CC BY
20
Philip Shushurin
New York University, New York — Institute of Linguistics, RAS,
Moscow
THE OBLIQUE CAUSER CONSTRUCTION IN LEZGIAN1 1. Introduction
Recent studies in argument structure have paid increasing attention to constructions which contain arguments which are not directly selected by the main predicate [Bosse et al. 2012; Kim 2011 etc.]. One such construction is the so-called Oblique Causer Construction, which is exemplified below by sentences from German (1) and Icelandic (2):
(1) Dem Mädchen zerbrach die Vase
the.dat girl broke the.nom vase
'The girl unintentionally broke the vase.' [Schäfer 2008]
(2) Blom fölna alltaf hja mer.
flower.nom wilt always by me. dat
'I'm always causing flowers to wilt.' [Wood 2013]
This construction has been widely discussed in recent literature. Thus, Kittilä [2005] provides a typological overview of this construction, which he calls Involuntary Agent Construction, Ganenkov et al. [2008] discuss this construction in Agul, a Dagestanian language closely related to Lezgian. Schäfer [2008] provides a survey of several languages which possess this construction and identifies several properties which characterize the Oblique Causer Construction (OblCC) in these languages. Thus, according to him, oblique causers can only be human and are only compatible with change-of-state verbs.
Schäfer proposes that oblique causers are arguments introduced by an applicative head and assigned an inherent case in SpecApplP.
In this paper, I present data from Lezgian (Lezgic, Nakh-Dagestanian), which, as I will show, also possesses the Oblique Causer Construction; see an example of such construction in (3):
1 This research was supported by the Russian Science Foundation grant No.14-18-02429.
(3) Nariman-awaj rak agal-na N.-adel door. abs close-aor 'Nariman unintentionally closed the door.'
This construction has been mentioned in Haspelmath's [1993], although not discussed at length.
Lezgian OblCCs, while sharing some important similarities with the corresponding constructions in Schäfer's sample, exhibit several unique characteristics.
The crucial difference between Lezgian and German, on which I will largely concentrate in this paper, is the compatibility of Lezgian OblCCs with transitive constructions, see below:
(4) Kerim-awaj pul gu-n xa-na K.-adel money bring-msd become-aor 'Kerim unintentionally brought the money.'
In contrast, transitive predicates are prohibited in German:
(5) *Dem Mädchen hat den Becher def.dat girl has def.acc glass gebrochen
broken
int. 'The girl unintentionally broke the glass.'
While the sentence in (5) contains a transitive construction (the object gets accusative case), the sentence in (1) does not: the internal argument has nominative marking and possesses subject properties.
In this paper, I investigate properties of OblCCs in Lezgian and propose that OblCCs in Lezgian are instances of light verb constructions — combinations of a semantically vacuous verb and a preverb, which, together, form a complex predicate. I suggest that oblique causers are introduced by a Voiceexpl — a projection which doesn't introduce a semantic role by its own but rather transmits it from an assigner of a semantic role contained lower in the structure. In this respect, OblCCs can be compared to such phenomena as possessor raising, where the possessor is syntactically realized outside of the DP which assings the possessor semantic role. Voiceexpl can only be merged with an expletive v, which is in most cases realized as the light verb xun. The expletive v, in its turn, merges with structures
which have not realized their external argument: with inchoative structures or subjectless non-finite clauses.
The rest of the paper is organized as follows. In section 2, I provide data on the morphosyntax of the Oblique Causer Construction. In section 3, I discuss syntactic properties of the auxiliary verb xun and propose that it should be analyzed as a light verb. In section 4, I outline the proposal and suggest that adelative phrases are introduced by an expletive voice projection. In section 5, I discuss syntactic properties of masdar OblCCs and propose that they contain subjectless voice phrases. In section 5, I propose that the expletive voice and the ordinary voice can be analyzed as two instances of the same projection. Section 6 concludes.
2. Morphosyntax of Lezgian OblCC
Lezgian is an ergative language, which, normally, marks subjects of transitive constructions with ergative case. The main predicate usually bears inflection. I will call constructions like (6) direct constructions and the external argument in such constructions canonical agent:
(6) Nariman-a rak agal-zawa N.-erg door.abs close-ipfv 'Nariman is closing the door.'
OblCCs (like (7)) are different from direct constructions in several respects. Firstly, they always possess an argument which is marked with adelative case:
(7) Kerim-awaj pul gu-n xa-na K.-adel money bring-msd become-aor 'Kerim unintentionally brought the money.'
Outside of OblCC, adelative is mostly used in spatial expressions, roughly corresponding to English from-PPs:
(8) Cun ad-awaj q:aq:at-na. we.abs he-adel go.away-aor 'We went away from him.'
I will follow Svenonius [2012] in assuming that case markers (at least of non-structural cases) in Lezgian are adpositions and the
whole words are postpositional phrases. I propose the following structure for adelative phrases:
(9) Structure of adelative phrases:
AdP
DP Ad'
Kerim -awaj
If an adelative argument is not superficially present in a OblCC, the sentence can be interpreted as containing an adelative pro, once there is a salient human referent in the discourse.
(10) pro Ali k'wal-iz xutax-un xa-na pro.adel A.abs home-dat bring-msd become-aor '(S/he) uninentionally brought Ali home.'
The presence of such an argument can be diagnosed in the following way. In Lezgian, emphatic reflexives must agree in case with their antecedents.
(11) Nariman-a wic-i/*wic rak agal-zawa N.-erg self-erg/self.abs door.abs close-ipfv 'Nariman is closing the door himself.'
The following example shows that sentences like (12) allow for emphatic reflexives marked adelative, even if the adelative argument is not overtly realized:
(12) (Kerim-awaj) wic-iwaj Ali k'wal-iz K.-adel self-adel A.abs home-dat xutax-un xa-na
bring-msd become-aor
'Kerim himself unintentionally brought Ali home. '
The other important distinction is the morphological form of the verb. In direct constructions, the predicate appears in a synthetic form (13), or, depending on tense/aspect, forms a periphrastic construction, where the main predicate is in the root form, and the inflection is borne on the auxiliary (ex. (14)-(15)).
(13) Ali-di car kxe-na
A.-erg letter.abs write-aor 'Ali wrote a letter.'
(14) Ali-di car kxin iji-zwa A.-erg letter. abs write.rf do-ipfv 'Ali is writing a letter.'
(15) Ali-di car kxin iji-da A.-erg letter. abs write.rf do-fut 'Ali will write a letter.'
In OblCCs, the morphological marking on the verb is not uniform, and three types of construction can be distinguished. In the first type, which I call masdar OblCCs, the verb appears in the so called masdar form; see the following example:
(16) Kerim-awaj Ali k'wal-iz xutax-un xa-na K.-adel A. abs home-dat bring-msd become-aor 'Kerim uninentionally brought Ali home.'
Masdar is a term used in descriptive grammars ([Haspelmath 1993] and others) to denote historically nominalized forms. It is not obvious whether synchronically, masdar clauses should be analyzed as non-nominalized structures or nominalized structures marked absolutive (in Lezgian, absolutive marking is always null). I will assume that masdar clauses are not nominals although this question may require some additional investigation. (Nothing in my analysis crucially depends on this choice).
Importantly, masdar clauses in OblCCs can contain both intransitive (17) and transitive (18) verbs:
(17) Kerim-awaj xkat-un xa-na K.-adel leave-msd become-aor 'Kerim left unintentionally.'
(18) Kerim-awaj ic basmis-un xa-na K.-adel apple.abs crush-msd become-aor 'Kerim unintentionally crushed the apple.'
Tense in masdar clauses is always anaphoric on the matrix clause. Thus, if the matrix verb bears present or future tense, the verb
in the masdar clause will denote a present respectively.
(19) Kerim-awaj fad-fad tup wehi-n K.-adel often ball.abs drop-msd 'Kerim often accidentally drops the ball.'
(20) Kerim-awaj tup wehi-n ze-da K.-adel ball.abs drop-msd become-fut 'Kerim will accidentally drop the ball.'
If the matrix clause and the embedded clause contain adverbs of different tense, the sentence becomes ungrammatical:
(21) *Naq' Kerim-awaj paka pul Yesterday K.-adel tomorrow money. abs gu-n xa-na
bring-msd become-aor
int. (roughly) 'It so happened yesterday that Kerim will unintentionally bring the money tomorrow.'
In the subsequent subsections I will propose that masdar clauses in masdar OblCCs are Voice Phrases which which do not project an external argument.
The second case concerns the so called labile verbs. The set of verbs identified as labile consists of predicates which participate in the inchoative-causative alternation. Thus, the root agal 'close' can be used both in inchoative and causative sentences:
(22) Dak 'ar agal-na Window close-aor 'The window closed.'
(23) Kerim-a dak'ar agal-na K.-erg window. abs close-aor 'Kerim closed the window.'
Importantly, in such cases, a periphrastic construction can also be used. In inchoative sentences, the auxiliary is xun2 while in
2 The intransitive auxiliary xun is subject to suppletive allomorphy, with the main allomorphs being xu-, xa- and ze-. I gloss this morpheme consistently as 'become'.
or a future action,
ze-zwa become-lpfv
causative sentences another auxiliary must be used — the transitive auxiliary awun.
(24) Dak'ar agal xa-na Window close become-aor 'The window opened.'
(25) Kerim-a dak'ar agal awu-na K.-erg window. abs close do-aor 'Kerim closed the window.'
Labile verbs in OblCCs can also be used both periphrasitically and non-periphrastically, as the following examples show. However, in periphrastic form, only the intransitive auxiliary can be used:
(26) Kerim-awaj dak'ar agal-na K.-adel window. abs close-aor 'Kerim unintentionally closed the window. '
(27) Kerim-awaj dak'ar agal xa-na K.-adel window. abs close become-aor 'Kerim unintentionally closed the window. '
(28) *Kerim-awaj dak'ar agal awu-na K.-adel window. abs open do-aor int. 'Kerim unintentionally opened the window.'
Labile verbs can also form masdar clause; in those cases a periphrastic construction with the verb awun is always possible (xun is disallowed (30)):
(29) Kerim-awaj dak'ar agal-un/ agal K.-adel window close-msd/ close awu-n xa-na
do-msd become-aor
int. 'Kerim unintentionally closed the window.'
(30) *Kerim-awaj dak'ar agal K.-adel window. abs close x-un xa-na become-msd become-aor
int. 'Kerim unintentionally closed the window.'
In addition to those two constructions, a third type of construction can be identified3. This type of construction consists of a predicative clause and an adelative argument:
(31) Kerim-awaj cal jaru xa-na K.-adel wall. abs red become-aor 'Kerim unintentionally painted the wall red. '
In contrast to the previous case, no other element in the clause, apart form the auxiliary, can bear inflection:
(32) *Kerim-awaj cal jaru-na K.-adel wall. abs red-aor
int. 'Kerim unintentionally painted the wall red'
I will call such construction predicative OblCC. Despite the differences, all three types of construction share some similarities: there are several other restrictions associated with all three types of OblCC. Thus, OblCC only occur with telic predicates (the telicity requirement), as the following example shows, which contains a non-telic predicate and is ungrammatical:
(33) *Kerim-awaj sa sajat kat-un xa-na K.-adel one hour. abs run-msd become-aor int. 'Kerim was running for one hour although he didn't intend it. '
This sentence can be contrasted with the following grammatical sentence where the same verb katun 'run' is telic:
(34) Kerim-awaj k'wal-iz kat-un xa-na K.-adel home-dat run-msd become-aor 'Kerim unintentionally ran home. '
The same requirement holds for two other constructions as well. Another requirement, which also holds for all three types of OblCCs, is that the oblique causer must be human, (the humanness requirement) as the following example shows:
3 One difference of this type from the previous one is that non-periphrastic constructions (58) are not allowed here.
(35) *Gar-awaj rak agal-un xa-na wind-adel door.abs close-msd become-aor int. 'The wind closed the door.'
Finally, oblique causers cannot be merged with agentive structures (transitive or unergative clauses) directly, in the absence of the auxiliary xun (the non-agentive requirement):
(36) *Kerim-awaj pul ga-na K.-adel money.abs bring-aor int. 'Kerim unintentionally brought money.'
(37) * Kerim-awaj k'wal-iz kat-na K.-adel home-dat run-aor int. 'Kerim unintentionally ran home'
These ungrammatical examples can be contrasted with minimally different (7) and (34) respectively. I will propose that the non-agentive requirement is a structural constraint on the proj ection which introduces oblique causers, while the humanness requirement and the telicity requirement are semantic constraints associated with the Initiator semantic role which is assigned to oblique causers. Before proceeding to the proposal, I will discuss the syntactic properties of the verb xun, and its transitive counterpart awun, which, as I will argue, are light verbs and are merged in v, below the projection which introduces the oblique causer.
3. Syntax of the auxiliary xun
In this subsection, I intend to show that the auxiliary verb xun, which OblCCs almost always contain, should be analyzed as a light verb which occupies the head of little v.
I begin with listing the syntactic positions in which xun can occur, apart from OblCCs. Firstly, xun can serve as a light verb and form a light verb construction (LVCs). Thus, xun can form intransitive LVCs. In the next example, the auxiliaries combine with the root ujax 'awake' and form the verb meaning 'to wake up'.
(38) Ali ujax xa-na Ali.abs awake become-aor 'Ali woke up.'
The auxiliary has a transitive counterpart awun, which also can form light verb constructions. Thus, combining with the same morpheme, ujax, it forms the predicate 'to wake (smb.) up':
(39) Ali-di Kerim ujax awu-na Ali-erg Kerim.abs awake do-aor 'Ali woke Kerim up.'
An important subclass of such constructions are certain complex psych predicates which denote change of state, such as kic'e xun 'get frightened', bizar xun 'get fed up', and others (experiencers of such verbs are marked dative):
(40) Ali-diz sew-reqhaj kic 'e xa-na Ali-dat bear-poel afraid become-aor 'Ali got frightened of the bear.' etc.
These predicates cannot form OblCCs.4
(41) *Ali-diwaj sew-reqhaj kic'e xa-na Ali-adel bear-poel afraid become-aor
int. (roughly) 'Ali got frightened of the bear by accident.'
Finally, in addition to the constructions discussed above, the verb xun, as well as its transitive counterpart awun, can be used as main predicates in the clause (or, to use a different terminology, are used as heavy verbs), see examples below:
(42) Jug nani ze-zwa day.abs evening.abs become-ipfv
'It is becoming evening.' [Haspelmath 1993]
(43) Ada dide-di laha-j-wal awu-na. he. erg [mother. erg say-aop-man] do-aor
'He did as his mother had said.' [Haspelmath 1993]
The these types of environment share one thing in common. The complement of the verb xun denotes a result state. The presence of
4 The ungrammatically of (41) is due to the fact that OblCCs cannot be formed unless there is an element assigning Initiator semantic role, this idea will be elaborated in subsequent subsections.
this state can be diagnosed in the following way. Adding the refactive prefix qh- 'again' to the verb yields the restituitive meaning:
• Predicative:
(44) Cal jaru xa-na wall.abs red become-aor 'The wall became red.'
(45) Cal jaru qh-xa-na wall.abs red again-become-aor
'The wall became red again.' (Presupposition: 'The wall was red before.')
• Eventive psych verbs:
(46) Ali-diz sew-reqhaj kic 'e xa-na Ali-dat bear-poel afraid become-aor 'Ali got frightened of the bear.'
(47) Ali-diz sew-reqhaj kic 'e qh-xa-na Ali-dat bear-poel afraid again-become-aor
'Ali got frightened of the bear again.' (Presupposition: 'Ali was afraid of the bear before.')
• Inchoative:
(48) Dak'ar aqha xa-na window.abs open become-aor 'The window opened.'
(49) Dak'ar aqha qh-xa-na window.abs open again-become-aor
'The window opened again.' (Presupposition: 'The window was open before.')
To sum up, we have seen that the verb xun is associated with the following four environments:
1. inchoative (change-of-state);
2. predicative clauses (change-of-state);
3. psych verbs (change-of-state);
4. masdar OblCCs (stative).
Constructions like ones described above, such as (38), have recently received much attention in literature. Thus, Folli et al. [2005] discuss similar data in Modern Persian. In Persian, complex predicates are formed by merging a semantically deficient verb (light verb, LV) with a preverbal element, or preverb (PV), which can have diverse syntactic characteristics.
Consider the following example with the light verb sodan 'become', which is an equivalent of the Lezgian sentence (38):
(50) Kimea bidar sod Kimea awake became
'Kimea became awake.', 'Kimea woke up.' [Folli et al. 2005] Folli et al. propose the following structure for (50):
(51) Structure of Persian complex predicates:
vP
DP A sod
Kimea bidâx
Sugimura [2012] discusses similar data in Japanese, which possesses the light verb su- 'do'. In Japanese, this verb can merge with PVs which participate in inchoative/causative alternation. She suggests that the preverbs are uncategorized roots and proposes the following structure for the complex predicate in (53), which accounts, among other things, for the lack of case-marking on the preverbal element:
(52) Matto-ga syokuji-sita Matt-nom dining-did
'Matt had dinner.' [Sugimura 2012]
(53) Structure of Japanese light verb constructions:
vP
^syokuji v su
The data that I presented above shows clearly that the Lezgian auxiliaries xun and awun are structurally equivalent to both Japanese and Persian light verbs5. Thus, Lezgian auxiliaries are similar to their Japanese and Persian analogues in their ability to verbalize preverbs, to merge with predicative clauses (for intransitive light verbs) and to participate in inchoative-causative alternation.
It should also be noted that, in contrast to English auxiliaries, such as have or be, xun and awun are not associated with any tense/ aspect form.
I propose that, firstly, the Lezgian verb xun in all four cases mentioned above, including the OblCCs, occupies the same structural position, and, secondly, that this position is the head of expletive little v.
The term expletive is intended to capture the following properties of this kind of little v. First of all, it doesn't assign a semantic role to any argument. Thus, in predicative construction, semantic roles to both arguments are assigned inside the PredP. In inchoative constructions, the semantic role to the internal argument is assigned by the root. In masdar OblCCs, all semantic roles are assigned within the masdar clause.
Secondly, the expletive v doesn't contribute to the semantic interpretation of the sentence.
Let's now consider each individual construction. In predicative sentences, xun is contained in the same v projection which takes as a complement a predicative clause. Thus, I propose the following structure for (54)6.
5 See, however, Ilkhanipur and Sugawara [2015] who propose that at least in some cases the Persian verb sod must be analyzed as a voice marker rather than a light verb.
6 Here and henceforth I don't indicate the structural position where the inflectional morphology is generated, showing it at places where it is superficially
(54) Cal jaru xa-na
wall.abs red become-aor 'The wall became red'
(55) Lezgian predicative construction:
In inchoative constructions, I propose, the expletive v takes a root phrase (see e. g. [Sœbo 2001] for a similar proposal). Thus, I propose the following structure of the predicate in (57):
(56) Dak'ar agal
window.abs close 'The window closed.'
xa-na
become-aor
(57) Inchoative structure (periphrastic):
As for constructions which allow non-periphrastic constructions, I propose that the position of the head of the expletive v in such cases
realized. Also, I don't discuss here whether subjects in this sentence, as well as in similar cases, remains in its thematic position or moves to SpecTP.
is occupied by a null allomorph of the auxiliary xun1. I propose the structure for (58) in (60).
(58) Dak'ar agal-na window.abs close-aor 'The window closed'.
(59) Dak'ar agal xa-na window.abs close become-aor 'The window closed.'
(60) Inchoative structure (non-periphrastic):
VexplP
vp 0
DP ^/aqal-im
A
dak'ar
I propose that in both cases the little v is merged with an uncategorized root, the head of which may or may not be realized overtly. I furthermore propose that the expletive v can only have a null realization when the inflection can be successfully realized, which is the case in (58) (inflection can attach to roots) and not in (61) (inflection cannot attach to masdars).
(61) *Kerim-awaj fad-fad tup wehi-n-zwa
K.-adel often ball.abs drop-msd-ipfv
int. 'Kerim often accidentally drops the ball.'
An important issue that arises at this point is that it seems that the semantic contribution of the verb xun in predicative cases, eventive psych verbs and inchoative constructions is different from that in masdar OblCCs. I follow [Wood, Marantz 2017] in assuming that the interpretation of the predicative clause is sensitive to the properties of
7 Speakers report that (58) and (59) only differ in register, from which
I conclude that there is the choice of the allomorph is not structurally conditioned.
the syntactic environment where it appears. In our case, the predicative clause is interpreted as change-of-state when it is a complement of little v.
Although I don't develop a semantic theory here, I propose here that the semantic contribution of the light verb in predicative clauses is null and thus is identical to that in masdar OblCCs.
4. The main claim
Schäfer [2012] discusses OblCCs in German and proposes that oblique causers are only allowed in the context of the resultative structure. More specifically, he proposes that oblique causers are introduced by an applicative projection which is merged with a constituent denoting change of state:
(62) Structure of oblique causers [Schäfer 2009]:
ApplP
DP Appl'
Appl change of state
In this subsection I will show that Schafer's generalization that oblique causers are only compatible with resultative structures doesn't hold for Lezgian.
My claim is that oblique causers are introduced by a special kind of Voice phrase (I call it Voiceexpl). Furthermore, I suggest that OblCCs are subject to several constraints. First, they are only compatible with telic predicates and are only compatible with human causers. Second, Voiceexpl is subject to the following structural constraints: I suggest that it cannot merge with agentive structures directly and can only be merged with structures which assign, but fail to realize, the Initiator semantic role. The latter two constraints account for the fact that languages only allow for OblCCs with inchoative structures (such structures are non-agentive and, by assumption, contain
an unrealized Initiator8 semantic role), unless the language in question has a strategy to embed transitive structures under intransitive light verbs. For the languages which do have this strategy, Schäfer's generalization doesn't hold.
I start with observing that masdar OblCCs show different constraints on predicates if compared to inchoative OblCCs. All inchoative OblCCs involve change-of-state predicates. Clearly, the restrictions on Masdar OblCCs are different. Thus, the following example shows that masdar OblCCs can involve predicates which do not denote a change of state: unlike previous cases, adding the prefix qh- to the predicate yields only repetitive readings, with restituitive readings unavailable.
(63) Kerim-awaj ic basmis-un xa-na K.-adel apple.abs crush-msd become-aor 'Kerim unintentionally crushed the apple.'
(64) Kerim-awaj ic basmis-un qh-xa-na K.-adel apple.abs crush-msd again-become-aor 'Kerim again unintentionally crushed an apple' (Presupposition: Kerim crushed an apple before)
However, all predicates in OblCCs must be telic. Consider the minimal pair below. These sentences involve the verb katun 'run', which can have a telic as well as non-telic use.
(65) Kerim sa sajat kat-na K.abs one hour.abs run-aor 'Kerim was running for one hour'
(66) Kerim k'wel-iz kat-na K.abs home-dat run-aor 'Kerim ran home (in an hour).'
In its telic use it can form an OblCC while in its non-telic use it cannot:
(67) *Kerim-awaj sa sajat kat-un xa-na K.-adel one hour.abs run-msd become-aor int. 'Kerim ran for one hour although he didn't intend it.'
8 I borrow Initiator as a cover term for the semantic roles agent, causer etc. from Bruening [2013]. By assumption, Experiencer semantic role is not included.
(68) Kerim-awaj k'wel-iz kat-un xa-na K.-adel home-DAT run-MSD become-AOR 'Kerim ran home unintentionally.'
Despite the difference between the two cases, they have one thing in common: the structures in both cases contain an expletive v. I propose that the Voiceexpl projection, which introduces an oblique causer, can only merge with an expletive v. I propose the following general structure of OblCC (69).
(69) General schema of OblCCs:
VoiceexpiP
oblique causer vP^pi voice
voiceP/rootP
I call the projection where the adelative argument is merged Voiceexpl although the exact label is not crucial: in the next subsection I will propose that this kind of Voice and the VoiceP which introduces canonical agents can be analyzed as two instances of the same projection. The head of this projection is null and the specifier of this projection contains an Adpositional Phrase (AdP), where the adelative argument is merged.
I want to propose that a different requirement is relevant in Lezgian, namely, the following:
(70) Voiceexpl can only merge with a non-agentive v.
By definition, an agentive v is a vP which contains an element capable of assigning an Initiator semantic role in its head. In Lezgian OblCCs, the vP with which the Voiceexpl merges, contains the semantically vacuous auxiliary xun which is unable to assign the Initiator semantic role (and any other semantic role, in fact).
As we have seen, apart from a small number of cases involving labile verbs, where, as I have proposed, the expletive v is nevertheless
present, all OblCCs involve an expletive v, which is, by definition, non-agentive. Formulating this requirement allows us to derive, in most cases, the correct distribution of OblCCs.
To sum up, I propose that in order for an OblCC to be merged and felicitously uttered, the following two requirements must be met:
(71) Structural requirement on Voiceexpl (to be revised)
Voiceexpl cannot be merged with an agentive v.
(72) Semantic requirement on oblique causers
Oblique causers, can only denote humans and are only compatible with telic predicates9.
Below I show that this system indeed derives the right distribution OblCC (apart from one class of exceptions which I will discuss below). Thus, the sentences which denote telic predicates but do not contain an expletive v are ruled out because the Structural requirement is not satisfied.
(73) *Kerim-awaj tup wehe-na K.-adel ball. abs drop-aor int. 'Kerim accidentally dropped the ball.'
Similarly, structures which contain an expletive v but which do not contain a telic predicate are disallowed because they violate the Semantic Requirement.
(74) *Kerim-awaj sa sajat kat-un xa-na K.-adel one hour.abs run-msd become-aor int. 'Kerim ran for one hour although he didn't intend it.'
Transitive structures cannot be directly merged with an adelative argument because any transitive verbal structure contains an agentive v. However, a transitive clause can be a complement of an expletive v. In this case, the expletive v does not make any semantic contribution to the sentence, but it makes possible for a Voiceexpl to merge.
However, these two requirements are not enough. The system fails to rule out unaccusative structures. For example, OblCCs cannot
9 I leave for future research whether or not these two requirements can be further generalized as a teleological capacity requirement on non-canonical agents (see [Folli and Harley 2008]).
be formed with predicates which are not agentive (i. e. do not assign the Initiator semantic role) but are not labile.
The following minimal pair shows that while inchoative structures (both periphrastic and non-periphrastic) can merge with an oblique causer, a minimally different unaccusative structures cannot:
(75) Kerim-awaj ttar jarx xa-na K.-adel tree.abs fall become-aor 'Kerim unintentionally felled the tree.'
(76) *Kerim-awaj ttar awat-na K.-adel tree.abs fall-aor
int. 'Kerim unintentionally felled the tree.'
While the verb in the former sentence is labile (it participates in inchoative-causative alternation: jarx xun 'fall (intr.)' — jarx awun 'fall' (tr.)), the verb in the latter one is not, although nothing in the system presented above should rule this sentence out.
In order to capture the right distribution, I propose the following. I assume that labile verbs are different from non-labile unaccusatives in that the former contain an element which is able to assign the Initiator semantic role. Thus, by assumption, the root terg is able to assign the Initiator semantic role, while the root kwax is not10. (The diagnostics for agentivity of roots is simple: only agentive verbal roots are compatible with canonical agents).
Having said that, we can now amend the structural condition in a way that would avoid overgeneration:
(77) Structural requirement on Voiceexpl (revised)
Voiceexpl can only be merged with structures which are non-agentive and which contain an assigned but non-projected Initiator semantic role.
Positing this requirement allows us now to cull out unaccusatives while leaving the rest of the distribution intact.
In the next subsection, I will show that this account can be extended to masdar OblCCs.
10 I take an agnostic position on whether the Initiator semantic role is always assigned by roots or it can be sometimes assigned by bigger structures.
5. Masdar OblCCs
Before I proceed to discussing how the outlined approach can be extended to masdar OblCCs I turn to discussing syntactic properties of such clauses. More specifically, I argue here that the masdar phrase in masdar OblCCs is a voiceP which doesn't contain the subject.
I begin with showing that adelative phrases are not arguments of the masdar clause. In other words, I will show that 1. is the right bracketing for (78) and 2. is not.
(78) Kerim-awaj tup wehi-n xa-na K-adel ball.abs drop-msd become-aor 'Kerim accidentally dropped the ball.'
1. Kerimawaj [[tup wehin] xa-na]
2. [Kerimawaj [tup wehin]] xa-na
There are two empirical arguments in favor of 1. Firstly, the emboldened structure in 1. can undergo movement, while the emboldened structure in 2. cannot.
In Lezgian, constituents can undergo rightward movement11 to the postverbal position. This movement is driven by information structure and can target only given constituents. The next sentence consists of a raising verb which takes a sentential complement, which, in the normal case, linearly precedes the matrix verb.
(79) [Kerim-a ktab-ar k'wel-iz] baslamis-na K.-erg book-pl read-inf begin-aor 'Kerim began to read books.'
The next examples show that the embedded clause in this case can move to the postverbal position.
(80) Baslamis-na [Kerim-a ktab-ar k'wel-iz] begin-aor K.-erg book-pl read-inf (roughly) 'As for reading books, Kerim began doing that.'
11 I don't commit to a particular analysis of this phenomenon in this case. The same argumentation can be applied if one assumes that such constituents are generated in the right periphery.
Turning to OblCCs, they also allow rightward movement of the bracketed structure in 1., while the analogous movement of the emboldened structure in 2. is dispreferred.
(81) Tup wehi-n xa-na Kerim-awaj ball drop-msd become-aor K.-adel 'Kerim accidentally dropped the ball. '
(82) ??Xa-na Kerim-awaj tup wehi-n become-aor K.-adel ball.abs drop-msd int. 'Kerim accidentally dropped the ball.'
In contrast, rightward movement of the phrase containing the verb and its internal argument is judged fully grammatical:
(83) Kerim-awaj xa-na [tup wehi-n] K.-adel become-aor ball.abs drop-msd 'Kerim accidentally dropped the ball. '
I take this as evidence that the adelative argument does not form a constituent with the bracketed structure in 1.
(I assume that the relative acceptability of (82) is due to the fact that this sentence has been generated by two consecutive rightward movements).
Furthermore, there is evidence that masdar clauses do not contain any adelative subjects. Let's begin with showing that masdar clauses do not contain any covert adelative subjects. The following sentence contains a masdar OblCC with an emphatic reflexive marked adelative:
(84) Kerim-awaj wic-iwaj tup wehi-n xa-na K.-adel self-adel ball.abs drop-msd become-aor 'Kerim himself accidentally dropped the ball'
One could analyze this construction in two ways, Firstly, one can suggest that the antecedent of the emphatic reflexive wiciwaj is the overt adelative subject of the matrix clause.
Secondly, one can assume that the antecedent of the emphatic reflexive is a covert adelative argument (presumably, a PRO) contained in the embedded masdar clause. The following sentence shows that the second option is incorrect.
(85) *Kerim-awaj xa-na [wic-iwaj tup K.-adel become-aor self-adel ball. abs wehi-n]
drop-msd
int. 'Kerim himself accidentally dropped the ball.'
When moved to the right periphery, the embedded clause cannot contain the adelative emphatic reflexive12. The grammatically of (86), where the emphatic reflexive has remained in situ, suggests that it is contained in the matrix clause and its immediate antecedent is the adelative argument in the matrix clause.
(86) Kerim-awaj wic-iwaj xa-na [tup K.-adel self-adel become-aor ball. abs wehi-n]
drop-msd
'Kerim himself accidentally dropped the ball.'
Similarly, using the same diagnostics, one can show that masdar clauses in OblCCs do not contain any covert subjects. Thus, the following examples show that masdar clauses do not license ergative or absolutive emphatic reflexives.
(87) *Kerim-awaj wic-i tup wehi-n xa-na K.-adel self-erg ball.abs drop-msd become-aor int. 'Kerim himself accidentally dropped the ball.'
(88) *Kerim-awaj wic xkat-un xa-na K.-adel self.abs leave-msd become-aor int. 'Kerim himself left unintentionally.'
The sentence (87) contains a transitive predicate, which, in direct construction, marks the subject ergative; however, the ergative marking on the emphatic reflexive is not possible. Similarly, the sentence (88), containing an intransitive predicate, disallows absolutive emphatic reflexives.
The minimally different sentences containing adelative emphatic reflexives are grammatical in both cases:
12 Additionally, one can show that, as an empirical rule, emphatic reflexives must immediately linearly follow its antecedent.
(89) Kerim-awaj wic-iwaj tup wehi-n xa-na K.-adel self-adel ball.abs drop-msd become-aor int. 'Kerim himself accidentally dropped the ball.'
(90) Kerim-awaj wic-iwaj xkat-un xa-na K.-adel self-adel leave-msd become-aor int. 'Kerim himself left unintentionally.'
Finally, the following sentences show that emphatic reflexives can have covert arguments as antecedents. Thus, they are licensed in control environments and agree with PRO in case, as the following examples show:
(91) Kerim-az [PRO wic xkat-un] k'an-zawa K.-dat PRO. abs self. abs leave-msd want-lpfv int. 'Kerim wants to leave by himself.'
(92) Kerim-az [PRO wic-i ic K.-dat PRO.erg self-erg apple. abs basmis-un] k 'an-zawa
crush-msd want-ipfv
int. 'Kerim wants to crush the apple himself.'
At the same time, I assume that the masdar clause in masdar OblCCs contains a voiceP13. One piece of evidence comes from the fact that the masdar clauses can license instrumental phrases:
(93) Kerim-awaj kutal-daldi cewg K.-adel hammer-srdir ant. abs q 'i-n xa-na
kill-msd become-aor
'Kerim unintentionally killed the ant with a hammer. '
The instrumental is contained in the masdar clause, since it can rightward-move with rest of the clause, as the following example shows:
(94) Kerim-awaj xa-na [kutal-daldi K.-adel become-aor hammer-srdir
13 I follow [Alexiadou et al. 2006; Harley 2009], among others, in assuming that vP and voice are distinct projections.
cewg q 'i-n] ant.abs kill-msd
'Kerim unintentionally killed the ant with a hammer.'
I follow Bruening [2013] who shows that instrumental phrases are only licensed by structures containing a Voice Phrase.
I propose that the masdar clause is a VoiceP which does not project an external argument. Thus, I propose the following structure for sentences like (96):
(95) Structure of masdar OblCCs:
(96) Kerim-awaj tup wehi-n xa-na
K.-adel ball.abs drop-msd become-aor
'Kerim accidentally dropped the ball.'
This structure is similar to the previously discussed inchoative construction in the following way: it has a light verb which embeds a structure which can (due to the presence of an embedded voice projection) but doesn't project an external argument.
This position is characteristic for complex predicates. Complex predicates can be defined as structures containing at least two predicates which require one argument which saturates them all. In our case, we are dealing with a subset of complex predicates: viz., light verb constructions (LVC), which consist of a light verb and a preverb.
In the simple case, the preverb of a LVC contains little structure (a nominal, a root phrase or a PredP). In the case of masdar clauses, I want to argue, the preverb of the complex predicate is a clause (which, however, does not contain an external argument).
The idea that preverbs of light verb constructions may contain clausal structures is not novel. For example, Tatevosov [2013] proposes that analytic causative constructions in Ossetian (Iranian, Indo-European) are instances of light verb constructions, where a preverb contains a clausal structure embedded under a nominal projection14. Thus, in his proposal, the infinitive clause in the next example is a nominalization, which is embedded under the light verb kodta, which corresponds to the transitive auxiliary awun in Lezgian.
(97) Alan [Aslan-i bad-in]m> kodta
A.nom A.-acc sit-inf do.pst.tr.3sg
'Alan was sitting Aslan down.' [Tatevosov 2013]
Another line of research has concentrated on investigating mechanisms of formation of complex predicates which are not instances of LVCs. Thus, Wurmbrand [2013] investigates properties of complex predicates in several languages (Chamorro, Spanish and others) and proposes that complex predicates may be formed by voice incorporation. Under this approach, the incorporated voice phrase doesn't project an external argument and is connected to the matrix voice phrase via an agree relation.
I want to unify these insights and propose that masdar OblCCs are light verb constructions where the preverb is a subjectless voice phrase. I take an agnostic position on whether the two Voice projections in masdar OblCCs are connected via voice incorporation or not.
Crucially, I want to propose that nothing prevents the embedded voice phrase from not projecting a specifier.
Indeed, such structures can be compared to possessive-raising structures. In English, DPs containing relational nouns may, and, in most cases, must contain a possessor:
(98) My/?an uncle will leave today.
14 However, he proposes that the light verb is connected with its preverb via a control relation, which is quite different from what I propose here.
However, in possessor raising constructions, such as (99), the DP contains relational noun (arm), whose possessor (John) is realized outside of the DP.
(99) John broke an arm. [Wood and Marantz 2017]
I take this as evidence that there is no syntactic requirement for a DP with a relational noun to contain a possessor.
Similarly, I assume that there is no syntactic requirement for a masdar clause to contain an Initiator15. In the next subsection, I will propose a system which formalizes the conditions where the Initiator can be realized non-locally.
To sum up, I propose that the oblique causer, both in inchoative and masdar OblCCs, is the realization of the Initiator semantic role, which is assigned low but is realized as an oblique causer higher in the structure, above the thematically void light verb.
At this point, the following questions may arise. First, why cannot an oblique causer be directly merged with a masdar clause or an agentive vP? In other words, what is the reason of the non-agentive requirement? Second, why cannot ergative subjects be merged with expletive v?
In the next subsection, I formalize several intuitions that I have formulated above and propose an account which is intended to answer these questions.
6. Voice and Voiceexpl as instances of i*
In the previous section I showed that oblique causers cannot be merged with agentive structures directly:
(100) *Kerim-awaj [ic basmis-na]agentwe K.-adel apple. abs crush-aor
int. 'Kerim accidentally crushed the apple.'
Similarly, ergative agents cannot be merged with expletive structures:
(101) *Kerim-a [ic basmis-un xa-na] expl K.-erg apple.abs crush-msd become-aor int. 'Kerim crushed the apple.'
15 I assume that the reason why a matrix clause in a direct constructions must contain an external argument is due to the fact that the EPP requirement of the matrix T cannot be otherwise satisfied.
I have also proposed that OblCCs are subject to the following rule:
(102) Structural condition on Voiceexpl
Voiceexpl can only be merged with structures which are non-agentive and which contain an assigned but non-projected Initiator semantic role.
In the previous subsection, I have shown that OblCCs share similarities with possessor raising constructions in that they realize an external argument far enough from the place where its semantic role is assigned. I will call such derivations semantic raising constructions.
Wood and Marantz [2017] discuss this and similar constructions (Icelandic figure reflexives, Japanese adversity causatives) in different languages. They propose that external arguments are introduced by one head which they call i*, different instances of which correspond to traditional VoiceP, ApplP, pP (and perhaps, others). The crucial assumption that they make is that the interpretation of the i* depends on the properties of the projection that it merges with. Thus, when merged with an agentive v the argument in the Spec i*P gets interpreted as a canonical agent; when merged with applicative morphology it gets interpreted as an applicative etc. Furthermore, they propose that in some cases, the i* projection is merged to a projection which doesn't project a semantic role.
I want to concentrate on this option and propose that Voiceexpl phrase, with which we are dealing in OblCCs, is an instance of a semantically vacuous i* (I borrow the term Voiceexpl from [Myler 2014]). I suggest that there are two characteristic features of this projection: first, since it doesn't merge with an agentive v, it is unable to assign an Initiator semantic role of its own. Second, it has a syntactic requirement to merge an adelative phrase in its specifier. Since the argument that Voiceexpl introduces must be nevertheless semantic marked (I assume there are no argument expletives in the language), from it follows that Voiceexpl can only be merged successfully in semantic raising environments. To recall, in these environments, the Initiator semantic role is generated quite far from the place where it is assigned to an argument (in our case, to an oblique causer). (I adopt Myler's [2016] distinction between Instant Gratification and Delayed Gratification and assume that OblCCs involve Delayed Gratification).
The adelative adposition that heads the AdP contained in the specifier of Voiceexpl, by assumption, is not a semantic assigner. I suggest that this adposition makes no semantic contribution (see [Collins 2004] who makes the same claim about the English preposition by in passive constructions). I leave the question open why this particular adposition is associated with OblCCs.
I assume that Voiceexpl has no special requirement to merge with an expletive v. However, it cannot merge with any other projection i* can merge with. To put it slightly differently, i* projection can merge with other projections, but only when it merges with an expletive v it can be interpreted as Voiceexpl and is able to merge an oblique causer: the semantic role in this case will be transmitted from a semantic assigner contained in the complement of the expletive v.
I will largely adopt Wood and Marantz's [2017] formalism. The most important assumption about the i* projection that I adopt is that it gets interpreted as Voice when merged with an agentive v. I will furthermore suggest that i* is interpreted as Voiceexpl (i. e., it makes no semantic contribution) when it is merged with a non-agentive v (technically, I will assume that Voiceexpl represents the elsewhere case: i* gets interpreted as Voiceexpl when it doesn't merge with either agentive v, applicative morphology or prepositions16). I will not provide semantic derivations here, while I will notate semantic marking in the tree.
Ergative marking is disallowed in OblCCs because the Voiceexpl, requires an adelative phrase in its specifier.
Thus, in this system, certain heads are semantic assigners which generate a semantic role which is percolated up the tree until it is assigned to the suitable argument. We will be only concerned with heads assigning Initiator semantic role. Such heads are telic verbs and agentive roots. This intuition is illustrated in the following diagram. The masdar phrase that it contains, assigns but doesn't realize the Initiator semantic role. Only when an expletive v is merged, an adelative phrase can be merged in the specifier of the expletive voice.
I will assume that the semantic interpretation characteristic for OblCCs arises from the fact that an OblCC contains an inchoative structure (which contributes 'change by itself' meaning, to which an oblique causer is merged (which contributes the 'Initiator' semantics).
16 The last two conditions are irrelevant in our case.
(103) Thematic structure of masdar OblCCs:
voiceexpiP
Initiator Missing
Now we are in the position to answer the questions raised on the previous subsection.
The reason why an oblique causer cannot merge with an agentive v is that if i* is merged with such a projection, it gets interpreted as Voice which will consequently introduce the canonical agent.
I assume that the reason why canonical agents cannot cannot merge with expletive v is because it has a requirement to merge an adposition in its specifier — the requirement that canonical agents fail to satisfy.
7. Conclusions
In this paper I discussed constructions with oblique causers. I proposed that oblique causer constructions are subject to two sorts of constraints: structural, disallowing them from from merging with agentive structures and several semantic constraints on the causer semantic role. I have proposed that OblCCs in Lezgian are light verb constructions. In the simple case, an LVC with an intransitive verb involves an inchoative structure. However, another option is also available: a light verb in Lezgian can take as a complement a voice phrase which doesn't project an external argument. If this analysis is on the track, oblique causer constructions can be regarded as a phenomenon of the same nature as possessor raising constructions, Japanese adversity causatives etc.
A light verb construction, in this system, can be seen as syntactic tool to extended thematic structure of the preverb, if the preverb cannot realize cannot realize its arguments locally.
There are several questions which I didn't discuss in the paper. First, I didn't discuss in detail semantic interpretation of OblCCs. Ganenkov et al. [2008] and Schäfer [2009] report that a range of semantic interpretations of OblCCs can be quite diverse: in addition to 'unintentionality' readings, other semantic interpretations are possible ('failed prevention readings' and 'success with difficulty readings', in their terminology). An elaborated semantic investigation is required to derive a theory which could account for these facts.
Second, I didn't discuss the morphological marking of oblique causers. While oblique causers are marked adelative in Lezgian, they are marked dative in German; in Icelandic oblique causers are prepositional phrases headed by the preposition hja.
Finally, in certain languages, such as Laz (Kartvelian) oblique causers trigger applicative morphology on the verb (see [Demirok 2017]). Note that the theory proposed here doesn't account for such variation. I leave elaboration of these and related questions for future research.
Abbreviations
abs — absolutive; acc — accusative; adel — adelative; aop — aorist participle; aor — aorist; dat — dative; def — definite; erg — ergative; fut — future; inf — infinitive; ins — instrumental; ipfv — imperfective; man — manner converb; msd — masdar; nom — nominative; poel — postelative; pst — past tense; rf — root form; srdir — superdirective; trs — transitive.
Bibliography
Alexiadou et al. 2006 — A. Alexiadou, E. Anagnostopoulou, F. Schäfer. The properties of anticausatives crosslinguistically // Phases of Interpretation 91, 2006. P. 187-211. Bosse et al. 2012 — S. Bosse, B. Bruening, M. Yamada. Affected experiencers //
Natural Language Linguistic Theory 30(4), 2012. P. 1185-1230. Collins 2005 — Chr. Collins. A smuggling approach to the passive in English //
Syntax 8(2), 2005. P. 81-120. Demirok 2017 — O. Demirok. Dative subjects in Laz: A modal approach.
Manuscript. Cambridge, MA: MIT, 2017. Folli, Harley 2008 — R. Folli, H. Harley. Teleology and animacy in external arguments // Lingua 118(2), 2008. P. 190-202.
Folli et al. 2005 — R. Folli, H. Harley, S. Karimi. Determinants of event type in Persian complex predicates // Lingua 115(10), 2005. P. 1365-1401.
Ganenkov et al. 2008 — D. Ganenkov, T. Maisak, S. Merdanova. Non-canonical agent marking in Agul // H. de Hoop, P. de Swart (eds.). Differential Subject Marking. Dordrecht: Springer, 2008.
Harley 2009 — H. Harley. The morphology of nominalizations and the syntax of vP // M. Rathert, A. Giannankidou (eds.). Quantification, Definiteness and Nominalization. Oxford: Oxford University Press, 2009. P. 320-342.
Haspelmath 1993 — M. Haspelmath. A Grammar of Lezgian. Berlin — New York: Mouton de Gruyter, 1993.
Ilkhanipour, Sugawara 2016 — N. Ilkhanipour, A. Sugawara. On the Semantics and Syntax of Persian 'become' // University of Pennsylvania Working Papers in Linguistics 22(1), 2016. P. 17.
Kim 2011 — K. Kim. High applicatives in Korean causatives and passives // Lingua 121(3), 2011. P. 487-510.
Myler 2016 — N. Myler. Building and interpreting possession sentences. Cambridge, MA: MIT Press, 2016.
Kittilä 2005 — S. Kittilä. A typology of involuntary agent constructions // Word 56(3), 2005. P. 381-419.
S^b0 2001 — K. J. S^b0. An analysis of the anticausative alternation. Ms. Universitetet i Oslo, Oslo, 2001.
Schäfer 2009 — F. Schäfer. The oblique causer construction across languages // Proceedings of NELS 38, 2009. P. 297-308.
Schäfer 2012 — F. Schäfer. Two types of external argument licensing — the case of causers // Studia Linguistica 66(2), 2012. P. 128-180.
Sugimura 2012 — M. Sugimura. A study of Japanese light verb construction and its implications for nominal architecture Mina Sugimura // Proceedings of ConSOLE XVII 289, 2012. P. 298.
Svenonius 2012 — P. Svenonius. Structural decomposition of spatial adpositions. Work. pap., Dep. Linguist., Univ. Troms0, 2012.
Tatevosov 2013 — S. Tatevosov. Causative in Ossetian: a complex predicate analysis. ICIL5 handout. 2013.
Wood, Marantz 2017 — J. Wood, A. Marantz. The interpretation of external arguments // R. D'Alessandro, I. Franco, Á. J. Gallego (eds.). The Verbal Domain. Oxford: Oxford University Press, 2017. P. 255-278.
Wood 2013 — J. Wood. The unintentional causer in Icelandic // Proceedings of the Forty-First Annual Meeting of the North East Linguistic Society. Vol. 2, 2013. P. 273-286.
