erg ,loc}} with the free absolutive of the {P}. Again, the position of John with respect to expected is determined by its failure to undergo subject formation with respect to that verb. John is a dependent of both expected and to, as well as of read; and in the case of the former two there is no ‘tangling’ whatsoever. There is ‘tangling’ of dependency arcs and asso- ciation lines above to, but this ‘tangling’ involves the association attached to 346 Modern Grammars of Case the ‘transparent’ periphrastic inWnitive marker. And, crucially, John occupies simplex modiWer position in relation to to; thus, the ectopicity condition determines its position in relation to to and read. The modal in (5) is not accompanied by an overt inWnitive marker. As is familiar, this is also characteristic of verbs of perception (20) and some ‘direct causatives’ (21): (20) Bert saw/heard Kate leave (21) Bert made/let Kate leave These classes of verb are also exceptions to the requirement embodied in (18b), and they must be marked as such in the lexicon: (22) InWnitive complementation exceptions {P;N{ * (18b)}} As well as the verbs in (20) and (21), modals and periphrastic to are also such exceptions, of course: this seems to involve a generalization that inWnitive- taking ‘periphrasts’ are exceptions to (18b), and need not be included in their lexical entry. It can be argued (Anderson 2005d) that the exceptional lexical verbs at least form two semantically natural classes (though not all causative verbs are exceptional). For many speakers ought is a modal that in taking a to-inWnitive is an exception to the exception, i.e. an exception to (22). However, other speakers regularize the verbal (Ought she leave?) or make it non-operative (She didn’t ought to leave) or don’t use it. We can, on the basis of the preceding, associate the structure in (23) with causative sentences (21): (23) {P} | {{abs}} {P;N/{erg}{P;N{*(18b)}}} : {{erg}} {{abs}} {P;N/{abs, erg}} | : {N} {{abs,erg}} : | : {N} : : : : : : : : : : : : : : : : Bert made/let kate leave Argument-Sharing I: Raising 347 Make is marked as an exception to (18b), so retains its lexical speciWcation; it can therefore be satisWed by the bare inWnitive form. Kate Wlls the free absolutive of made, as well as the intransitive agent of leave. Whereas we can say in general that between a predicator and a dependent {P;N} there must appear an overt inWnitive marker, the classes of verbs that we have identiWed are exceptional in this regard. 11.3 Conclusion A consideration of the category of the inWnitive has taken us away somewhat from the main theme of this chapter, which concerns the characterization of raising and control and in particular the role of free absolutives in this. But this diversion has allowed us not just to X esh out the structure of the subordinate clauses that occur in raising constructions but also to extend a little further our elaboration of the lexical apparatus underlying syntax. Chapter 12 returns to the main theme, and is again based initially on the discussion in Anderson (2001a). The main conclusion to be drawn from the present chapter is the crucial role of free absolutives in the syntax of raising sentences. Their capacity to share an argument with subcategorized-for arguments in lower predications provides the licensing environment for representations that involve tangling, and thus avoid reference to ‘movement’ and ‘empty categories’. The next chapter looks at suggestions that this role can be extended to control structures. So that the same limited apparatus can allow for phenomena that have been regarded as involving ‘movement’ and phenomena that at one time were considered to involve deletion (cf, for example, Rosenbaum 1970) but more recently have been described in terms of a particular type of ‘empty’ category, diVerent from the ‘traces’ left by ‘movement’. These latter too are an unnecessary enrichment of syntactic theory. As suggested in §11.1, such ‘empty categories’ weaken the role of the sign in regulating the establishment of units: there are units of content that are uncorrelated. Given absence of grounding, they are also units of content form (syntax) that involve radical autonomy, rather than merely categorial autonomy. In this respect, as I have indicated, their adoption pushes the notion of autonomy beyond anything previously contemplated—and, I sug- gest, beyond anything that warrants being contemplated. 348 Modern Grammars of Case 12 Argument-Sharing II: Control We now turn to control, against the background of the discussion of raising in Chapter 11. Anderson (1977:§3.4; 1992:§4.4) and Bo ¨ hm (1982:§3.1), indeed, suggest that the syntax of ‘obligatory control’ verbs such as those in (11.1b), renumbered for convenience simply (1b), and in (2b) involves ‘raising’: (1) a. John seemed to like Rasselas b. John tried to read Rasselas (2) a. Roger believes John to be a fool b. Roger persuaded John to escape In each instance the John element is an argument of the escape verb—and indeed is its designated ‘subject’—which is also, by virtue of raising, an argument of the Wnite verb. Let us see how this can be reconciled with the (motivations for the) traditional distinction drawn between the constructions in (1b) and (2b), on the one hand, and the typical raising constructions illus- trated by (1a) and (2a). 12.1 The role of the absolutive Traditional raising , interpreted as in Chapter 11, is to the absolutive argument position,whichis subjectinintransitives(1a),objectintransitives(2a):thesubject functor phraseof the inWnitive is associated, via raising, with the absolutive of the raising verb. In control structures, on the other hand, the ‘subject’ of the lower predicator has commonly been considered to be related in some way to an argument of the upper that, in ‘case grammar’ terms, bears a relation other than absolutive: in the case of (1b), an agentive, in (2b) an experiencer. And this relation can impose restrictions on its argument, of course. So that (3a) and (3b), for instance, require heavy contextualization for viable inter- pretations to be achieved: (3) a. Bertrand tried to be a younger son b. Roger persuaded Bertrand to be a younger son This is so even though in both cases Bertrand is quite acceptable as a ‘subject’ of the embedded predication. ‘Traditionally’ (at least since the time of McCawley (1967), Rosenbaum (1967)), the distinction between the (a) and (b) examples in (1) and (2) has not been attributed to variants of raising. The two arguments related by the control relation in the (b) examples are both subcategorized for. Thus the analysis of these has not been taken to be parallel to that appropriate for the (a) examples. The former have been associated with the occurrence of two coindexed positions—one in the upper predication, the other corresponding to the ‘subject’ of the lower—either in both the (b) examples (for example Rosenbaum 1967) or even only in (1b), with the (2b) instance showing in this case neither distinct positions nor raising. The superWcially rather similar constructions in the (a) and (b) examples thus involve rather diVerent syntactic descriptions. This requires substantial motivation, motivation not immediately apparent, except on the basis of severely theory-internal assump- tions. What motivations there are for diVerentiating between these construc- tions are semantic, with minimal syntactic consequences. Such a traditional v iew also does not suggest any account of a salient distinction among control structures. The verbs in the above (b) examples speciWcally require their coindexed argument—the ‘controller’ (whether agentive or experiencer)—to be matched with an argument in an embedded predication which is also agentive, so that (4a) is strange (unless construed, say, as an elliptical reference to some experiment whose desired result was a hangover): (4)a. * Muriel tried to suVer a hangover b. Muriel expected to suVer a hangover c. Muriel expected to be famous/late/in London/a grandmother The experiencer control verb in (4b) (also a raising verb, as exempliWed in §11.2.4), however, imposes no such requirement. This is shown particularly by (4c), wherein the subordinate predication doesn’t even involve an experien- cer. Expect is compatible with any subordinate predication in which the ‘null’ ‘coindexee’ is human (given pragmatic constraints on what one can plausibly expect to be the case with Muriel). The extent of control is variable, in accordance with the semantics of the control verb. In this section we look at an analysis of (obligatory) control constructions (based on Anderson (2001a)) which is also (like that suggested for raising) non-mutative and in which the free absolutive also plays a crucial role. We start, in §12.1.1, with a consideration of experiencer control, beginning with verbs (like expect) that allow either a raising or a control derivation. It is 350 Modern Grammars of Case suggested that a free absolutive may be introduced either independently or in association with a subcategorized-for semantic role: the latter situation characterizes control. §12.1.2 explores the relationship of experiencer and agentive control to the agentive controller condition underlying the strangeness of (31a), as well as making note of a distinction between control structures that involve a mediated inWnitive (contracting a labelled semantic role) and a direct inWnitive, as in the examples discussed above. And §12.1.3 ascribes some of the traditional problems in specifying selection of controller to the internal (lexical) argument structure of diVerent control verbs, particularly those which convey an illocution or perlocution. The discussion of the suggestions of Anderson (2001a) concludes in §12.1.4 with an overview. No attempt is made, any more than in Anderson (2001a), to survey the large literature on control (large even in relation to English): see, for a sample, Sag and Pollard (1991), Pollard and Sag (1994: ch. 7); Panther (1994); Jack- endoV and Culicover (2003); and see also, for example, (contributions to) Wilkins (1988c), and Anderson (1992:§§3.5, 4.4), for discussion of some earlier work, including other studies invoking semantic relations. But, though the focus is on the role of free absolutive, I attempt here to present what Anderson (2001a) regarded as an account of some of the basic distinctions involved in ‘obligatory’ or ‘unique’ control. JackendoV and Culicover (2003) oVer a diVerent, much fuller account of diVerent control types that also assumes a semantic-relational basis for control. We cannot establish within the conWnes of the present work to what extent all of their description, couched in a highly stratiWed framework, is compatible with the proposal oVered by Anderson (2001a) described in what follows, with its more restricted focus. However, some extension of his proposal is o Vered in §12.1.3, and particularly in §12.2, concerned with control by locatives. Some recent accounts within a ‘minimalist’ framework have also proposed a more uniWed approach to raising and control. Hornstein (1999), for instance, argues that obligatory control structures, as well as raising struc- tures, are formed by movement. He too argues (1999: 71) that, as traditionally conceived, ‘the distinction between raising and control multiplies the inven- tory of empty categories’. Hornstein’s account involves an abandonment of the ban on movement into ‘ u-positions’. Likewise, the proposal concerning control structures oVered in Anderson (2001a) requires sharing of an argu- ment which has a semantic role in two clauses. Regarding control as involving movement is argued against by JackendoV and Culicover (2003); however, from the perspective of Anderson (2001a), neither does raising involve movement. Argument-Sharing II: Control 351 On Anderson’s account, the selection of controller is made on semantic- relational grounds: although the shared argument can, in its function as ‘controllee’, be characterized as (potential) ‘subject’ of the lower clause (and thus, in any one of a variety of neutralized semantic relations to its verb), the controller is speciWed by its semantic relation (agent, experiencer)—see too, for example, Pollard and Sag (1994:§7.2). This is in accord with the (deriva- tionally based earlier) ‘case grammar’ hypothesis that grammatical relations are derived cyclically, so not available in the cyclic clause. In terms of a non- derivational framework, the control relationship illustrates the role of (potential) subject as designated identiWer of ectopic shared arguments: we know that the shared argument is (potential) subject of the lower clause, despite its ectopic placement. In the upper clause, the shared argument bears (whatever else) an absolutive relation to its verb, whether control or raising is involved. This absolutive is not part of the subcategorization frame of the control/raising verb; it is free, and present by virtue of a universal requirement that every predication contain an absolutive. Control thus diV ers from raising, on Anderson’s (2001a) account, in the free absolutive’s being associated with a subcategorized-for semantic relation ({erg}, in the case of (1b), for instance). This argument bears both a free absolutive and a lexically speciWed relation or relations. The discussion of how this might be more precisely formulated begins with the familiar observation of an overlap in the classes of raising and control verbs. 12.1.1 Raising versus control Compare with the raising structures in (5) the control variants in (6), with the same verbs: (5) a. Bert expected Kate to have (had) a good time b. Bert remembered Kate to have had a good time (6) a. Bert expected to have (had) a good time b. Bert remembered to have (had) a good time In (5), on the analysis of §11.2.4, Kate realizes the argument shared between the subject of the lower verb and the free absolutive of the upper predication; Bert is a subcategorized-for experiencer in the latter. In (6) Bert corresponds to both this experiencer and the subject of the lower predication, both of them subcategorized-for. Otherwise, we have structures parallel in syntax and interpretation. Anderson (2001a) concedes that with various verbs there are, to be sure, constraints involving temporal reference and other factors limiting 352 Modern Grammars of Case these parallels. This is illustrated by the necessarily perfect form of the subordinate in (5b). Thus too the control structure (7a), for example, is ‘paralleled’ in temporal proWle not by (7b) but by the lexical causative in (7c): (7) a. Bert remembered to take his duck b. * Bert remembered Kate to take her duck c. Ber t reminded Kate to take her duck d. Bert remembered Kate to have taken her duck (7c), however, obviously adds a dimension absent from (7a) and (5, 6). All of these last, however, are again suYciently like each other in interpretation and syntax as to prompt a consideration of whether an analogous analysis is in order. The diVerences in temporal proWle associated with the main verbs in (5) and (6), illustrated further by the viability of ( 7 d) as against (7b), whereas both the parallels with expect are viable, relate to what a person can reasonably be conceived of as remembering to be necessary and possible future under- takings. And we can relate this, as well as to the meaning of the verbs, also to the diVerence between control and raising structures proposed by Anderson (2001a), as will emerge in what follows. And this should not be allowed to obscure the basic similarities between these structures. We associated a free absolutive with the raising construction exempliWed in (5), on the assumption that the subcategorized-for arguments of the raising verbs in (5) are {erg,loc} and {P;N}; these verbs are not subcategorized for absolutive. In (6), too, there is no obvious candidate for being a subcategor- ized-for absolutive argument of expected/ remembered; these verbs too take an experiencer and a verb. Suppose, then, all these verbs are uniformly—in both pairs—lacking in a valency involving absolutive. With the sentences in (5) the absolutive introduced as a consequence of the universality of absolutive requirement is Wlled by the hierarchically highest argument of the inWnitive: Kate in (5) is shared by the free absolutive and the lower subject relations. Suppose, further, that this also occurs in the building of structure associated with the (6) sentences: a free absolutive is introduced (in the absence of a subcategorized-for absolutive) which shares its argument with the lower subject. But in (6) there is no overt argument present distinct from Bert . This would lead us to the conclusion that Bert in (6) is simultan- eously the {{erg,loc}} argument of the Wnite verb and (by default) the Wlled absolutive, i.e. the free absolutive argument Wlled by the subject argument of the inWnitive. Argument-Sharing II: Control 353 [...]... to show that the apparatus of case grammar’ can allow, without structural mutation, for further phenomena that have been seen as necessitating appeal to transformations and ‘empty’ categories, notably those involving control The analysis of raising of Chapter 11 is basically that of Anderson ( 199 7: §3.6), while the raising account of control phenomena (§12.1) is drawn from Anderson (2001a) Both accounts... structures, takes us on to the topic of the next subsection We turn now to what appears to be a class of causatives that, unlike the simple raising verbs of (20), involve control, and which introduce a traditional problem with the description of control What follows modiWes the analysis of Anderson (2001a) in the light of Anderson (2005a) 12.1.3 Causatives and control Anderson (2001a) recalls that apparently... usually the victim of control—the ‘origin’ of the shared argument—is the agentive argument of this predicator, as illustrated by the typical examples in (21): 360 Modern Grammars of Case (21) a Bert managed to leave town in time b Kate decided to read Rasselas (22) *Bill managed/decided to feel unwell Interpretation of (22) would require some play-acting, or some other infringement of normal expectations... attributed to a property, rather— in the case of the lock in (7.15a), of ‘having been easy to pick’ This all makes sense if in this variety of control it is a simple locative (locating the property) that has the free absolutive associated with it We can associate with (7.15a) the representation in (36) (which, as usual, ignores much that is irrelevant): 368 Modern Grammars of Case (36) {P} : {{abs}} : : : {{(loc),abs}}... mugged 372 Modern Grammars of Case Be does not otherwise function as a locative-subject verb; but then it isn’t one here either, but merely a periphrasis allowing the goal–subject conWguration in (45) to be Wnite The locative-control stage accounts for why in the case of the passive raising appears (unusually) to take the next highest argument of the base verb However, the relative lack of transparency... return to this in a moment In 9. 3.2 I distinguished syntactic causatives like that in (9. 63) from the periphrastic causative in (9. 77), as well as, of course, from morphological and lexical causatives: (9. 63) John caused Bill to die (9. 77) Il fait fondre la neige it makes melt:INF the snow (‘It makes the snow melt’) (9. 77) was interpreted as undergoing the Wrst part of the lexical causativization... Control (56) 377 a Mehmet Hasan-a bavul- u ac- tIr- dI ¸ Mehmet Hasan-DAT suitcase-ACC open-CAUS-PST (‘Mehmet had Hasan open the suitcase’) b Bavul (Mehmet tarafIndan) Hasan-a ac- tIr- Il- dI ¸ suitcase (Mehmet by) Hasan-DAT open-CAUS-PASS-PST (‘The suitcase was caused (by Mehmet) to be opened by Hasan’) (Aissen 197 9: 15; Rosen 199 0: 225) This pattern doesn’t always correlate with morphological causativization,... also Anderson ( 199 2: §4.4) and Pollard and Sag ( 199 4: ch 7) for some suggestions along these lines Rather, I merely want here to indicate factors that might in principle account for the apparent aberrancy of complex predicators with respect to the generalizations established for core control predicators in the previous subsections We return in §12.2.3, however, to a reconsideration of the analysis of. .. abbreviated the associative structure ‘{erg,loc} {abs}’ of (8b) as ‘{{(erg,loc),abs}}’ Bert realizes successively, moving upwards, the {{erg,loc}} argument of have and the free absolutive of to, the free absolutive of expected, together with the {{erg,loc}} of expected (by virtue of its association with the free absolutive), and the free absolutive of the Wniteness element {P}; and the structure-building... hierarchization too looks like a sign of routinization 370 Modern Grammars of Case Any element of ‘aVectedness’ associated with the passive subject is stronger in the get-passive: (43) a Beryl got mugged b Bob got sacked Get basically seems to be a goal-subject verb (as well as agentives based on this): (44) a b Bob got a present Bob got the sack I suggest that this is also the case with the passives in (43): . and Pollard ( 199 1), Pollard and Sag ( 199 4: ch. 7); Panther ( 199 4); Jack- endoV and Culicover (2003); and see also, for example, (contributions to) Wilkins ( 198 8c), and Anderson ( 199 2:§§3.5, 4.4),. subor- dinate predication of the argument they share with it. Thus, Anderson (2001a) takes the syntactic structure of (6a) to be as in (9) : 354 Modern Grammars of Case (9) {P} | {{abs}} {P;N/{erg,loc}{P;N}} . to our 344 Modern Grammars of Case main concerns here. I present one analysis that is compatible with the framework we have been adopting. Anderson (2001a) again follows Anderson ( 199 7)—and ultimately