Download Stacked passives in Turkish and more Schemes and Mind Maps Voice in PDF only on Docsity!
Stacked passives in Turkish
Andrew Murphy*
Abstract
This paper deals with ‘stacked passives’ in Turkish, an impersonal passive construction with two occurrences of passive morphology and two instances of argument reduction. The aim will be to adequately capture the fact that each instance of passivization seems to be mirrored by a morphological reflex on the verb. I will adopt the theory of passivization in Müller (óþÕ¦), who assumes that passivization involves merging and subsequently removing the external argument from the structure. The analysis of double passivization will involve assuming two Voice projections above vP, each headed by a passive suffix. This approach will allow us to capture the relevant data for Turkish and it will also be shown how this approach extends to another case of double passivization in Lithuanian as well as antipassive constructions.
Õ. Introduction
In this paper, I propose an analysis of stacked passives, that is passives with two occurrences of passive morphology, which assumes that each passive morpheme corresponds to the head of a Voice projection. Furthermore, I argue that argument reduction in passivization is triggered by each of these heads and is carried out by an operation I call Slice (proposed by Müller óþÕ¦), which removes arguments from the structure. Accordingly, two Voice projections in SP entails that two instances of argument reduction must take place. This analysis will capture two main properties of stacked passives as identified by Postal (ÕÉä): (i) They are only possible with transitive verbs, (ii) both internal and external arguments are understood as implicit. Since two Voice projections entail two argument reduction operations, only transitive verbs will allow for all relevant features to be checked. Furthermore, the fact that these syntactic
- (^) Thanks go to Gereon Müller for the inspiration for this paper as well as the participants of the seminar Passives in summer semester óþÕ¦ at the Universität Leipzig. Particular thanks go to Murat Özgen for providing judgements for Turkish and Doreen Georgi and Sandhya Sundaresan for helpful comments. Any errors in exposition or interpretation are entirely my own.
Topics at InfL, óäì–ìþ¦ A. Assmann, S. Bank, D. Georgi, T. Klein, P. Weisser & E. Zimmermann (eds.) L¶«±«hu A§fu±« Bu§h±u Éó, Universität Leipzig óþÕ¦
óä¦ Andrew Murphy
arguments were present in the structure at some point in the derivation, will allow their traces to be existentially bound and therefore derive the implicit argument property of passives in (ii). In Section ó, I will present the data pertaining to passives and stacked passives in Turkish. Section ì will discuss various approaches to argument reduction and to what extend they can be applied to stacked passives. Section ¦ presents an analysis of stacked passives based on the operation Slice. Section ¢ will discuss some further implications of this approach to the passive such as accusative absorption as well how this analysis extends to argument demotion in double passives in Lithuanian and antipassive constructions with argument reduction.
ó. The data
Passivization is typically analysed as suppression of the ‘most prominent’ (external) argument coupled with promotion of the internal argument to the subject. In this paper, I will focus on a particular passive construction in Turkish involving two occurrences of passive morphology as well as what looks like two instances of passivization, i.e. reduction of both the internal and external argument. In the following, I use the term ‘stacked passive’ to refer to double passivization coupled with two instances of a passive suffix. Thus, the passive suffixes are ‘stacked’ on top of each other The term ‘double passivization’ is more general and used for cases of dual argument reduction without two morphological reflexes of passivization. Instances of ‘stacked’ or double passives have received little attention in the literature for perhaps two reasons: (i) They are typologically rare (Kiparsky óþÕì attributes this to a ‘morphological bottleneck’), (ii) They do not exist in English and other languages where the passive has been particularly well studied. Nevertheless, instances of double passives have been reported in Turkish (Özkaragöz ÕÉä), Kazakh (Şahan Güney óþþä), Lithuanian (Timberlake ÕÉó, Keenan and Timberlake ÕÉ¢), Sanskrit (Ostler ÕÉßÉ) and Irish (Nerbonne ÕÉó). Its existence has important implications for any theory of the passive since a theory of the passive designed to handle one instance of argument reduction should be able to be extended to account for instances of dual argument reduction. As will be shown, this is not always straightforward for many of the analyses we will encounter.
óää Andrew Murphy
Note that in each example we have a transitive verb and therefore two instances of argument reduction. Furthermore, each sentence has an implied internal and external argument, which follows from the fact that only transitive verbs are possible in the construction. Finally, all above examples exhibit aorist tense. These are identified as the main three characteristics of stacked passives by Postal (ÕÉä) and can summarized as follows:
(ä) Characteristics of stacked passives in Turkish (Postal ÕÉä):
a. Only passives of transitive verbs are possible. b. Both arguments must be understood as implied arguments. c. They are only possible with aorist tense.
In this paper, I will focus on providing an explanation of the first two character- istics as the third is perhaps largely semantically motivated.Õ^ Evidence for (äa,b) comes from the fact that stacked passives are not possible with non-transitive verbs such as unergatives (ß) and unaccusatives ():
(ß) *Bura-da here-h
koş-ul-un-ur. run-£Z««-£Z««-Z§ Int.‘There is running here.’ (Unergative)
() *Okyanus-ta ocean-h
bat-ıl-ın-ır sink-£Z««-£Z««-Z§ Int.‘In this ocean, there is sinking.’ (Unaccusative)
ì. Previous approaches to argument reduction
A central characteristic of passive clauses is that often involve argument reduc- tionó^ Every theory of personal passives has to explain argument reduction and thus, an analysis should be applicable to instances of dual argument reduction.
Õ (^) Furthermore, it seems to be more of a strong tendency than an inviolable property of the construction (Göksel and Kerslake óþþ¢: Õìä). The fact that both arguments are implied may strongly lend itself to a generic interpretation and thus explain the use of the aorist. (Özkaragöz ÕÉä: ß) provides some examples with past tense marking rather than aorist. However, she also claims that these are not ‘genuine’ double passives as the passive marker can be used disambiguate cases where the passive marker -n is syncretic with the reflexive marker. ó (^) Or ‘argument demotion’, i.e. realization of an argument in an oblique case, but the focus of this paper is on argument reduction as in the Turkish cases. See Section ¢.Õ for a tentative analysis of argument demotion in Lithuanian, however.
Stacked passives in Turkish óäß
Ideally, one should simply be able to apply a passizivation operation twice (once to the active structure and again to the resulting personal passive) and arrive at double passive. In the following section, I will review the main approaches to argument reduction in the literature and assess how each analysis can be extended to double passivization. We will see that extending these analyses to the problem at hand is not always without problems.
ì.Õ. Lexical approaches Some of the earliest approaches to argument reduction in passive assumed that this process takes place in the lexicon (Chomsky ÕÉÕ, Bresnan ÕÉó, Jackendoff ÕÉß, Booij ÕÉÉó, Wunderlich ÕÉÉì). Although these theories obviously differ in the details of the frameworks they are couched in, the basic idea is the same: Passivization modifes the valency requirements/theta grid/argument structure of a verb to remove one of the arguments (namely, the external argument). If we consider a possible lexical entry for a transitive verb taking to two arguments in (Éa) and compare it with the passivized form in (Éb) , we see that we can derive the latter from the former using the rule in (Õþ).
(É) a. Active verb: V(x,y) b. Passive verb: V(x,-) (Õþ) Lexical passivization rule: V(x,y) → V(x,-)
For cases of double passivization, one could simply propose an additional rule removing both arguments rather than one.ì^ Such a rule would look as in (ÕÕ).
(ÕÕ) Double passivization rule: V(x,y) → V(-,-)
The question at this juncture is whether this approach can derive the properties in (ä). A rule such as the one in (ÕÕ) will only apply to transitive verbs and as such will derive this property, however, the second property (implied arguments) is more difficult. Lexical approaches would have to assume that there are no arguments in double passives. The lexical entry is modified before syntax and will thus bleed any combination of a verb and its arguments. As a result, one
ì (^) Another option would be to apply the passivization rule twice, although see Müller (óþÕì, to appear) for problems with this approach in certain frameworks.
Stacked passives in Turkish óäÉ
another technical problem: if the external argument is syntactically present, then it is unclear why it does not count as an intervening goal for movement to Spec-TP. In (Õó), the silent external argument is higher and thus Minimality considerations should block movement of the internal argument (Õì).¦
(Õì) (^) TP
T′
VoiceP
Voice′
VP
V tDPint
Voice
DP
pro
T
DPINT
Furthermore, the existence of a small pro does not receive any kind of inde- pendent motivation for passives in non pro-drop languages such as English (Wanner óþþÉ: Õ¦¢). Its postulation is only motivated a solution for this prob- lem in passives and is therefore an ad hoc solution to a technical problem. Furthermore, the question of how this pro argument is semantically-linked to a DP in a by-phrase is far from trivial and will certainly entail more than simple co-indexation (cf. Sternefeld ÕÉÉ¢).
ì.ì. Passive morphemes as arguments (Baker, Johnson & Roberts ÕÉÉ) Following Jaeggli (ÕÉä), Baker, Johnson and Roberts (ÕÉÉ) propose approach to the passives in the framework of Government & Binding, where the passive morpheme (-en in English) has argument status. I summarize the details of the approach and how it captures some important properties of passives, but I will
¦ (^) This is what Collins (óþþ¢) worked hard to avoid with his ‘smuggling’ analysis. Nevertheless, I will not discuss his approach here as it is essentially a pro approach: He assumes that the external argument in by-phrases is in the canonical Spec-vP position, however, in passives without a by-phrase he is forced to assume a pro argument.
óßþ Andrew Murphy
not recount all the details since much of the theoretical apparatus utilized by them is now obsolete. Under their approach, the passive morpheme is present in syntax and behaves like an NP argument in that it can be assigned θ-roles and case. Following assumptions in Chomsky (ÕÉÕ), the passive morpheme is base-generated in I and then assigned accusative case and the external argument θ-role. It is assumed that the passive morpheme behaves like a clitic syntactically. Thus, BJR propose a ‘downgrading’ operation where -£Z«« lowers onto the verb:
(Õ¦) Passivization in English (Baker, Johnson and Roberts ÕÉÉ): IP
I′
VP
V tÔ
-en V
I
t-en
NP
The assumption here is that accusative case is assigned to the passive morpheme under government and this ultimately forces movement of the internal argument to Spec-IP in order for it to receive case. The captures both case-driven movement of the internal argument to subject position and absorption of accusative case in passive clauses. Baker, Johnson and Roberts (ÕÉÉ: óìóf.) discuss stacked passives such as those discussed here in Turkish and those in Lithuanian. They claim that these can be captured by making different assumptions about the syntactic nature of the passive morpheme in such languages. Double passivization is ruled out in English as this would entail having two passive morphemes outside the VP yet only having one external θ-role to assign. They claim that the passive morpheme in languages such as Lithuanian and Turkish, which allow stacked passives, is actually an N element and not INFL. This element is then base generated directly in argument positions. This allows for the possibility of having two instances of passive morphology: one in subject position and the other in argument position. Their proposed derivation of stacked passives is given in (Õ¢):
óßó Andrew Murphy
This may work if one is willing to entertain the idea that adjunction can be to the right for one kind of head and to the left for another, but this goes against assumptions about incorporation in Baker (ÕÉ, ÕÉÉ), namely that adjunction is always to the left of the targeted head (Baker ÕÉÉ: óÉ). Another major drawback of their approach is that the argument status of the passive morpheme means that there is no external argument syntactically present. This is problematic in the light of evidence suggesting that there is a syntactically present external argument. For example, it is possible for this phonologically absent subject to control a PRO in a lower clause (Manzini ÕÉì, Sternefeld ÕÉÉ¢):
(Õß) a. Theyi decreased the price [PROi to the help poor]. b. The price was decreased [PROi to the help poor].
Furthermore, it possible for so-called ‘subject-oriented adverbs’ to occur in passives. In (Õb), the passivized variant of (Õa) , it is still possible for the subject-oriented adverb to occur. This suggests that there is in fact a syntactically/semantically present external argument at some point of the derivation in order to establish control and adverbial modification of the subject.
(Õ) a. Die the
Mädchen girls
haben have
die the
Cocktails cocktails
nackt naked
serviert. served ‘The girls served the cocktails (while) naked.’ b. Die the
Cocktails cocktails
sind have
nackt naked
serviert served
worden. been ‘The cocktails were served naked.’ (Sternefeld ÕÉÉ¢)
Baker, Johnson and Roberts (ÕÉÉ) do in fact discuss some of these data and want to claim that the passive suffix can fulfil all the functions of a genuine referential DP, but as far as I can see, these assumptions are implausible from a semantic point of view. Another issue is that their analysis employs a number of non-standard operations not only resulting from the out-datedness of the analysis. Even if we restate the analysis in modern terms, the situation does not improve. We could instead, assume the following clause structure:
(ÕÉ) [TP [vP - £Z«« [v′^ v [VP V DPint ]]]]
Stacked passives in Turkish óßì
Here, the passive morpheme is base-generated in Spec-vP. Crucially, the downgrading operation (lowering to V) has to take place in syntax proper. If we were to conceive of it as a postsyntactic ‘lowering’ operation in a Distributed Morphology approach, then it would still be present in Spec-vP at syntax and thus act as intervener for Relativized Minimality (cf. Section ì.ó). The way out of this quandary is to assume ’downward’ movement in Narrow Syntax, which would be highly problematic under a Minimalist approach as it is (i) unmotivated in other areas of grammar, (ii) not clearly feature-driven and (iii) violates the Extension Condition (Chomsky ÕÉÉ¢). Questions also arises with regard to the syntactic status of the suffix. Why is it only passive morphology that has this privileged argument status? Why does tense morphology (e.g. -ed), also assumed to be in I not also fulfil an argument function? Lastly, it is unclear what the semantics of these quasi-arguments is. We have seen that there is an ‘implied’ characteristic of suppressed arguments under passivization, this would require that the suffixes actually be -type arguments semantically and crucially, only passive morphemes would have this status.
ì.¦. Argument reduction by existential closure (Bruening óþÕì)
A different approach is proposed by Bruening (óþÕì), who claims that argument reduction is carried out by existential binding of the external argument slot. His assumption is that the Voice head (=vP) introduces the external argument. He assumes a semantics for Voice (v) in active clauses that is very similar to that of Kratzer (ÕÉÉä) as in (óþ):
(óþ) The lobbyist bribed the senator. VoiceP λe.bribing(e,the senator)^ ∧^ Initiator(e,lobbyist)
Voice′^ λe.λx.bribing(e,the senator)^ ∧^ Initiator(e,x)
VP λe.bribing(e,the senator)
the senator
V DP
bribe
Voice λf⟨s,t⟩.λe.λx.f(e) ∧ Initiator(e,x)
DP
The lobbyist
Stacked passives in Turkish óߢ
TP
T
∃x.∃y.beat(x,y) PassPò
Passò
-(I)n λf.∃x.f(x)
λx.∃y.beat(x,y) PassPÔ
PassÔ
-(I)l λf.∃x.f(x)
λy.λx.beat(x,y) VoiceP
Voice
...
VP
The problem with this approach is that the lower Pass head (PassÔ) first re- duces the internal argument (by closing the y slot) and then the higher head existentially binds the external argument slot. Thus, the derivation of the passive in Turkish under this approach would be transitive → antipassive → passive. If this analysis were correct, a single passivization operation in Turkish should result in an antipassive and then stacked passives would be derived by a second passivization operation. The antipassive is characterized by reduction or demotion of the internal argument rather than the external argument (see Silverstein ÕÉßó and Section ¢.ó). The fact that the antipassive construction is impossible in Turkish yields this analysis untenable.
ì.¢. The passive in Relational Grammar In Relational Grammar (Perlmutter ÕÉþ), grammatical relations are represented by arcs annotated with relations such as P (predicate), Õ (subject), ó (direct object) and ì (indirect object). Unlike in other theories of grammar (e.g. GB, Minimalism, HPSG), these relations constitute primitives of the theory. A simple example of the analysis of an active clause in Relational Grammar is given below:
óßä Andrew Murphy
(óì) Mary cooked the steak.
cooked steak Mary
P ó^ Õ
Here, the subject Mary bears the relation Õ corresponding to the subject and the object steak bears the relation ó. Grammatical function changing operations such as passivization are captured by assuming that these relations are not fixed, but rather fluid, i.e. they can change (Perlmutter and Postal ÕÉìa, Õɦa,b). Relational Grammar makes use of different levels of representation, or strata, to capture this. It is possible for a given lexical item to bear a relation (e.g. Õ for subject) in one stratum and a different relation (e.g. ó for object) in an another. For example, in the RG analysis of the passive (Perlmutter and Postal ÕÉìa), the thematic object can bear the object relation (ó) in the first stratum (cÕ) and the relation Õ in the second stratum:
(ó¦) The steak was cooked (by Mary).
cooked steak Mary
P ó Õ cÕ P Õ Cho
Much in the same way as transformational approaches, we have promotion of the internal argument to the highest grammatical role (subject). In the same way, the demotion/suppression of the external argument in passives is captured by demotion of the external argument from Õ to Cho, which stands for chômeur (the process is referred to as chômage). I will not go into the nature of this relation here (see Perlmutter and Postal ÕÉìa, Õɦa for discussion), it suffices to view this as an argument, which has undergone some kind of demotion/reduction. Now, let us consider how double passives can be analyzed in Relational Grammar. Özkaragöz (ÕÉä) discusses stacked passives in Turkish in this framework and claims the phenomenon constitutes passivization of a ‘personal passive’. Essentially, this results in two instances of passivization or multiple
óß Andrew Murphy
second passive suffix would then advance to Õ. This would capture double passivization but remains still problematic under the assumption of the Õ Advancement Exclusiveness Law (Perlmutter and Postal ÕÉìa, Õɦa) banning multiple advancement to Õ. One of the main questions here, and a problem that arose with Baker, Johnson & Roberts’ (ÕÉÉ) approach, is that we are giving argument status to passive morphology. In a framework such as RG where grammatical roles are primitives of the theory, it is not clear what it would mean to say that a dummy element or passive morpheme is the subject of a clause. Many analyses in this framework (e.g. Özkaragöz ÕÉä) are not too explicit about the exact semantics of their proposals but this a more general problem with analyses that account for argument reduction by assigning argument status to non-canonical arguments such as suffixes. One redeeming quality of the Relational Grammar approach also shared by Baker, Johnson and Roberts (ÕÉÉ) is that we capture the intuition that double passives are simply two instances of a standard passivization operation applied successively (i.e. transitive → passive → impersonal passive). In the RG approach, this is two instances of advancement to Õ and chômage and in BJR’s approach it is two instances of lowering to the INFL from Spec-IP.
ì.ä. Summary
In this section, we encountered various approaches to argument reduction and how they could be applied to instances of double passivization. Recall the problematic situation that we have evidence both for and against the existence of an external argument in passive constructions. Since lexical approaches simply deny the existence of an external (since its introduction is bled in the lexicon), they are not compatible with evidence for some syntactically-present external argument (e.g. control, subject-oriented adverbs). Furthermore, it is not made explicit how the link between argument reduction and passive morphology is captured, meaning that the link between two instances of passive morphology and dual argument reduction is not accounted for. Approaches that assume a syntactically-present, silent external argument do not suffer from the same problems but instead incur into Minimality violations as it is not clear how the object DP can raise over the higher subject DP. Furthermore, the assumption of a pro element in passives seems independent of the availability of pro-drop in the language in question and thus has a decidedly ad hoc flavour to it.
Stacked passives in Turkish óßÉ
The approaches by Baker, Johnson and Roberts (ÕÉÉ) and Özkaragöz (ÕÉä) share many similarities despite being couched in different frameworks. The BJR approach can circumvent many of the problems with regard to the external argument if one adopts the (controversial) assumption that lowering can take place in syntax proper. Nevertheless, both analyses inevitably suffer from the same problem of the argument status assigned to passive morphology, whose implications for both syntax (e.g. selection) and semantics are far from trivial. Thus, it seems that whether one assumes that there is an external argument syntactically present or not, different problems arise in each case. The question at this juncture is whether there is a third possibility that avoids all these problems. An alternative recently explored by Müller (óþÕ¦) assumes that the external argument is present for part of the derivation and is then later removed. This is the approach to argument reduction that I will adopt in the analysis to follow.
¦. The analysis
The problem we are facing with regard to the external argument in passives is that there seem to be arguments both for and against its syntactic presence. Thus, a completely satisfactory analysis would need to ‘have its cake and eat it’ by assuming that external argument is both syntactically present and absent. Rather pursue deep metaphysical questions of how a syntactic object can be both present and absent at the same time, I will follow Müller (óþÕ¦) in assuming that the external argument is present for only part of the derivation and is then later removed. This is what he calls the ‘short life-cycle of external arguments’. This will allow an external argument to present in the structure long enough to established downward relations such as binding/control as well as absorbing accusative caseÉ^ (óß), but be removed from the structure at later stage of the derivation early enough for it not to act as an intervener for movement of the subject (ó):
É (^) This assumption is not entirely unproblematic, however. See Section ¢.ì for discussion
Stacked passives in Turkish óÕ
The question at this point is what kind of operation can be proposed in a Minimalist framework to achieve this result. Such an operation will be presented and discussed in the following section.
¦.Õ. Reversing Merge: Slice In this section, I will present a syntactic operation recently discussed in Müller’s (óþÕ¦) analysis of the passive that will allow us implement the analysis sketched above. Whereas External Merge takes elements from the workspace/numeration and adds them to the existing structure, the operation I call Slice removes elements from the tree.Õþ^ Note that this is very similar to Sideward Movement (Nuñes óþþ¦), which is an operation that moves elements between workspaces. An important difference between Slice and Sideward Movement is that Slice occurs in a very strict structural configuration; in a Spec-Head configuration with a head bearing a [–D–] feature. As such it is very much the reverse operation of Merge (Merge = workspace → tree, Slice = tree → workspace).ÕÕ^ We can view this structure removal therefore as movement back into the workspace. Müller (óþÕ¦) assumes that, like Merge, this operation only applies at the root node and therefore does not violate the Extension Condition (Chomsky ÕÉÉ¢). Following the notational conventions in Heck and Müller (óþþß), Müller (óþÕÕ), there are two types of structure building features: ‘bullet’ features triggering (External or Internal) Merge [• D • ], and ‘star’ features triggering Agree operations [F]. Assuming that Slice is also feature-driven, we can then add a corresponding Slice feature to our list of structure building features:
(óÉ) Structure-building features a. Merge features: [•D•] b. Probe features: [F] c. Slice features: [–D–]
In the following section, we will see how these features can be combined to successfully derive both passives and stacked passives in Turkish. Õþ (^) Müller (óþÕ¦) calls this operation Cut. I feel Slice is more appropriate as it implies that only a thin layer is removed. As will be discussed below, it is assumed that this operation only applies at the root node and therefore it is only ever the top layer of a given structure that can be removed. ÕÕ (^) The triggers for Sideward Movement are somewhat unclear and thus it remains less-restricted than Slice, although see Nuñes (óþÕó) for recent discussion of this point.
óó Andrew Murphy
¦.ó. A Slice-approach to standard passives
Now, let us see how Slice can be applied to standard passives. I follow Merchant (óþÕì), Harley (óþÕì) and the growing body of literature, suggesting that Voice and v constitute distinct heads (contra Kratzer ÕÉÉä) in the clausal spine. The morphological reflex of passivization is captured by assuming the passive suffix is the head of VoiceP above vP and that it bears a Merge-triggering feature [•D • ] and a Slice-feature [–D–]. These are featured are ordered with regard to one another so that [•D • ] precedes [–D–]. The derivation will precede as follows: The [•D • ] feature first triggers internal merge of the closest DP (since the numeration is empty); in (ìþ), the external argument. This DP is then moved back into the workspace in order to check the Slice feature [–D–] and is therefore no longer present at the point of the derivation where T probes for a goal for movement to Spec-TP:
(ìþ) The steak was cooked. TP
T′
VoicePÔ
Voice′
vP
v′
VP
V tthe steak cook
v
tDPext
Voice -ed [•D•]>[–D–]
T
[•D•]
DP
the steak
DPEXT
As a result, it is the internal argument that is moved to Spec-TP, thus deriving a standard passive construction. Now that the analysis of a personal passive, the
