Abstract

The article discusses the OC-NC generalization (Landau 2004, 2015) that establishes a relationship between the obligatory control nature of a construction and its semantic (tensedness/attitude) and formal (morphological agreement) properties. I discuss three counterexamples attested in languages of the Caucasus and demonstrate that complements of desiderative verbs in these languages violate predictions of the generalization: agreeing tensed complements instantiate obligatory control, or uninflected tensed complements are noncontrolled. I also give references to other potential counterexamples and discuss issues raised by the empirical evidence, ultimately arguing that no such typological universal holds.

1 The OC-NC Generalization

Landau’s (2004, 2006) crosslinguistic research into the syntax of control structures has established a relationship between the obligatory control nature of a construction, on the one hand, and its semantic and formal properties, on the other hand. The initial formulation proposed that semantic tense (understood as the possibility of independent temporal modification of the embedded verb) and agreement morphology on the embedded verb determine whether a construction is an obligatory control (OC) or no control (NC) structure, as shown in (1).

(1) The OC-NC generalization (Landau 2015:7) In a fully specified complement clause (i.e., a clause in which the I head carries slots for both [T] and [Agr]):

  • If the I head carries both semantic tense and agreement ([+T, +Agr]), NC obtains.

  • Elsewhere, OC obtains.

According to Landau’s generalization, OC constructions represent an elsewhere case; that is, OC PRO is licensed in environments where lexical DPs and pro fail to be licensed. The latter, according to the generalization, require the presence of semantic tense and morphological agreement on the embedded predicate. Failure to satisfy either of the two conditions leads to the OC pattern. Importantly, Landau’s generalization offers clear typological implications, as spelled out in (2).

(2) Typological predictions about OC complements (Landau 2015:8)

  • Untensed complements (excluding small clauses) will universally be in the OC class, regardless of inflection.

  • Uninflected complements (i.e., [−Agr]) that are nevertheless non-defective (so a [T] slot is present) will universally be in the OC class, regardless of semantic tense.

More recently, Landau (2015) has reformulated the generalization in line with his latest theoretical developments in the analysis of OC. Specifically, he reformulates the difference between two familiar types of OC—OC into tensed and untensed complements (Landau 2000, 2004)—in terms of (non)attitude complements. Landau’s revised generalization is given in (3).

(3)

  • Landau’s revised generalization (Landau 2015:20)

  • [+Agr] blocks control in attitude complements but not in non-attitude complements.

  • Or: [+Agr] blocks logophoric control but not predicative control.

Remarkably, as Landau (2015:7) explicitly states, neither of these two formulations needs a reference to factors other than attitude/tensedness and agreement, such as finiteness (e.g., finite but not nonfinite), a specific ϕ-feature (e.g., person but not gender or number), or structural size of the complement (e.g., CP but not TP).1

Empirical support for the proposed generalization has been adduced from languages belonging to different branches of Indo-European as well as from Basque (isolate), Turkish (Turkic), Hungarian (Finno-Ugric), Arabic (Semitic), and some other language families. Overall, the OC-NC generalization is claimed to be “the only successful description of patterning of OC and NC complements” (Landau 2015:8) and thus receives considerable theoretical significance: a theory of control that can derive it would clearly be superior to those theories that cannot. In fact, both theories of control proposed by Landau—his Agree model (2004) and his two-tiered model (2015)—were developed with this goal in mind, so that the derivation of the generalization is hardwired into their design.

The goal of the current article is to discuss three counterexamples to Landau’s generalization attested in languages of the Caucasus and ultimately to argue that no such typological universal holds. In section 2, I begin with an extended discussion of Khuduts Dargwa, a language of the Nakh-Daghestanian family. I demonstrate that agreeing complements of desiderative verbs instantiate OC in this language, thus violating Landau’s generalization. In section 3, I document a similar counterexample from Georgian (Kartvelian). In section 4, I show that Udi, another Nakh-Daghestanian language, exhibits another violation of Landau’s generalization: its uninflected tensed complements are noncontrolled. In section 5, I give references to other potential counterexamples and discuss issues raised by the empirical evidence presented in the article.

2 [+Agr] in Khuduts Dargwa OC Complements

Khuduts is a language of the Dargwa branch of the Nakh-Daghestanian family spoken by around 1,200 native speakers living in or originating from the village of Khuduts (Dakhadaevskij rajon, Daghestan, Russia). It is a morphologically ergative, pro-drop, SOV language.

2.1 Basics of Khuduts Agreement

Like virtually all languages of the Nakh-Daghestanian family, Khuduts possesses verbal gender agreement, which, together with case marking, operates on an ergative basis: intransitive subjects and transitive objects in the (unmarked) absolutive case determine gender agreement on the verb, whereas transitive subjects in the (marked) ergative case never do so (see Van den Berg 2005:155–159 for an overview of gender agreement in Nakh-Daghestanian). Argument case marking and verbal gender agreement are exemplified in (4) and (5).2

(4)

  • rursːi r-isː-ul=ca-r

  • girl(ABS) F.SG-cry:IPF-CVB=AUX-F.SG

  • ‘The girl is crying.’

(5)

  • rasul-li ħinc-bi {d/*w}-iʕ-un

  • Rasul-ERG apple-PL(ABS) N.PL/*M.SG-steal:PF-AOR

  • ‘Rasul (male name) stole the apples.’

Not all verb stems can have a gender agreement prefix/infix: each verb stem is lexically specified to either host or not host a gender agreement slot. The following example shows the perfective stem of the verb ‘go’, which has no agreement prefix and thus looks the same regardless of the gender of the absolutive argument:

(6)

  • {rursːi / gal / q’waʕl} ag-ur

  • girl(ABS) boy(ABS) cow(ABS) go:PF-AOR

  • ‘The girl / boy / cow went away.’

For verb stems that do host such a slot, gender agreement is obligatory and appears on all verbal forms, both finite and nonfinite; the appearance of such verb stems without a gender marker is simply ungrammatical. In other words, morphologically, the gender prefix may be conceived of as part of the verbal stem. In the absence of a clausemate absolutive argument, the gender agreement slot on the verb is filled with the default (neuter singular) marker b-.3 The gender agreement prefix in Khuduts reflects both gender (masculine, feminine, or neuter) and number (singular or plural) features of the absolutive argument; see table 1.

Table 1

Morphological exponence of gender-number agreement in Khuduts

NFM
Sg b- r- w- / zero 
Pl d- b- b- 
NFM
Sg b- r- w- / zero 
Pl d- b- b- 

Moreover, Khuduts has what can be called person agreement parasitic on gender agreement; that is, person features of some arguments can be seen in the system of gender agreement. More specifically, morphological exponence for agreement with the plural personal pronouns nusːa ‘we’ and ʕušːa ‘you all’ is different from the exponence with other human plural arguments. The latter require the human plural marker b-, as shown in table 1, whereas first and second person plural pronouns obligatorily employ the marker d-, as in (7).

(7)

  • nusːa d-isː-ul=da

  • we(ABS) 1/2PL-cry:IPF-CVB=1

  • ‘We are crying.’

The gender agreement marker on the verb thus makes gender, number, and, in the plural, [participant] features of the clausemate absolutive argument visible.4 Example (7) also shows that Khuduts features person agreement proper expressed in the suffixal position.5 Unlike gender agreement, person agreement is restricted to finite contexts, with one exception to be shown below. Either the subject or the direct object may serve as the controller of person agreement, whichever is higher on the person hierarchy 2 > 1 > 3; that is, a second person subject or direct object, when present, always controls person agreement. In the absence of second person arguments, it is the first person subject or direct object that triggers person agreement. Third person agreement appears only in sentences where both core arguments are third person.6

The two subsystems of verbal agreement are independent of each other and cross-reference ϕ-features of different arguments, as shown in (8).7

(8)

  • du sːika b-uc-ib-da

  • I(ERG) bear(ABS) N.SG-catch:PF-AOR-1

  • ‘I caught a bear.’

Here, the gender agreement prefix b- (neuter singular) shows the gender and number of the absolutive direct object (‘bear’), whereas the first person marker -da cross-references the person of the ergative subject.

In what follows, I assume that the person-number probe is associated with T0 in Khuduts, whereas the gender-number probe is located on v0. Primary evidence about the structural locus of person agreement comes from an observation in the literature on Dargwa languages (Sumbatova and Lander 2014:118–119) that person agreement markers and overt tense markers occupy the same morphological slot in the verb. This is seen especially clearly in the progressive and resultative, where the existence of pairs of present and past forms allows us to detect the relative morphological position of tense and person markers. Present tense forms of the progressive and resultative are overtly marked for person, but have no overt (present) tense marking. Past tense forms, however, demonstrate the opposite: they feature an overt past tense marker, but do not show person distinctions.8 Examples (9) and (10) illustrate the progressive forms.

(9) Present progressive: No overt tense marking, overt person inflection

  • b-irq’-ul=da

    N.SG-do:IPF-CVB=1

    ‘(I) am doing (it).’

  • b-irq’-ul=di

    N.SG-do:IPF-CVB=2SG

    ‘(You) are doing (it).’

  • b-irq’-ul=ca-b

    N.SG-do:IPF-CVB=AUX-N.SG

    ‘(She / He) is doing (it).’

(10)

  • Past progressive: Overt tense marking, no overt person inflection

  • b-irq’-ul=di

  • N.SG-do:IPF-CVB=PST

  • ‘(I / You / We / You all / He / She / They) was / were doing (it).’

This complementary distribution of (finite) tense and person marking naturally follows if we assume that they represent the spell-out of the same functional head T0. The morphological resources of the language do not allow the overt expression of both tense and person, so person marking can appear when tense is zero-marked but is blocked when tense is overtly marked.

With regard to gender-number agreement, low (vP) nominalizations are usually used as a primary diagnostic of the locus of gender agreement in Nakh-Daghestanian languages (see Gagliardi et al. 2014 for Lak and Tsez, Polinsky 2016 for Archi). Such evidence is not available in Khuduts, which apparently lacks low nominalizations. It is clear, however, that case licensing and gender-number agreement, unlike person agreement, are not related to finiteness, as seen from the ability of (high?) nominalizations to agree in gender-number and license appropriate cases on clausal arguments; see (11) and (12).

(11)

  • aba-j b-uχːu-l=ca-b [nusːa sa-d-ač’-ni ]

  • mother-DATN.SG-know:IPF-CVB=AUX-N.SG we(ABS) PV-1/2PL-come:PF-NMLZ

  • ‘Mother knows that we came.’

(12)

  • aba-j b-uχːu-l=ca-b [atːa-l q’uʕl-i d-ic-ni ]

  • mother-DATN.SG-know:IPF-CVB=AUX-N.SG father-ERG cow-PL(ABS) N.PL-sell:PF-NMLZ

  • ‘Mother knows that Father sold the cows.’

On the basis of these examples, I conclude that the gender-number probe resides either on v0 or on defective (nonfinite) T0. Further research is definitely needed; for the time being, none of these options would affect the following discussion.

2.2 Infinitival Complementation in Khuduts

Khuduts Dargwa possesses two different verb forms that can be used in complements of modal, aspectual, implicative, and desiderative verbs (the following discussion is confined to desiderative verbs). One form, traditionally called “infinitive,” bears (like any other verb form in Khuduts) obligatory gender agreement with the clausemate absolutive argument (subject of intransitive embedded verb or direct object of transitive embedded verb), as (13a–b) demonstrate.

(13) Desiderative verbs: ‘want’

  • madina-j [ħinc-bi d-ic-ij ] b-ikː-ul=ca-b

    Madina-DAT apple-PL(ABS) N.PL-sell:PF-INFN.SG-want:IPF-CVB=AUX-N.SG

    ‘Madina (female name) wants to sell the apples.’

  • madina-j [ka-r-iž-ij ] b-ikː-ul=ca-b

    Madina-DATPV-F.SG-sit:PF-INFN.SG-want:IPF-CVB=AUX-N.SG

    ‘Madina wants to sit down.’

In (13a), the embedded verb is transitive and thus agrees with its absolutive direct object; in (13b), the embedded verb is intransitive and therefore agrees with its (unexpressed) absolutive subject. Below, I refer to this form as the gender-inflected infinitive.

In addition, Khuduts, as well as a few neighboring Dargwa languages, has developed a new infinitive on the basis of an adverbial form originally expressing purpose (‘in order to’), which also can appear in complements of modal, aspectual, implicative, and desiderative verbs. Again, like any other verb form in Khuduts, it includes an obligatory gender agreement prefix. An important difference regarding this new infinitive is that it also includes a morphological slot for person agreement, the only nonfinite verb form that does so in Khuduts. This is shown in (14); compare (13).

(14) Desiderative verbs: ‘want’

  • ba-j [ħinc-bi d-ic-ar-aj ] b-ikː-ul=ca-b

    mother-DAT apple-PL(ABS) N.PL-sell:PF-3-INFN.SG-want:IPF-CVB=AUX-N.SG

    ‘Mother wants to sell the apples.’

  • madina-j [ka-r-ig-ar-aj ] b-ik:-ul=ca-b

    Madina-DATPV-F.SG-sit:PF-3-INF N.SG-want:IPF-CVB=AUX-N.SG

    ‘Madina wants to sit down.’

In what follows, I refer to this form as the person-inflected infinitive. As with finite verbs, the prefixal and suffixal agreement slots are independent of each other. With intransitive embedded verbs, both cross-reference the ϕ-features of the absolutive embedded subject, as in (14b). With transitive verbs, prefixal agreement is always with the absolutive direct object, whereas suffixal agreement depends on the relative person specification of the subject and direct object, as shown in (15) and (16).

(15)

  • ʕui q’ast b-arq’-ib-di-w [Δi ħinc-bi d-iʕ-atː-aj ]

  • you.SG(ERG) decision N.SG-do:PF-AOR-2SG-QERG apple-PL(ABS) N.PL-steal:PF-2-INF

  • ‘Did you decide to steal apples?’

(16)

  • atːa-li q’ast b-arq’ib [Δi ʕu ʡaħ r-arq’-itː-aj ]

  • father-ERG decision N.SG-do:PF-AORERG you.SG(ABS) good F.SG-do:PF-2-INF

  • ‘Father decided to heal you.’

In (15), person agreement on the embedded verb is with the null embedded subject, since the latter is higher on the person hierarchy than the embedded direct object (‘apples’). By contrast, in (16), the embedded direct object (‘you’) is higher than the null embedded subject and thus controls person agreement in the embedded clause. We thus see that both infinitives have obligatory prefixal agreement reflecting gender, number, and [participant] features of the clausemate absolutive argument. In addition to gender agreement, the person-inflected infinitive features a separate person agreement slot reflecting the person of the argument higher on the hierarchy 2 > 1 > 3.

It is not a straightforward task to determine the predictions of Landau’s (2004, 2015) approach with respect to the control behavior of such agreeing complements. Both versions of Landau’s theory handle control structures in languages with pure subject agreement structurally represented as features of a ϕ-probe on T0.9 Neither gender agreement nor person agreement in Khuduts is subject-oriented: gender agreement is determined on the basis of morphological case and thus can be with the subject or the direct object; person agreement is regulated by a person hierarchy and therefore can also be controlled by either of the two core arguments.

In a similar vein, both modifications of Landau’s approach do not explicitly address the special role of T0: would another agreeing functional head in the clausal spine—say, v0—play a similar role in the distribution of OC and NC patterns? As far as I can see, it would, and the identity of the agreeing functional head would make no difference in either the Agree theory (2004) or the two-tiered theory (2015).

Assuming that the crux of both versions of Landau’s theory lies in the valuation of an agreeing functional head by the subject’s ϕ-features, one could extend the scope of the theory: namely, one could suggest that it predicts that whenever a functional head (be it v0 or T0) in the tensed/attitude embedded clause receives the subject’s ϕ-features, the embedded clause will be noncontrolled. By contrast, agreement with other arguments in the embedded clause must have no effect on control properties of the embedded clause.

If this interpretation is on the right track, Landau’s generalization predicts that tensed/attitude infinitival complements of desiderative verbs should display different behavior in Khuduts depending on the transitivity of the embedded predicate and the relative position of the core argument on the person hierarchy. Transitive complements where gender agreement is with the direct object should be OC structures, whereas intransitive complements where agreement is with the null subject should be NC structures. In a similar vein, the generalization predicts that person agreement may also affect the OC or NC status of the person-inflected infinitival complement, depending on the particular configuration of the person values of the subject and direct object.

In what follows, I concentrate on intransitive infinitival complements where both gender and person agreement on the infinitive show the ϕ-features of the null absolutive subject, which, according to Landau’s generalization, should lead to the absence of OC. I argue that neither gender-inflected nor person-inflected infinitival complements in Khuduts obey Landau’s generalization: tensed (attitude) infinitival complements are uniformly OC structures.10

2.3 Diagnosing OC

For the purposes of this article, I have checked seven desiderative predicates in Khuduts: bikː- ‘want’ (stative), hut bik’wij ‘strive, be eager’, razi biχwij ‘agree’, uruχ biχwij ‘become afraid’, ħadur biχwij ‘get ready’, q’ast barq’ij ‘decide’, and pikri barq’ij ‘think’ (the last verb takes an infinitival complement only in the meaning ‘plan, intend’, never in its use as a propositional verb). For reasons of space, I only provide examples with the matrix verb q’ast barq’ij ‘decide’.11 If not specifically indicated, other desiderative verbs listed above demonstrate identical behavior in all relevant aspects. (17) provides a baseline example of infinitival complementation with desiderative verbs (when showing identical behavior, gender- and person-inflected infinitives are both given in a single example, separated by a slash).

(17)

  • aba-l [urkaraqi r-uq’ʕ-ij / r-uq’-aʕn-aj ] q’ast b-arq’-ib

  • mother-ERG to.Urkaraq F.SG-go:PF-INFF.SG-go:PF-3-INF decision N.SG-make:PF-AOR

  • ‘Mother decided to go to Urkaraq.’

The presence of a null subject in such examples is evidenced by two facts. First, only absolutive arguments trigger gender agreement on lexical verbs in Khuduts—no feminine absolutive argument besides the embedded subject can be identified in (17). Second, the embedded subject may be lexicalized under focus in the form of an overt reflexive or personal pronoun, as in (18).

(18)

  • abal-li [ca-ri=gina urkaraqi r-uq’ʕ-ij / r-uq’-aʕn-aj ] q’ast

  • mother-ERG self(ABS)-F=only to.Urkaraq F.SG-go:PF-INFF.SG-go:PF-3-INF decision

  • b-arq’-ib

  • N.SG-make:PF-AOR

  • ‘Mother decided to go to Urkaraq herself / alone.’

In this example, the infinitival complement features its own subject in the absolutive case, as required by the embedded intransitive verb buq’ʕij ‘go’, different from the matrix subject in the ergative case. The embedded subject, however, still must be covalued with the matrix subject, and disjoint reference is impossible. That the absolutive case of the reflexive in (18) indeed stems from selectional requirements of the embedded predicate can be seen from the fact that only dative case of the reflexive is grammatical with embedded dative subject verbs, as in (19).12

(19)

  • aba-li [cin-iji=gina prezident či-w-až-ij / či-w-ag-ar-aj ]

  • mother-ERG self-DAT=only president(ABS) PV-M.SG-see:PF-INFPV-M.SG-see:PF-3-INF

  • q’ast b-arq’-ib

  • decision N.SG-make:PF-AOR

  • ‘Mother decided to see the president herself.’

The Agree version of Landau’s generalization (2004) applies to complements that are both inflected for the ϕ-features of the subject and tensed. With regard to ϕ-feature inflection in infinitival complements, a full match in all ϕ-features is required between the matrix subject and the embedded subject in intransitive infinitival complements of desiderative verbs. A partial match (e.g., only in gender but not number or person) is impossible.13 See the examples in (20), where the agreement shown is the only grammatical option.

(20)

  • atːa-lii urkaraqi uq’ʕ-ij / uq’-aʕn-aj ] q’ast

    father-ERGABS to.Urkaraq (M.SG)go:PF-INF (M.SG)go:PF-3-INF decision

    b-arq’-ib

    N.SG-make:PF-AOR

    ‘Father decided to go to Urkaraq.’

  • {rursː-ba-l / gul-a-l } [Δi urkaraqi b-uq’ʕ-ij / b-uq’-aʕn-aj ]

    girl-PL-ERG boy-PL-ERGABS to.Urkaraq M/F.PL-go:PF-INFM/F.PL-go:PF-3-INF

    q’ast b-arq’-ib

    decision N.SG-make:PF-AOR

    ‘The {girls / boys} decided to go to Urkaraq.’

  • ʕušːaii urkaraqi d-uq’ʕ-ij / d-uq’-aʕtː-aj ] q’ast

    you.PL(ERG) ABS to.Urkaraq 1/2PL-go:PF-INF 1/2PL-go:PF-2-INF decision

    b-arq’-ib-da

    N.SG-make:PF-AOR-2PL

    ‘You decided to go to Urkaraq.’

The following example demonstrates that infinitival complements of desiderative verbs may host a future-denoting temporal adverbial and thus should be diagnosed as tensed:14

(21)

  • aba-l [čːaʕal urkaraqi r-uq’ʕ-ij / r-uq’-aʕn-aj ] q’ast

  • mother-ERG tomorrow to.Urkaraq F.SG-go:PF-INFF.SG-go:PF-3-INF decision

  • b-arq’-ib

  • N.SG-make:PF-AOR

  • ‘Mother decided to go to Urkaraq tomorrow.’

Landau’s two-tiered theory of control (2015) deals with matrix verbs in terms of attitude vs. nonattitude rather than tensedness. The Khuduts desiderative verbs listed above have been tested for the de re / de dicto ambiguity (Landau 2015:18–19, Pearson 2015) indicative of attitude contexts, and they appear to yield the same entailment pattern as their English counterparts when one definite description is substituted for another in such contexts. Imagine the following scenario. There are two friends, Rasul and Ali. When they were young and lived in their native village, they spent a lot of time together and often threw vodka-drinking parties. At some point, Rasul moved far away from the village for a new job. After many years away and without having maintained any contact, he goes back and one of the important points on his to-do list is to throw a party to drink vodka with Ali again. What he does not know, however, is that in the years that have elapsed, Ali has become a totally different person: he is very religious and now serves as a mullah at the local mosque, and thus no longer drinks vodka at all. (22) is a truthful description of Rasul’s intentions, whereas (23) does not truthfully describe his epistemic/bouletic state of mind.

(22)

  • rasul-li [ʕali-cːil ʕaraq’i d-erčː-ij / d-erčː-ar-aj ] q’ast

  • Rasul-ERG Ali-COM vodka(ABS) N.PL-drink:PF-INFN.PL-drink:PF-3-INF decision

  • b-arq’-ib

  • N.SG-make:PF-AOR

  • ‘Rasul decided to drink vodka with Ali.’

(23)

  • rasul-li [malla-cːil ʕaraq’i d-erčː-ij / d-erčː-ar-aj ] q’ast

  • Rasul-ERG mullah-COM vodka(ABS) N.PL-drink:PF-INFN.PL-drink:PF-3-INF decision

  • b-arq’-ib

  • N.SG-make:PF-AOR

  • ‘Rasul decided to drink vodka with a / the mullah.’

As can be seen, the definite description ‘Ali’ cannot be substituted for the definite description ‘mullah’ inside the infinitival complement of a desiderative verb without changing the truth value, which shows that desiderative verbs in Khuduts are attitude verbs and must thus be within the scope of Landau’s two-tiered theory of control (2015).

Infinitival complements of desiderative verbs are thus attitude, tensed, and inflected complements and therefore must be subject to Landau’s generalization in both the 2004 and 2015 versions; that is, they are predicted to be noncontrolled when agreeing with the null subject in gender and/ or person. Contrary to this prediction, empirical evidence suggests that they constitute a textbook example of OC.

The two types of infinitival complements behave identically with respect to OC diagnostics. The only difference between them concerns availability of disjoint embedded subjects expressed by an overt DP.15 Gender-inflected complements of all desiderative verbs uniformly prohibit referentially disjoint overt subjects, as seen in (24).

(24)

  • *aba-l [rasul urkaraqi uq’ʕ-ij ] q’ast b-arq’-ib

  • mother-ERG Rasul(ABS) to.Urkaraq (M.SG)go:PF-INF decision N.SG-make:PF-AOR

  • Intended: ‘The mother decided that Rasul would go to Urkaraq.’

Instead, a different complementation pattern with a perfective converb must be used, as in (25).

(25)

  • aba-l [rasul urkaraqi ag-ur-ri ] q’ast b-arq’-ib

  • mother-ERG Rasul(ABS) to.Urkaraq go:PF-AOR-CVB decision N.SG-make:PF-AOR

  • ‘The mother decided that Rasul would go to Urkaraq.’

However, person-inflected complements of different desiderative verbs behave differently. The verb razi biχwij ‘agree’ allows a disjoint overt embedded subject without restrictions, as seen in (26).

(26)

  • rasul razi w-iχ-ub [aba čːaʕal urkaraqi

  • Rasul(ABS) agree M.SG-LV:PF-AOR mother(ABS) tomorrow to.Urkaraq

  • r-uq’-aʕn-aj ]

  • F.SG-go:PF-3-INF

  • ‘Rasul agreed that Mother would go to Urkaraq tomorrow.’

Disjoint overt subjects of person-inflected complements with the verbs bikː- ‘want’ and pikri barq’ij ‘plan, intend’ are considered marginally acceptable by some speakers (three out of five consultants) but only when in scope of the focus-sensitive clitic =gina ‘only, alone’, as in (27). Other speakers (two out of five consultants) find such examples unacceptable. None of the five consultants accepts nonfocused lexical embedded subjects in person-inflected complements of these verbs.

(27)

  • ???rasul-li-j b-ikː-ul=ca-b [madina=gina čːaʕal

  • Rasul-OBL-DATN.SG-want:IPF-CVB=AUX-N.SG Madina(ABS)=only tomorrow

  • urkaraqi r-uq’-aʕn-aj ]

  • to.Urkaraq F.SG-go:PF-3-INF

  • ‘Rasul wants only Madina to go to Urkaraq tomorrow.’

The remaining verbs hut bik’wij ‘strive, be eager’, uruχ biχwij ‘become afraid’, ħadur biχwij ‘get ready’, and q’ast barq’ij ‘decide’ allow no DPs, either plain or focused, in the position of the subject of a person-inflected complement, as shown in (28).

(28)

  • *rasul-li q’ast b-arq’-ib [madina(=gina) urkaraqi

  • Rasul-ERG decision N.SG-make:PF-AOR Madina(ABS)=only to.Urkaraq

  • r-uq’-aʕn-aj ]

  • F.SG-go:PF-3-INF

  • Intended: ‘Rasul decided that (only) Madina would go to Urkaraq.’

This restriction on disjoint overt subjects equally concerns first and second person arguments, as seen in (29).

(29)

  • *rasul-li q’ast b-arq’-ib [ʕu(=gina) urkaraqi

  • Rasul-ERG decision N.SG-make:PF-AOR you.SG(ABS)=only to.Urkaraq

  • r-uq’-aʕtː-aj ]

  • F.SG-go:PF-2-INF

  • Intended: ‘Rasul decided that (only) you would go to Urkaraq.’

Importantly, the matrix subject is the only eligible antecedent for the null embedded subject independently of whether or not an overt disjoint subject is allowed, as confirmed by a number of diagnostics.16 First, only sloppy readings are allowed under complement ellipsis, as shown in (30).

(30) ʕui=ra q’ast b-arq’-ib-di [Δi r-uq’ʕ-ij / you.SG(ERG)=ADD decision N.SG-make:PF-AOR-2SGABSF.SG-go:PF-INF r-uq’-aʕtc-aj ], aba-lj=ra q’ast b-arq’-ib F.SG-go:PF-2-INF mother-ERG=ADD decision N.SG-make:PF-AOR

  • ‘You decided to go, and Mother decided (to go), too.’

  • *‘You decided to go, and Mother decided (that you would go), too.’

Second, the null embedded subject cannot be interpreted as referentially disjoint from the matrix subject, nor can it have an arbitrary reading, as shown in (31)–(33).

(31)

  • *aba-lij urkaraqi r-uq’ʕ-ij / r-uq’-aʕn-aj ] q’ast

  • mother-ERGABS to.Urkaraq F.SG-go:PF-INFF.SG-go:PF-3-INF decision

  • b-arq’-ib

  • N.SG-make:PF-AOR

  • Intended: ‘Motheri decided that shej would go to Urkaraq.’

(32)

  • *aba-lij urkaraqi r-uq’ʕ-ij / r-uq’-aʕt:-aj ] q’ast

  • mother-ERGABS to.Urkaraq F.SG-go:PF-INFF.SG-go:PF-2-INF decision

  • b-arq’-ib

  • N.SG-make:PF-AOR

  • Intended: ‘Motheri decided that youj would go to Urkaraq.’

(33)

  • *aba-liarb urkaraqi b-uq’ʕ-ij / b-uq’-aʕn-aj ]

  • mother-ERGABS to.Urkaraq M/F.PL/N.SG-go:PF-INFM/F.PL/N.SG-go:PF-3-INF

  • q’ast b-arq’-ib

  • decision N.SG-make:PF-AOR

  • Intended: ‘Motheri decided that onearb would go to Urkaraq.’

Third, the antecedent must c-command the null embedded subject, as shown in (34).

(34)

  • *rasul-lai aba-lji urkaraqi uq’ʕ-ij / uq’-aʕn-aj ]

  • Rasul-GEN mother-ERGABS to.Urkaraq (M.SG)go:PF-INF (M.SG)go:PF-3-INF

  • q’ast b-arq’-ib

  • decision N.SG-make:PF-AOR

  • Intended: ‘Rasuli’s motherj decided that hei would go to Urkaraq.’

Fourth, the null embedded subject must find an antecedent in the immediately dominating clause and cannot be associated with an antecedent two clauses up, as illustrated in (35).

(35)

  • *rasul-li-j b-uχː-ul=ca-b [aba-l [Δi urkaraqi

  • Rasul-OBL-DATN.SG-know:IPF-CVB=AUX-N.SG mother-ERGABS to.Urkaraq

  • uq’ʕ-ij / uq’-aʕn-aj ] q’ast b-arq’-ni ]

  • (M.SG)go:PF-INF (M.SG)go:PF-3-INF decision N.SG-make:PF-NMLZ

  • Intended: ‘Rasuli knows that Motherj has decided that hei will go to Urkaraq.’

An anonymous reviewer notes that constructions with embedded pro also often prefer a closer antecedent and that (34) and (35) thus do not necessarily show the OC PRO nature of the null embedded subject.17 Note, however, that the closest higher argument is excluded from the set of antecedents in those examples due to gender-number agreement on the embedded verb: the latter shows masculine singular agreement, whereas the closest antecedent, ‘mother’, is feminine singular. The only masculine singular argument in the sentence is the more distant DP Rasul, which makes it the only possible antecedent for the null subject of the embedded infinitive. The sentences are nevertheless not grammatical, which proves that non-c-commanding or nonclosest c-commanding arguments are not eligible antecedents of the infinitival subject.18

More generally, if an antecedent other than the closest matrix subject were ever in principle possible, we would expect to observe mismatches between the ϕ-features of the matrix subject and the gender-person agreement on the intransitive embedded verb. The robust fact of the grammar of Khuduts, however, is that ϕ-features appearing on the intransitive infinitive (partial control factored out; see footnote 13) can never be different from the ϕ-features of the closest matrix subject.

On the basis of the diagnostics above, we can conclude that both types of complements with desiderative verbs—gender- and person-inflected infinitives—instantiate OC with both transitive and intransitive embedded verbs. This is unexpected: according to (my interpretation of) Landau’s generalization, the constellation of properties we observe in Khuduts infinitival complements should lead to NC or OC, depending on the transitivity of the embedded verb.

Importantly, this pattern is not just a quirk of Khuduts; rather, it is a trait typical of the whole family, at least in the part concerning the behavior of gender-inflected infinitives (person-inflected infinitives are restricted to Khuduts and a few neighboring Dargwa languages). A Nakh-Daghestanian language typically has only gender-number agreement and no person agreement. Gender agreement across Nakh-Daghestanian looks very much the same as described above for Khuduts: verb stems are lexically specified as either having or not having a gender agreement prefix; those verb stems that do have the prefix must have it in all verb forms, finite and nonfinite; and ϕ-features shown by the agreement prefix include gender and number of the clausemate absolutive argument (an extensive discussion of gender agreement in the distantly related language Archi of the Lezgic branch can be found in Bond et al. 2016).

As in Khuduts, the presence of gender-number agreement in infinitival complements of desiderative verbs does not automatically lead to NC in other Nakh-Daghestanian languages either. The variation may concern availability of disjoint lexical subjects in complements of some individual verbs (similar to what is observed in Khuduts person-inflected complements). For example, languages of the Avar-Andic and Tsezic branches typically allow lexical embedded subjects in infinitival complements of the verb ‘want’, whereas Tabasaran (Lezgic branch) does so in infinitival complements with the verb ‘be afraid’. The nature of the null subject in such complements, pro or PRO, requires further investigation. However, the majority of desiderative predicates in these languages still ban overt lexical DPs in the position of the infinitival subject and can only host a bound PRO or reflexive/pronominal subject (see the discussion of Tsez matrix verbs in Polinsky 2015).

3 [+Agr] in Georgian OC Complements

Georgian is a morphologically split-ergative Kartvelian language with pro-drop and person agreement. Person agreement in transitive finite sentences is determined by the first or second person direct object or, in case the direct object is third person, by the first or second person subject (see Harris 1981:28–32 for details of person agreement). All intransitive sentences and a subset of transitive sentences thus show person agreement with the subject.

According to Vamling’s (1989:60) description, desiderative verbs like vpikrob ‘think, intend’, vap’ireb ‘intend’, vp’irdebi ‘promise’, vubrʒaneb ‘order’, movitxov ‘demand’, varc’muneb ‘persuade’, msurs ‘want’, and vgegmav ‘plan’ take a subjunctive (optative) complement where the verb bears agreement marking for person-number. The controller of person agreement in subjunctive complements is determined in the same way as in indicative clauses. As Landau’s generalization predicts, many desiderative predicates allow noncontrolled complements with either a lexical subject or pro, as shown in (36).

(36)

  • m-inda, rom (ma-n) ga-ak’et-o-s

  • I-want that he-ERGPV-make-OPT-3

  • ‘Ii want that hej / proj shall do it.’

  • (Vamling 1989:86)

However, some desiderative verbs allow only a controlled reading of the null embedded subject and require that the subjunctive complement bear agreement features identical to those of the matrix subject. Vamling (1989:82–87) indicates that the subject control verb vap’ireb ‘intend’ and several object control verbs—for example, vubrʒaneb ‘order’, nebas vrtav ‘permit’, vemudarebi ‘beg’, and varc’muneb ‘persuade’—require obligatory coreference between the matrix argument (subject or object, respectively) and the embedded subject of the subjunctive complement. I have independently confirmed the pattern reported by Vamling for the verb vap’ireb ‘intend’ (other verbs have not been tested).

This verb takes tensed complements, as evidenced by the possibility of a temporal mismatch between the matrix and complement clauses. Subjunctive complements feature a subject argument that can either materialize in the form of a personal pronoun or remain unexpressed, as in (37).

(37)

  • samšabat-s me v-ap’irebd-i, rom (me) st’at’ia ga-m-e-gzavn-a

  • Tuesday-ADV I(NOM) 1-intend-PST that I(NOM) article(NOM)PV-1-PLUPRF-send-3

  • xutšabat-s

  • Thursday-ADV

  • ‘On Tuesday, I intended to submit the paper on Thursday.’ (Today is already Sunday, and I haven’t submitted it yet.)

However, the verb allows no disjoint reference of the embedded subject, either overt or null; see (38).

(38)

  • *v-ap’ireb, rom {Vano / is / Δ} c’a-vid-e-s tbilis-ši

  • 1-intend that Vano(NOM) he(NOM) NOMPV-goes-OPT-3 Tbilisi-ALL

  • Intended: ‘I intend that Vano / he / pro goes to Tbilisi.’

Example (39) shows that the null embedded subject cannot pick up its reference from a c-commanding antecedent two clauses up.

(39)

  • *vano-mi ici-s, rom me v-ap’ireb, rom Δi c’a-vid-e-s tbilis-ši

  • Vano-ERG know-3 that I(NOM) 1-intend that NOMPV-goes-OPT-3 Tbilisi-ALL

  • Intended: ‘Vano knows that I intend that he goes to Tbilisi.’

Georgian thus appears to provide another counterexample to Landau’s generalization. Despite its geographical proximity to Nakh-Daghestanian, Georgian is genealogically unrelated to and structurally very different from the latter, both in general and with respect to control constructions. Nakh-Daghestanian infinitival constructions are nonfinite and usually only inflect for gender and number. Georgian subjunctive clauses are finite (in the sense that they may be used as an independent utterance), inflect for person, and are introduced by the complementizer rom. Georgian subjunctive complements with the verb ‘intend’ thus should probably count as an independent counterexample.

4 [−Agr] in Udi NC Complements

So far, we have seen two counterexamples attesting to the possibility of inflected OC complements of desiderative verbs. Recall, however, that Landau’s generalization makes one more prediction about tensed (attitude) complements: [−Agr] complements should universally be controlled. In this section, I document a counterexample to this second prediction and show that noninflected tensed complements can be noncontrolled.

Udi is a language of the Lezgic branch of the Nakh-Daghestanian family originally spoken in Azerbaijan, now also in Georgia and Russia, by around 10,000 native speakers. Udi is only distantly related to Khuduts; both go back to the same Proto-Nakh-Daghestanian language dated to around 4,000 years ago. Udi is morphologically ergative, allows free pro-drop, and has basic SOV word order. Udi has been heavily influenced by Iranian languages, Armenian, and Azerbaijani, which are dominant in the area. In many respects, it is structurally more similar to these languages than to its fellow Nakh-Daghestanian languages. Udi is one of three Nakh-Daghestanian languages lacking prefixal gender agreement of the type discussed in section 2. Instead, it developed (independently of Dargwa) obligatory person-number marking in finite clauses, always controlled by the subject, regardless of its morphological case—ergative or absolutive. Morphophonologically, the subject markers are clitics attached to the focused constituent (a detailed description of person marker placement in Udi can be found in Harris 2000, 2002). Syntactically, they function like regular agreement markers; in particular, a default third person singular marker appears in “failed agreement” contexts (see Preminger 2009).

Let us consider first the verb čureces ‘want’, which can combine with three different types of complements: (a) a finite subjunctive complement, (b) a nominalization in absolutive case, and (c) a nominalization in dative case. The finite subjunctive complement is of the NC type and thus is totally well-behaved from the point of view of Landau’s generalization. Nominalized complements, however, are more unusual. Neither of the two types of nominalization is inflected for person or number, and both are thus predicted to be OC complements.19 The actual distribution is different: the absolutive nominalization is an OC construction with a PRO subject, as expected, whereas the dative nominalization is an NC construction that allows either a lexical DP or pro as embedded subject. To illustrate null subjects in Udi, example (40) shows the PRO subject of an absolutive nominalization, while (41) shows the proi subject of a dative nominalization and the proj subject of a finite verb.

(40)

  • zui [PROi taj-sun ]=ez čur-esa

  • I(ERG) ABS go-NMLZ(ABS) =1SG want-LV:PRS

  • ‘I want to go.’

(41)

  • [proi ǯuhutː-χo-j pačːaʁ-a vaʕ-jnakː čːevkː-sun-a ] proj čur=nan=sa

  • 1SG Jew-PL-GEN king-DAT you.PL-BEN pull.out-NMLZ-DAT 2PL.ERG want=2PL=LV:PRS

  • (Context: You have a custom that I release one prisoner to you at Passover.)

  • ‘Do you want me to release the king of the Jews to you?’ (John 18:39, unpublished translation by Georgi Kechaari)

As (41) shows, the dative nominalization allows a noncontrolled reading of the null embedded subject. Furthermore, some desiderative verbs only select for nominalized complements in the dative that allow both lexical DPs and null pronouns in subject position. While the coreferential interpretation of the null subject is more easily available, disjoint readings are also possible if a discourse topic has been established.

(42) [tːija taj-sun-a ]=z qːiʕ-b-sa there go-NMLZ-DAT =1SG afraid-LV-PRS

  • ‘I am afraid to go there.’

  • ‘Ii am afraid of proj going there.’

Furthermore, nominalized complements in the dative are regularly used with propositional predicates like ‘say’, ‘think’, and ‘believe’ and with interrogative predicates like ‘ask’, ‘understand’, ‘know’, and so forth. Again, both overt DPs and pro are allowed in the subject position of such complements.

(43)

  • proi veʕ=te=z [proj mija ej-sun-a ]

  • ABS believe=NEG=1SGABS here come-NMLZ-DAT

  • ‘Ii don’t believe hej / shej / theyj came here.’

(44)

  • proi [proj mani ga-n-uχun bak-sun-a ] p-es te=tːun bak-saj

  • ERGABS which place-OBL-ABL become-NMLZ-DAT say-INFNEG=3PL can-PST

  • ‘Theyi (villagers) could not say where theyj (newcomers) come from.’ (Georgi Kechaari,

  • Orayin. Baku: Azərbaycan dövlet nəşriyyatı, 2001)

As the above examples illustrate, noncontrolled pro subjects are in principle possible in Udi nominalized complements with desiderative predicates, contrary to the predictions of Landau’s generalization.

5 Discussion

Previous work apparently showed that tensed/attitude complements in many languages obey the predictions of Landau’s generalization. The question then arises: what makes OC constructions in languages of the Caucasus so different from what we see in other languages? While a number of different parameters could be considered responsible for this deviation, I propose that the first necessary step is to reconsider the status of Landau’s generalization itself and ask another question: is it really a typological universal?

As mentioned at the outset, a small but nontrivial sample of languages initially appears to support the generalization. Moreover, some apparent counterexamples identified in the literature—for example, in Brazilian Portuguese and Finnish—have been shown to fall outside the scope of the generalization due to the nature of the null embedded subject (pro rather than PRO), and thus not to be actual counterexamples (see Landau 2013:94–95 and references therein).

Nonetheless, the OC-NC generalization has never been tested on a reasonably large sample. Languages known so far to obey it belong to a limited number of linguistic families and are almost exclusively spoken in the Europe-Balkans-Middle East linguistic area, thus representing only a small portion of the world’s linguistic diversity. While admitting that the pattern observed by Landau (2004, 2015) is probably more frequent typologically (though only a balanced sample of languages would allow us to estimate its frequency worldwide), I suggest that the counterevidence attested in the languages of the Caucasus shows that it is not universal.

I should note here that Caucasian languages are not the only case against Landau’s generalization. Although descriptive work often does not properly discuss properties of null subjects with control verbs, especially those with verbs other than ‘want’, several potential candidates can be located. In Maltese (Semitic), for example, embedded verbs obligatorily inflect for person and number of the subject. Haspelmath (2013:54) indicates that the verb ‘want’ can be used with or without the complementizer li. In the former case, the construction represents an NC structure; in the latter case, the null embedded subject is obligatorily interpreted as coreferential with the matrix subject; see (45) and (46).

(45)

  • It-tifel j-rid j-iġi d-dar kmieni.

  • DEF-boy 3SG.M.IPFV-want 3SG.M.IPFV-come DEF-house early

  • ‘The boy wants to come home early.’

(46)

  • It-tifel j-rid li j-iġi d-dar kmieni.

  • DEF-boy 3SG.M.IPFV-want that 3SG.M.IPFV-come DEF-house early

  • ‘The boy wants him to come home early.’

Crucially, some desiderative verbs that require an overt complementizer also appear to license only OC complements. The verb ipprepara ‘prepare’, for example, requires the complementizer biex but allows no disjoint reference of the null subject (an overview of complementation in Maltese can be found in Borg and Fabri 2016).

(47)

  • *L-istudent ipprepara biex n-aħsel il-kelb.

  • DEF-student prepare.PFV.3SG.M to 1SG.IPFV-wash DEF-dog

  • Intended: ‘The student prepared for me to wash the dog.’

  • (Ray Fabri, pers. comm.)

Another promising area is Meso-America, where languages of different families have been reported to select for different types of clausal complements depending on whether the embedded subject is the same as or different from the matrix subject (Palancar 2012 (Otomi, Oto-Manguean), England 1983:298–303 (Mam, Mayan), Clemens et al. 2015:435 (Ch’ol, Mayan)). The distribution is similar to that found with Maltese ‘want’: inflected complements introduced by a complementizer are noncontrolled, whereas complementizerless inflected complements are reported to require obligatory coreference of the matrix and embedded subjects.

Landau’s own earlier work also identified counterexamples to the generalization. In Hebrew, finite complement clauses with desiderative verbs show the OC pattern (see Landau 2004 for an extensive discussion). European Portuguese has also been spotted as deviating from the predictions of Landau’s generalization (Landau 2013:94–95).

As far as uninflected complements are concerned, Imbabura Quechua seems to provide a counterexample parallel to the Udi one discussed above. Imbabura Quechua has two types of clausal (nonnominalized) complements, neither of which is inflected for ϕ-features (Cole 1985:37–38). With the verb ‘want’, -ngapaj complements require obligatory coreference between the matrix subject and the null embedded subject, whereas -chun complements are noncontrolled. The latter complement type allows both overt DPs, as shown by examples in Cole 1985:37, and pro subjects, as shown by examples from the Bible, translated into Imbabura Quechua (transcription preserved, glosses mine).20

(48)

  • Muna-ni Juzi pay-paj mama-ta riki-chun.

  • want-1 José he-of mother-ACC see-SBJV

  • ‘I want José to see his mother.’

(49)

  • Maijan-da-ta cacha-ri-chun muna-nguichi?

  • which-ACC-Q go-CAUS-SBJV want-2PL

  • (Context: It was the governor’s custom at the festival to release a prisoner chosen by the crowd. So when the crowd had gathered, Pilate asked them:)

  • ‘Which one do you want me to release?’ (Matthew 27:17)

Finally, English subjunctive complements also represent a counterexample to Landau’s generalization. As tensed complements of desiderative verbs that do not show morphological agreement with the subject, they must obey the 2004 version of Landau’s generalization.21 As long as they are also attitude complements, they must also be in the scope of the 2015 version. Yet they allow no controlled null (or for that matter overt) pronoun in subject position, as shown in (50).22

(50)

  • Her father preferred that she go to school in Boston.

  • We agreed that she go back home.

  • I have decided that she go on leave due to injury.

In the face of contradicting evidence, a natural move would be to ensure some escape routes that would allow us to explain away emerging counterexamples as merely apparent. For example, in Landau’s 2004 system, the deviant behavior of Hebrew inflected complements has been accounted for by the interaction of two independent phenomena. On the one hand, the properties of C and I heads in Hebrew lead to licensing of PRO in such complements. On the other hand, the (language-specific) absence of third person pro excludes a noncontrolled reading of third person null subjects and thus only leaves a third person PRO subject. The problem, however, is that such explanations often look like ad hoc stipulations. In the Hebrew case, licensing of PRO in inflected tensed complements heavily relies on the C head being specified as [+T,+Agr]. Given that Hebrew complementizers never show overt agreement, this explanation not only is unfalsifiable but in fact lacks any empirical support at all (see McFadden 2014 for a similar point about Landau’s Agree model).23

Another way to deal with counterexamples to the generalization could be to introduce additional qualifications and thus restrict its empirical scope. As noted in section 1, the current formulations of the generalization do not include any reference to properties other than tensedness/attitude and agreement. However, both theoretical models developed by Landau (2004, 2015) only derive the generalization on the assumption that tensed/attitude complements are full CPs. Moreover, the 2015 two-tiered model even predicts that less-than-CP attitude complements should not be affected by the presence of agreement morphology and will always be cases of OC, similar to what is observed in untensed/nonattitude complements.

Restricting Landau’s generalization to CP complements would indeed cast doubt on the counterevidence presented above. In fact, gender-inflected complements in Khuduts and elsewhere in Nakh-Daghestanian lack an overt complementizer and thus may instantiate a structure less than CP (though not necessarily so; see below). The same is true for other counterexamples mentioned earlier in this section. Two considerations are relevant here.

On the one hand, some of the counterevidence would survive despite this restriction: Georgian and English subjunctive complements are introduced by overt complementizers but still violate Landau’s generalization. In addition, there is some evidence that Khuduts person-inflected infinitives may also represent a structure bigger than TP. Morphologically, they include a person agreement slot associated with the T head and the person-infinitival marker -aj. Importantly, only C-level elements may follow person markers in Khuduts: aside from the person-inflected infinitive marker, those include question markers (yes/no-question marker -u, wh-question marker -i), the logophoric complementizer hebli, and the embedded-question complementizer -l. Syntactically, both gender- and person-inflected infinitival constructions pattern with biclausal sentences rather than simple clauses with respect to binding of complex reflexives. The complex reflexive can be bound by a c-commanding clausemate argument but can never have an antecedent outside its clause. Examples (51) and (52) illustrate.

(51)

  • ʕali cin-erka ca-w pikri ik’-ul=akːu

  • Ali(ABS) self-ABL self-M.SG(ABS) thought (M.SG)say:IPF-CVB=AUX:NEG

  • ‘Alii does not care about himselfi.’

(52)

  • *madina-j b-uχːu-l=ca-b ʕali cini-j cin-erka pikri

  • Madina-DATN.SG-know-CVB=AUX-N.SG Ali(ABS) self-DAT self-ABL thought

  • ʕa-ik’w-ni

  • NEG-(M.SG)say:IPF-NMLZ

  • Intended: ‘Madinai knows that Alij does not care about heri.’

Similar to the biclausal example in (52) and in contrast to the monoclausal example in (51), infinitival constructions with desiderative verbs do not allow binding of the complex reflexive inside the embedded clause by the matrix subject; see (53).24

(53)

  • *rasul-li q’ast b-arq’-ib [cin-il cin-erka pikri

  • Rasul-OBL(ERG) decision N.SG-make:PF-AOR self-ERG self-ABL thought

  • ʕa-ik’w-ij / ʕa-ik’w-ar-aj ]

  • NEG-(M.SG)say:IPF-INFNEG-(M.SG)say:IPF-3-INF

  • Intended: ‘Rasuli decided not to care about himselfi.’

Further investigation is definitely needed to determine whether Khuduts gender- and person-inflected infinitives demonstrate any syntactic behavior indicative of a major clausal (phasal) break.

On the other hand, the restriction to CP complements would equally void much of the evidence presented in support of Landau’s generalization. Turkish and Basque nominalized complements (Landau 2006, 2013:93) would then probably be discarded. Only part of the remaining evidence represents indisputable CPs, such as Arabic, Persian, Romanian, and Albanian structures introduced by overt complementizers. If the CP status of a clausal complement is a matter of empirical evidence rather than theoretical assumption, then it has yet to be shown that other examples brought in support of the generalization, such as Hungarian and Portuguese infinitives or Greek and Bulgarian subjunctives, represent full-fledged CPs and not, say, TPs or FinPs.25 Some evidence will inevitably be disqualified. For example, Landau (2006:167–169) claims that Welsh desiderative complements are a good example of the OC-NC generalization at work, assuming that they represent CPs. However, Borsley, Tallerman, and Willis (2007:86–87) report that infinitival complements in Welsh have binding properties different from those of indicative complements in complex sentences but similar to what is found in monoclausal configurations: a pronominal subject of the infinitival clause must be disjoint in reference from a matrix NP, whereas an anaphor in the infinitival clause can be bound by an NP in the matrix clause (see also Tallerman 1998:92). While this fact can probably be accommodated within a CP approach, it indicates that the latter cannot be taken for granted.

As noted above, Landau concentrates on the familiar pattern of person agreement between the subject and the finite verb, while leaving unspecified a number of other parameters, thus making straightforward application of the generalization to other types of agreement difficult.

Agreement in Khuduts differs significantly from that canonical pattern with respect to the nature of the ϕ-features involved (separate morphological slots/syntactic probes for argument-verb agreement in gender-number and in person), the identity of the controller (ergative agreement in gender-number, hierarchical agreement in person), and the identity of the agreeing functional head (finite T0 for person, defective T0 or v0 for gender-number). As long as the embedded subject’s ϕ-values are present on a functional head in the clausal spine, current versions of Landau’s generalization block the formation of the control relation between the matrix controller and the embedded subject, and thus fail to predict the OC status of infinitival complements with less-canonical agreement properties, such as in Khuduts. The role of these parameters still needs to be clarified.

In a wider perspective, the whole idea that the PRO or pro status of the null embedded subject should correlate with agreement apparently stems from the assumption that the latter is a necessary factor involved in the licensing of pro. While this is indeed so in some languages, such as Germanic or Romance, it is not universally true, since radical pro-drop languages are known to disassociate pro from agreement. Crucially, the latter type includes not only languages without agreement, such as Chinese and Korean, but also regular subject agreement languages, like Turkic and Dravidian (see the discussion in Holmberg 2010:122–123). If agreement is not a theoretical prerequisite to pro licensing, then we can expect the two phenomena to diverge. Presence of agreement in tensed complements then does not necessarily mean that the category in question is pro. This is exactly what we see in languages of the Caucasus that allow free omission of arguments depending on appropriate discourse conditions but are insensitive to whether the arguments agree or not. As we have seen, agreeing tensed/attitude complements feature OC PRO subjects in these languages. In a similar vein, absence of agreement morphology in tensed/attitude configurations is not a sufficient condition for PRO licensing either.

The examples discussed in this article also reiterate one important point long known from previous work (Wurmbrand 2001): matrix verbs do not always easily lend themselves to being treated in terms of classes and often need to be handled on a verb-by-verb basis. As we have seen, Khuduts person-inflected complements with desiderative verbs, though uniformly licensing PRO null subjects, vary with respect to the availability of disjoint overt subjects. Similarly, Georgian subjunctive complements vary with regard to the nature of their null subject: some license OC PRO, whereas others license overt DPs or pro. Udi nominalized complements demonstrate that typical nonfinite (nominalized) complements within the same language may also diverge with respect to the nature of their null subjects (cf. Sundaresan and McFadden’s (2009) generalization restricting but not banning pro from nonfinite complements).

Overall, the theoretical significance of the OC-NC generalization seems to be overstated in Landau’s work. Although apparently having initial empirical support, the current version of the generalization cannot stand up against additional empirical evidence. Before incorporating the derivation of the generalization into the universal syntactic theory of OC, we still need to further investigate the interaction between OC and morphological agreement. Inspecting the behavior of matrix verbs in languages with different agreement properties, we will hopefully arrive at a more fine-grained typology of such interaction and then see whether anything like the OC-NC generalization arises in a nontrivial set of clausal complements. Such a generalization may well arise. However, much work is needed even in well-studied languages like English, let alone less well-known languages such as Nakh-Daghestanian and Kartvelian.

Notes

1 As Landau (2015:7) puts it:

The complements may differ in category (IP vs. CP vs. DP), mood (indicative vs. subjunctive), or morphological tense (subjunctive vs. infinitive), but each one falls on a definite side of the [±Agr] divide. The other grammatical features do not matter. The fundamental claim of the Agree model [retained in a modified form in the two-tiered theory of control—DG] is that all we need to know about the control behavior of a complement (whether it displays OC or NC) are two pieces of information: the value of its [T] feature and the value of its [Agr] feature.

2 The abbreviations follow the Leipzig glossing rules (https://www.eva.mpg.de/lingua/resources/glossing-rules.php); I have modified others’ glosses to match the rules where possible, and have added the following items not in the Leipzig list: ADD – additive clitic, AOR – aorist, IPF – imperfective stem, LV – light verb, OPT – optative, PF – perfective stem, PLUPRF – pluperfect, PV – preverb.

3 Following Landau (2004:839; 2015:6, 88), I consider the feature [Agr] to be purely morphological: [+Agr] is present when there is overt agreement morphology; [−Agr] (sometimes also called “abstract Agr”) is present when no overt morphological realization of ϕ-features is observed on the verb. Importantly, [+Agr] does not mean that every single lexical root and/or cell in a paradigm must have overt exponence. It is enough to have at least some ϕ-feature distinctions in morphology. In particular, this means that all Khuduts verbs lexicalize a syntactic structure with a gender probe, regardless of whether or not a specific verb hosts the gender agreement slot. This point is also relevant in the discussion of nonagreeing desiderative complements in sections 4 and 5.

4 The same pattern of gender agreement with plural personal pronouns is found in most other Dargwa languages (e.g., Van den Berg 2001:13 on Aqusha) as well as some other Nakh-Daghestanian languages: Archi from the Lezgic branch (Chumakina, Kibort, and Corbett 2007), and Chechen and Ingush from the Nakh branch (Nichols 2011). A typologically oriented discussion of this agreement pattern can be found in Baerman and Corbett 2013. In particular, they conclude that person agreement parasitic on gender agreement cannot be analyzed as a kind of default agreement and needs a reference to a person feature.

5 Khuduts possesses two series of person markers: one used in the aorist, present, and past progressive, and the other in the generic present, past habitual, and the person-inflected infinitive. Phonologically, person markers in the first series are clitics, whereas those in the second series are suffixes. I know of no difference between the two series in their morphosyntactic behavior.

6 It is impossible to establish which of the two arguments controls agreement in this case, since third person inflection does not show distinctions in other features, such as number or gender.

7 First and second person pronouns do not morphologically distinguish between ergative and absolutive forms in Khuduts. Interlinear glosses in (8) and other examples show the case of such pronouns as appropriate for the syntactic position of the pronoun.

8 An unrelated quirk of the Dargwa system is that the past tense marker -di looks the same as the second person singular marker -di. Whether they are diachronically related is not clear. Synchronically, they are usually considered to be simply homonymous.

9 Note that this fact alone seriously undermines claims about the universality of the OC-NC generalization.

10 Transitive infinitival complements of desiderative verbs display identical behavior in all aspects examined below, except for the controller of gender agreement: as explained earlier, transitive embedded verbs agree in gender with their absolutive direct object rather than with the ergative subject.

11 The verb q’ast barq’ij ‘decide’ is a transitive verb with an ergative subject. Morphologically, this is a complex verb consisting of the transitive light verb barq’ij ‘do, make’ and the noninflecting component q’ast ‘decision’. Gender agreement on the light verb is invariably neuter singular; person agreement is with the subject.

12 Transitive embedded predicates also license the ergative reflexive pronoun in the subject position. In constructions with the verb q’ast barq’ij ‘decide’, however, the licenser of the ergative reflexive in the embedded clause is not immediately seen on the surface, since both the matrix and embedded verbs are transitive. When a desiderative verb belongs to a different valency class (intransitive or dative subject), the embedded transitive verb is the only possible licenser of the ergative case on the reflexive pronoun.

  • madina-j b-ikː-ul=ca-b [cini-l=gina q’waʕl b-ircː-ij / b-ircː-ar-aj ]

    Madina-DATN.SG-want:IPF-CVB=AUX-N.SG self-ERG=only cow(ABS) N.SG-milk:PF-INFN.SG-milk:PF-3-INF

    ‘Madina wants to milk the cow herself.’

13 This is true without exceptions for singular antecedents/agreement. Plural agreement on the embedded intransitive infinitive is possible in the presence of a singular matrix subject in partial control environments. I leave the discussion of partial control in Khuduts and other Dargwa languages for another occasion.

14 Recall that tensed is understood here as availability of temporal mismatches between the matrix and embedded clauses. Whether or not this property should be called tensedness is an open issue (see Wurmbrand 2014 for a different approach to tense in infinitives). What is important here is that it is the property deemed relevant for Landau’s generalization.

15 Following Sundaresan and McFadden (2009) and Sundaresan (2014), I assume that availability of the disjoint overt subject in the complement clause is orthogonal to the question about the category of null subjects. The latter must be established on its own using standard diagnostics to distinguish between PRO and pro.

16 Data from the neighboring language Qunqi Dargwa may also be relevant in this respect. In this language, only a person-inflected infinitive can be used in the complement of a desiderative verb. In contrast to Khuduts, however, all desiderative verbs (I have inspected seven verbs, cognate to the Khuduts ones discussed here) uniformly prohibit overt DP subjects in embedded clauses; that is, they follow the pattern observed above for Khuduts gender-inflected complements rather than the pattern observed for Khuduts person-inflected complements.

17 The reviewer suggests that the most straightforward way to establish whether the embedded subject is PRO or pro is to exclude the matrix subject from the set of potential antecedents by setting up a Condition B situation in examples like ‘Rasul knows that Mother wants ____ to go with her’. If the sentence can be read as ‘Rasuli thinks that his motherj wants himi to go with herj’, then the null subject behaves like pro, rather than PRO. However, the proposed diagnostic is not applicable, since third person pronominals in Khuduts have antisubject orientation and can never be bound by a subject, be it local or nonlocal. ‘Mother’ therefore could not serve as an antecedent of the overt pronoun even if the null embedded subject were pro with reference to a more distant antecedent.

18 As another anonymous reviewer points out, (34) and (35) could also result from an intervention effect such that the closest potential controller must always be chosen even though there are other legitimate controllers. To the best of my knowledge, however, the two options cannot be teased apart in Khuduts. In particular, no impersonal (passive) constructions can be formed from desiderative verbs in the language.

19 The conclusion that Udi nominalized clauses must be qualified as [−Agr] is confirmed by two facts. First, no overt morphological agreement can ever be seen between the subject and the head of the nominalized clause. Second, the matrix clause and the nominalized embedded clause constitute a single domain with respect to person agreement, as shown by the ability of matrix person clitics to be attached to the constituents of the nominalized clause. No agreeing clause in Udi, indicative or subjunctive, can host a person clitic associated with the subject of another agreeing clause (Harris 2002).

20https://www.bible.com/bible/264/mat.27.17 (last accessed 25 August 2017).

21 As mentioned in footnote 3, [−Agr] means that overt morphological agreement is never present, while [+Agr] indicates that agreement is seen at least in some cells of the ϕ-feature paradigm of at least some lexical item(s). The English finite past tense is thus minimally different from the English subjunctive: the former shows agreement on the single verb be, enough to be qualified as [+Agr], whereas the latter shows agreement not even on the verb be and thus is [−Agr].

22 Unlike subjunctive complements in some other languages (Landau 2004:855–860), English subjunctive constructions are noncontrolled, without an additional disjoint reference requirement: overt pronouns in the subjunctive complement can still be accidentally coreferential with the matrix subject, as in Sheiasked that sheibe allowed to stay here.

23 It is also not clear whether/how this analysis could be recast in terms of the two-tiered theory of control. Landau (2015) does not discuss the Hebrew agreeing complements.

24 That the complex reflexive in (53) is (intended to be) bound by the matrix subject rather than the null embedded subject is seen from the case marking on the reflexive. Each component of the complex (double) reflexive in Khuduts bears its own case marking: one is in the case of the antecedent, the other is in the case of the reflexivized argument (for more on complex reflexives in Nakh-Daghestanian, see Ganenkov and Bogomolova in preparation). The ergative marking of the reflexive in (53) appropriate for the subject of the transitive matrix verb q’ast barq’ij ‘decide’ is ungrammatical. The absolutive case of the reflexive as required by the intransitive embedded verb pikri bik’wij ‘care’ would be fine here; compare the case marking of the complex reflexive in (51).

25 Note that even structures with overt complementizer-like elements may be analyzed as less than CP complements; see the discussion of the status of the complementizer ke in Persian in Ghomeshi 2001, Darzi 2008, and Ilkhanipour 2014. In addition, as McFadden (2014) also suggests, a more detailed classification of CP structures may be needed, which will also affect the scope and limits of typological generalizations like the one discussed in this article.

Acknowledgments

This article was first submitted as a squib, and I would like to thank all the reviewers and editors for their feedback on earlier versions. I am also grateful for comments from Maria Polinsky, Sandhya Sundaresan, and the audience at Control Workshop 2016 (Humboldt Universität zu Berlin, October 2016). Thanks also to my language consultants Hasan Rasulov, Huriyat Rasulova, Marziyat Rasulova, Bahammad Rasulov, and Salih Ahmedov for their patience and invaluable input on Khuduts Dargwa, and to Vladislav Dabakov, who provided judgments on examples from his native Nij Udi reported in section 4. Alexander Rostovtsev-Popiel was conveniently on his fieldtrip to Georgia when this manuscript was being prepared, and kindly agreed to check the Georgian examples reported in section 3 with native speakers. Financial support from the Basic Research Program of the National Research University Higher School of Economics (Moscow) is gratefully acknowledged.

References

Baerman,
Matthew
, and
Greville
Corbett
.
2013
. Person by other means. In
Languages across boundaries
, ed. by
Dik
Bakker
and
Martin
Haspelmath
,
1
14
.
Berlin
:
Mouton de Gruyter
.
Bond,
Oliver
,
Greville G.
Corbett
,
Marina
Chumakina
, and
Dunstan
Brown
, eds.
2016
.
Archi: Complexities of agreement in cross-theoretical perspective
.
Oxford
:
Oxford University Press
.
Borg,
Albert
, and
Ray
Fabri
.
2016
. Semantic functions of complementizers in Maltese. In
Complementizer semantics in European languages
, ed. by
Kasper
Boye
and
Petar
Kehayov
,
413
448
.
Berlin
:
Mouton de Gruyter
.
Borsley,
Robert D.
,
Maggie
Tallerman
, and
David W. E.
Willis
.
2007
.
The syntax of Welsh
.
Cambridge
:
Cambridge University Press
.
Chumakina,
Marina
,
Anna
Kibort
, and
Greville
Corbett
.
2007
.
Determining a language’s feature inventory: Person in Archi. In “Endangered languages,” ed. by Peter K. Austin and Andrew Simpson, special issue
,
Linguistische Berichte
14
:
143
172
.
Clemens,
Lauren Eby
,
Jessica
Coon
,
Pedro Mateo
Pedro
,
Adam Milton
Morgan
,
Maria
Polinsky
,
Gabrielle
Tandet
, and
Matthew
Wagers
.
2015
.
Ergativity and the complexity of extraction: A view from Mayan
.
Natural Language and Linguistic Theory
33
:
417
467
.
Cole,
Peter
.
1985
.
Imbabura Quechua
.
Amsterdam
:
North-Holland
.
Darzi,
Ali
.
2008
.
On the vP analysis of Persian finite control constructions
.
Linguistic Inquiry
39
:
103
116
.
England,
Nora
.
1983
.
A grammar of Mam, a Mayan language
.
Austin
:
University of Texas Press
.
Gagliardi,
Annie
,
Michael
Goncalves
,
Maria
Polinsky
, and
Nina
Radkevich
.
2014
.
The biabsolutive construction in Lak and Tsez
.
Lingua
150
:
137
170
.
Ganenkov,
Dmitry
, and
Natalia
Bogomolova
.
In preparation
. Binding and indexicality in the Caucasus. To be published in
The Oxford handbook of the languages of the Caucasus
, ed. by
Maria
Polinsky
.
Oxford
:
Oxford University Press
.
Ghomeshi,
Jila
.
2001
.
Control and thematic agreement
.
Canadian Journal of Linguistics
46
:
9
40
.
Harris,
Alice C
.
1981
.
Georgian syntax: A study in Relational Grammar
.
Cambridge
:
Cambridge University Press
.
Harris,
Alice C
.
2000
.
Where in the word is the Udi clitic?
Language
76
:
593
616
.
Harris,
Alice C
.
2002
.
Endoclitics and the origins of Udi morphosyntax
.
Oxford
:
Oxford University Press
.
Haspelmath,
Martin
.
2013
.
On the cross-linguistic distribution of same-subject and different-subject ‘want’ complements: Economic vs. iconic motivation
.
SKY Journal of Linguistics
26
:
41
69
.
Holmberg,
Anders
.
2010
. Null subject parameters. In
Parametric variation: Null subjects in Minimalist theory
, ed. by
Theresa
Biberauer
,
Anders
Holmberg
,
Ian
Roberts
, and
Michelle
Sheehan
,
88
124
.
Cambridge
:
Cambridge University Press
.
Ilkhanipour,
Negin
.
2014
.
On the CP analysis of Persian finite control constructions
.
Linguistic Inquiry
45
:
323
331
.
Landau,
Idan
.
2000
.
Elements of control: Structure and meaning in infinitival constructions
.
Dordrecht
:
Kluwer
.
Landau,
Idan
.
2004
.
The scale of finiteness and the calculus of control
.
Natural Language and Linguistic Theory
22
:
811
877
.
Landau,
Idan
.
2006
.
Severing the distribution of PRO from case
.
Syntax
9
:
153
170
.
Landau,
Idan
.
2013
.
Control in generative grammar: A research companion
.
Cambridge
:
Cambridge University Press
.
Landau,
Idan
.
2015
.
A two-tiered theory of control
.
Cambridge, MA
:
MIT Press
.
McFadden,
Thomas
.
2014
.
On subject reference and the cartography of clause types: A commentary on the paper by Biswas
.
Natural Language and Linguistic Theory
32
:
115
135
.
Nichols,
Johanna
.
2011
.
Ingush grammar
.
Berkeley
:
University of California Press
.
Palancar,
Enrique
.
2012
. Clausal juxtaposition and subordination: New light from Otomi. In
Clause linkage in cross-linguistic perspective: Data-driven approaches to cross-clausal syntax
, ed. by
Volker
Gast
and
Holger
Diessel
,
37
76
.
Berlin
:
Mouton de Gruyter
.
Pearson,
Hazel
.
2015
.
Attitude verbs
. http://semanticsarchive.net/Archive/mIwMTc1Z/Attitude%20verbs.pdf (last accessed 19 January 2018).
Polinsky,
Maria
.
2015
.
Tsez syntax: A description
.
Ms., Harvard University, Cambridge, MA
.
Polinsky,
Maria
.
2016
. Agreement in Archi from a Minimalist perspective. In
Archi: Complexities of agreement in cross-theoretical perspective
, ed. by
Oliver
Bond
,
Greville G.
Corbett
,
Marina
Chumakina
, and
Dunstan
Brown
,
184
232
.
Oxford
:
Oxford University Press
.
Preminger,
Omer
.
2009
.
Breaking agreements: Distinguishing agreement and clitic doubling by their failures
.
Linguistic Inquiry
40
:
619
666
.
Sumbatova,
Nina R.
, and
Yuri A.
Lander
.
2014
.
Darginskij govor selenija Tanty: Grammatičeskij očerk, voprosy sintaksisa
.
Moscow
:
Jazyki slav’anskoj kul’tury
.
Sundaresan,
Sandhya
.
2014
.
Making sense of silence: Finiteness and the (OC) PRO vs. pro distinction. Commentary on Kissock (2013)
.
Natural Language and Linguistic Theory
32
:
59
85
.
Sundaresan,
Sandhya
, and
Thomas
McFadden
.
2009
.
Subject distribution in Tamil and other languages: Selection vs. case
.
Journal of South Asian Linguistics
2
:
5
34
.
Tallerman,
Maggie
.
1998
.
The uniform case-licensing of subjects in Welsh
.
The Linguistic Review
15
:
69
133
.
Vamling,
Karina
.
1989
.
Complementation in Georgian
.
Lund
:
Lund University Press
.
Van den Berg,
Helma
.
2001
.
Dargi folktales: Oral stories from the Caucasus. With an introduction to Dargi grammar
.
Leiden
:
Universiteit Leiden, Research School of Asian, African, and Amerindian Studies
.
Van den Berg,
Helma
.
2005
.
The East Caucasian language family
.
Lingua
115
:
147
190
.
Wurmbrand,
Susanne
.
2001
.
Infinitives: Restructuring and clause structure
.
Berlin
:
Mouton de Gruyter
.
Wurmbrand,
Susi
.
2014
.
Tense and aspect in English infinitives
.
Linguistic Inquiry
45
:
403
447
.