Subjunctives in Romanian Languages: Micro-Parametric Variation in Complement CPs and the Periphrastic Future

Abstract

This paper aims to (i) establish the micro-parametric variation in the development of the subjunctive CP in Romanian languages (Daco-Romanian/DR; Aromanian/AR; Megleno-Romanian/MR; Istro-Romanian/IR) and (ii) account for derivations in which the subjunctive is integrated into the formation of the periphrastic future in these languages. Briefly, the analysis points out that the subjunctive CP in Romanian languages can display a split Fin (unlike in other Balkan languages) and that the remerging of the split Fin finds itself at different stages: complete in DR, but incomplete at different degrees in AR, MR, and IR. The compatibility of the subjunctive morphology with the derivation of the periphrastic future follows from the semantic bleaching and grammaticization of the volitional ‘will’ and ‘have’ verbs, together with the Balkan Sprachbund subjunctive mood marking, which combine in a monoclausal construction via a serial verb derivation to compositionally check a Fin marked [+finite, modal].

1. Introduction

All Balkan languages (including Balkan Romance) have a preverbal morpheme to mark subjunctive mood (Mišeska-Tomić 2006). Variation in the Balkan Sprachbund occurs with subjunctive complements insofar as only Romanian languages may also display a complementizer of the ‘that’ type in addition to the subjunctive marker (Hill and Mišeska-Tomić 2009). An illustration of this contrast is provided in (1): ‘that’ is ungrammatical with Bulgarian subjunctive complements, but it is optional in the same environment in Aromanian (AR).

(1)	a.	Zapovjadax	[(*če)	(Marija)	da	dojde.]		Bulgarian
		ordered.1sg	that	Maria	sbjv	comes.perf
		‘I gave an order for Maria to come.’
	b.	Deade	naredba	[(tă)	(Maria)	s-	yină.	AR
		gave.3sg	order	that	Maria	sbjv	come.3sg.sbjv
		‘I gave an order for Maria to come.’ (adapted from Hill and Mišeska-Tomić 2008, 2009)

The ‘that’ type complementizer with subjunctive clauses is reminiscent of Romance que (e.g., in French or Spanish), whereas the preverbal subjunctive marker (i.e., AR s’; Bulgarian da; Greek na) is a Balkan Sprachbund property, as is the subjunctives replacement of infinitives (Mišeska-Tomić 2006).

The first objective of this paper is to account for the complementizer cooccurrence in (1b) and establish the micro-parametric variation in the development of the subjunctive CP in Romanian languages (Daco-Romanian/DR; Aromanian/AR; Megleno-Romanian/MR; Istro-Romanian/IR). The proposal is that, diachronically, the subjunctive marker in Old Romanian arises from a series of reanalyses by which the clause typing (Force) complementizer să ceases to be an operator in selected clauses and becomes specified for modality/mood and non-finiteness; i.e., it merges as a Fin [-finite, modal] complementizer in subjunctives. At a certain stage of reanalysis, props like ‘that’ were needed to clarify the status of să in Fin by splitting Fin over the two features (i.e., non-finite complementizer ‘that’ for Fin [-finite] and să for Fin [modal]). When sufficient learners adopt the analysis of să as both non-finite and modal, Fin remerges, and the props are reduced or completely disappear. This process is implemented at a different pace in each Romanian language.

The second objective of this paper is to account for derivations in which the subjunctive is integrated into the formation of the periphrastic future in Romanian languages. Old Romanian and DR display two options for expressing the future tense: (i) the analytic future available to Romance languages more generally, consisting of an auxiliary followed by the infinitive form of the verb (i.e., voi pleca ‘will.1sg go.inf’; and (ii) the periphrastic future, seen in Balkan languages, which consists of an auxiliary of the ‘will’ type followed by subjunctive complementation, as in (2). This is a Balkan Sprachbund property (Mišeska-Tomić 2006), with minimal variation arising from independent language-specific properties. For example, consider the Future Perfect data in (2) where, in Macedonian, perfectivity can be encoded either on the auxiliary, (2a), or on the verb, (2b), whereas in DR, it may occur only on the ‘will’ auxiliary, (2c), and not as the free morpheme ‘be’ (2d), typically associated with perfectivity on the verb in other DR contexts.

(2)	a.	Nemaše		da	dojdam/dojdeš/dojde.				Macedonian
		not.will.Impers.Past		sbjv	come.1/2/3Sg.Perf.Pres
		‘I will not have come.’
	b.	Nema		da	dojdev/dojdeše.
		not.will.Impers		sbjv	come.1/2/3Sg.Perf.Past
		‘I will not have come.’ (from Hill and Mišeska-Tomić 2009)
	c.	O/	are/avea		să	devină		scriitor.	DR
		will	have.fut/have.fut.perf		sbjv	become.3sbjv		writer
		‘He will/will have become a writer.’
	d.	*O/are		să		fi	devenit	scriitor
		will/ have.fut		sbjv		be	become	writer
		Intended: ‘He will have become a writer.’

The observation is that the contrast between (2a, 2b) and (2c, 2d) arises from distinct underlying structures, which are biclausal in (2a, 2b) but monoclausal in (2c, 2d). In Macedonian, there are two TP fields available for encoding perfectivity, whereas in DR, there is only one TP field. We show that tests of scopal properties and constituent movement confirm the monoclausal structure of the periphrastic future.

The data in this paper come from corpora of AR and IR listed in the References section, which consists of texts transcribing oral narratives by native speakers. For MR, we used data arising from the interviews of native speakers recorded by Olga Mišeska-Tomić between 2008 and 2009. For DR, the native judgments of the authors are the main source. The Old Romanian discussion is based on the corpus of texts and tests in Hill and Alboiu (2016).

This paper is organized as follows: Section 2 provides a list of theoretical tools (Section 2.1) and a summary of the properties of subjunctive clauses in Old Romanian, which are relevant to the foregoing discussion (Section 2.2). Based on this background, Section 3 offers an analysis of subjunctive complements in Romanian languages. Section 4 introduces the periphrastic future constructions in Romanian languages, which receive a formal analysis in Section 5. Section 6 concludes the paper.

2. Background

2.1. Theoretical Framework

In this paper, we work within a feature-checking version of the Minimalist Program (Chomsky 1995, 2000, 2001, 2004, 2007, 2008) across the clausal spine. This framework assumes a bottom–up approach to structure and, in principle, allows for as much structure as economically necessary. The standard minimalist CP > TP > vP nucleus is insufficient to map all relevant features present with Romanian subjunctives, so we employ the cartographic mapping of Rizzi’s (1997, 2004) system, which splits the CP/left-periphery of the clause into several heads. In this system, a low/Fin and a high/Force C head sandwich discourse features, such as Topic and Focus. This clause hierarchy is given in (3), with (3) also showing that we expect to group the Romanian languages with Romance more generally, where the highest verbal element moves to some Infl/T head (Verb Raising),1 and Neg signals the border between C and T (Zanuttini 1997).2

(3)	[ForceP [TopP [FocP [FinP [NegP [TP V…]]]]]]

Nonetheless, we depart slightly from Rizzi’s feature mapping system. First, Rizzi (1997) has a lower TopP (familiar) between FocP and FinP, which is excluded from (3) because there is no evidence for it in Romanian languages, so (3) is rather similar to the proposal in Benincà (2001, 2006). Second, we follow D’Alessandro and Ledgeway (2010), who split modality and mood between Fin and T, respectively, arguing that Fin encodes only semantic modality, while grammatical [mood], as an inflectional property of the verb, is {XE “features:[mood]”} associated with T.

Finally, irrealis modality, as defined in Bybee and Fleischman (1995), is mapped to Force (not to Fin) as a clause typing operator, such as conditional (e.g., Haegeman 2010) or directive, in imperatives (e.g., Han 1998). These clause-typing operators further map a modal feature on the selected Fin, which ensures a compatible verb inflection by selecting T with an agreeing [mood] feature. Alternatively, in selected contexts, the modality feature of Fin does not reflect the properties of an unselected clause typing operator; rather, it reflects the semantics of the matrix verb requiring an irrealis modality connoting wishes, beliefs, and attitudes instead of conditions, options, and commands (Saeed 2003), with the same effects on T selection. In particular, Fin has an underspecified [modal] feature that can be valued as realis or irrealis. For the irrealis value, the exact reading arises compositionally from the sentence. Accordingly, Corbeanu and Hill (this volume) consider that the configuration for this mapping has the hierarchy in (4).

(4)	V/OP > Fin[modal] > T[mood]

Lastly, we follow previous work in Alboiu and Hill (2021) and Hill and Alboiu (2016) and take the notion of finiteness to correlate with temporal deixis/[+finite] (a property of the Force/phasal C head), or anaphoric tense/[-finite] (a property of Fin), rather than with the presence of phi-features on the verb form. Consequently, we take phi-features to be a T property (rather than a Fin property) since they depend on types of mood instantiation, which is also a T property: [+Agr] in indicatives and subjunctives; [-Agr] in supines, infinitives, and gerunds in Old Romanian. Nonetheless, given the matching requirement imposed by the Fin-T selectional relationship, the agreement specification of T reflects the syncretic mapping of finiteness and modality in Fin.3

Unlike in other Romance languages, the presence or absence of inflectional morphology on the verb stem is not a reliable criterion for independent lexical subjects (i.e., obviation) and Case in Romanian. Alboiu (2006, 2007, 2009) shows that the latter properties correlate with the presence of a phasal C head and trigger non-obligatory control. In obligatory control structures, a full-fledged CP/ForceP domain is never projected. Specifically, complements to obligatory control predicates are untensed (i.e., bear anaphoric tense), on par with other Balkan languages (see Dobrovie-Sorin 1994; Farkas 1984, for Romanian; Krapova 2001, for Bulgarian; Landau 2004, for Balkan languages more generally), so are FinP domains (Alboiu 2007, inter alia).4 Hill and Alboiu (2016) show that this also holds true of Old Romanian, while Corbeanu and Hill (this volume) show this for IR. Crucially, in obligatory control contexts (where obviation is excluded), Fin has an anaphoric tense, so it is non-finite, regardless of the presence or absence of phi-features/agreement in T (i.e., subjunctive inflection).

2.2. Old Romanian Să

The micro-variation in the syntax of subjunctive complements has its source in the distribution and use of să and of the subjunctive verb in Old Romanian. Disparities arose because Romanian languages split from Old Romanian at some point after the Balkan Sprachbund template for subjunctive clauses had emerged in the language, but at a time when their relation to să and to the existing types of clausal complements and adjuncts was still fluid. The point of population and, therefore, language splitting would be no later than the 10th century for AR (Vrabie 2000), around the 10th–11th centuries for MR (Capidan 1925), and the latest by the 13th century for IR (Lozovanu 2008). DR developed within the same geo-ethnical area as Old Romanian.

The Balkan subjunctive template had been adopted since Proto-Romanian when the speakers of the Romance language north of the Danube coexisted for centuries with the Bulgars (Turkic semi-nomadic warrior tribes), who were spread above and below the Danube (around the 5th–8th centuries), a situation conducive to language contact and bilingualism. Relevant to this paper is that during this balkanization of the language, the Latin morphemes si or sic became a homophonous să whose syntactic status was ambiguous.5

The earliest Old Romanian texts preserved date from the mid-16th century, which is too late to capture the emergence and early development of subjunctive clauses. However, even in the 16th century, the syntactic status of să was unstable. It occurs in root clauses that qualify as surrogates for imperatives, in prepositional adjunct clauses, in conditional protasis, and in selected complement clauses. Hill and Alboiu (2016) capture this distribution under the analysis of să as a Force complementizer with operator features, such as needed for imperatives and non-selected subordinate clauses, including conditionals. The use of să in the CP field of selected complement clauses must, then, be the result of a reanalysis of să from a Force complementizer to a Fin one by gradually stripping it of the operator feature and of the clause typing feature. The detailed gradual process of să reanalysis is summarized in Corbeanu and Hill (this volume).

The important observation is that AR, IR, and MR split from Old Romanian long before the 16th century, that is, at a time when să and the subjunctive paradigm were even more unstable than attested in Old Romanian. The development of these constructions occurred not only in isolation from Old Romanian but also under language contact and bilingualism with Balkan Slavic, Albanian, or Greek. Predictably, variation is expected in the syntax of subjunctive complements. Even more variation is expected in the emergence and development of the periphrastic future, as shown in (2c) since there is no evidence for such a construction in Proto-Romanian. The construction is barely attested in the 16th-century texts (SOR 2016); so its presence or productivity in the previous centuries is doubtful. At the time of the split, the productive mechanism for the expression of future tense in Old Romanian was the analytic future (i.e., auxiliary + infinitive verb; voi pleca ‘will.1sg go.inf’), which follows the Romance template.

3. Subjunctives across Romanian Languages

3.1. Morphology

All Romanian languages have preverbal subjunctive markers based on Old Romanian să, as shown in

Table 1. Complementizers.

	Indicative ‘That’	Subjunctive ‘That’	Subjunctive Mark
Aromanian	ca	tă	s-
Daco-Romanian	că	ca	să
Istro-Romanian	che	che	se, neca
Megleno-Romanian	ca	ca	si

. Only IR displays an alternative marker that is replacing să (i.e., neca; see Corbeanu and Hill, this volume). The inflection of the verb related to să varies very slightly for the present subjunctive (see Mišeska-Tomić 2006): In AR, DR, and MR, the forms are identical to the indicative for the first and second person but display specific endings in the third person. This contrast is similar to Old Romanian paradigms. In IR, the subjunctive is completely parasitic on the indicative paradigm (except for the inflection of ‘be’).

Remarkably, să is specialized for subjunctive only in DR. In AR, IR, and MR, să is also used as a conditional marker, with verbs inflected for conditional mood, as was the case in the 16th-century Old Romanian texts. This use is very productive, to the extent that s- is obligatory with the conditional inflection in AR (e.g., AR s-cântárim vs. *cântárim ‘I would sing’; Vrabie 2000).6

In subjunctive complements, să may be preceded by a ‘that’ complementizer, whose spellout may or may not be identical to the ‘that’ complementizer of indicative complements. Table 1 introduces the ‘that’ complementizers comparatively between indicatives and subjunctives, together with the să variant specific to each language.

The next section will assess the syntactic behavior of these complementizers.

3.2. Syntax

In general, the syntax of subjunctive complements displays sensitivity to referential control over the embedded subject. In Romance languages, this involves subject obviation with subjunctive complements but obligatory control with infinitive complements (see references in Landau 2013). In the Balkan Sprachbund, including Romanian (Farkas 1984, a.o.), both types of referential assignment ares compatible with subjunctive complements. However, the left periphery cartography of the clause is not identical for the two options: subject obviation entails a more complex CP than obligatory control, as outlined in Section 2.1. To recap, for Old Romanian, Hill and Alboiu (2016) have argued that the subject obviation option involves a ForceP structure, whereas the obligatory control option involves a FinP, that is, a truncated option with an unprojected ForceP level. This same correlation is argued for in Alboiu (2007) for DR.

Keeping this structural contrast in mind, we can proceed to tests of word order that clarify the status of each item in Table 1 within the subjunctive CP. Let us focus on the location of negation in relation to the subjunctive marker. Following (3), NegP is linearized between FinP and TP; hence, in (5), the subjunctive să/se/s- is in Fin because it precedes negation.

(5)	a.	Tsă	dzăsh		s-nu		yuvuseshci!	AR
		cl.2Sg.Dat	say.1Sg.Aor		sbjv-not		read.2Sg
		‘I told you to not read.’ (from Mišeska-Tomić 2006, p. 565)
	b.	Ţi-am		spus să	nu		citeşti.	DR
		cl.2Sg.Dat-have.1		said sbjv	not		read.2Sg
		‘I told you to not read.’
	c.	ţezaru-l’		ziţe che	se	nu-l’	hiie	rusire spure	IR
		prince.the-cl.2Sg.Dat		says that	sbjv	not- cl.2Sg.Dat be. sbjv.3sg shame say.inf
		‘The prince tells her to not be shy to tell him’ (SP 126,18)
	d.	ăţ	ziş		s-nu		conţ!	MR
		cl.2Sg.Dat	say.1Sg.Aor		sbjv-not		sing.2Sg
		‘I told you to not sing.’ (Mišeska-Tomić 2006, p. 547)

The word order in (5) locates the subjunctive să at the border between C and T, an insight first captured by Dobrovie-Sorin (1994). Perhaps unsurprisingly, its mapping has resulted in a variety of labels from a mood/I head (Alboiu 2002; Cornilescu 2000; Rivero 1994, a.o.) to a low C/Fin head (Alboiu 2007; Hill and Alboiu 2016) and the lowest Fin/Fin2 under a split Fin analysis (Alboiu and Hill 2021; Hill and Alboiu 2016). In sum, our current mapping of the subjunctive free morpheme in Fin is in line with previous analyses.

Crucially, as in (6), the subjunctive marker in Fin cooccurs with ‘that’ in AR, IR, and MR but not in standard DR.

(6)	a.	Pots	(tă) s-ti		duts.	AR
		can.2sg	that sbjv-refl.2sg		go.2sg
		‘You can go.’
	b.	Poţi	(*ca) să te		duci.	DR
		can.2sg	that sbjv-refl.2sg		go.2sg
		‘You can go.’
	c.	Pόţi	(che)	se	åvzi.	IR
		can.2sg	that	sbjv	see.2sg
		‘You can see.’
	d.	Poţ	(ca) s-ti		duţ.	MR
		can.2sg	that	sbjv-refl.2sg	go.2sg
		‘You can go.’

In (6), the verb ‘can’, which is adjacent to the subjunctive sequence, requires constructions with obligatory subject control. Hence, its complement can only be a reduced CP, as FinP (versus ForceP). Availability of ‘that’ and the subjunctive marker cooccurrence in AR, IR, and MR, as in (6), shows evidence for the necessity of a split FinP into Fin1 (‘that’) and Fin2 (sbjv), as in (7a).7

If we follow the hierarchy in (3), the subjunctive marker cannot be pushed down into TP because it is higher than negation, as shown in (5). Thus, a split Fin analysis follows, as in (7a), where Fin1 checks [finite] and Fin2 checks [modal]. Specifically, să was not initially capable of checking the [finite] feature of Fin and only checked irrealis modality (i.e., [modal] in Fin2). However, the optionality of ‘that’, as indicated by the brackets in examples, means that the alternative structure, with a remerged Fin, is also available, as in (7b): the subjunctive marker in Fin is also reanalyzed (upward reanalysis; Roberts and Roussou 2003) as being able to check [finite] in addition to [modal] (Alboiu and Hill 2021; Hill and Alboiu 2016). These are intra-language variations. Standard DR has lost the option in (7a) and generalized (7b).

(7)	a.	Vputea’can’ [FinP1 ‘that’ [FinP2 sbjv[NegP[TP….
	b.	Vputea’can’ [FinP sbjv [NegP[TP….

Comparing the cooccurrence of ‘that’-sbjv in Romanian languages with Old Romanian, it appears that DR has not only generalized the remerging of subjunctive Fin but has also restricted the use of să to subjunctives. This is different from AR, IR, and MR, where să preserves an ambiguous status as either Force or Fin item, with or without subjunctive clauses (especially in IR; Corbeanu and Hill, this volume). Thus, the reanalysis of să is more advanced in DR than in other Romanian languages.

In constructions with subject obviation, as in (8), the situation remains similar in that there is a contrast between DR and the other Romanian languages.

(8)	a.	Narăncio	(Maria) ca (Maria) si	vină	ună shi ună.	MR
		order.3Sg.Aor Maria that Maria sbjv come.3Sg.sbjv one and one
		‘(S)he asked/ordered for Maria to come immediately.’
	b.	Deade	naredba (Maria) tă (Maria) s-	yină	tunoară	AR
		gave.3Sg	order	Maria that Maria sbjv- come.3Sg.sbjv immediately
			‘(S)he asked/ordered for Maria to come immediately.’
	c.	A	cerut (*Maria) ca Maria să	vină	imediat	DR
		has asked Maria that Maria sbjv come.3Sg.sbjv			immediately
		‘(S)he asked/ordered for Maria to come immediately.’

In (8), the preverbal subject Maria may surface either before or after ‘that’ in AR and MR. Preverbal subjects in the CP field (i.e., higher than să) are in Spec,TopP, as a general Balkan Sprachbund rule, where the canonical order is VSO. Accordingly, when the subject is to the left of ‘that’, Fin1 contains ‘that’, whereas Fin2 contains sbjv, and the clause typing feature of Force is checked by ‘that’ through a long-distance Agree, as in (9a). Alternatively, the subject, still in Spec,TopP, may surface between the two complementizers, which indicates the movement of ‘that’ from Fin1 to Force for feature checking, as in (9b). If ‘that’ is skipped (which is also possible), Fin is remerged, as in (9c), and să checks the features of Force through a long-distance Agree. The subject is banned from surfacing between să and the verb (i.e., Spec,TP is not available to subjects in Balkan languages; Alexiadou and Anagnostopoulou 1998).

(9)	a.	V’order’ [ForceP [TopP Maria [FinP1 ‘that’ [FinP2 să sbjv[NegP[TP….
	b.	V’order’ [ForceP ‘that’ [TopP Maria [FinP1 <‘that’> [FinP2 să sbjv[NegP[TP….
	c.	V’order’ [ForceP [TopP Maria [FinP să sbjv[NegP[TP….

In DR (8c), the subject Maria in Spec,TopP can only follow ‘that’. Consequently, ‘that’ must be assumed base-generated in Force, denoting that Fin1 and Fin2 are no longer distinct head options and that the subjunctive particle is mapped in a remerged Fin head, as in (10).

(10)	V’order’ [ForceP ‘that’ [TopP Maria [FinP să sbjv[NegP[TP….

For IR, the situation is more complex since the subjunctive morpheme se/neca is still analyzed as a Force complementizer in free alternation with split Fin or remerged Fin. For example, it can precede wh-elements in Spec,FocusP, as seen in (11).

(11)	A	lui	zapisęit	neca če-m	då
	has cl.3sg.dat		wrote	sbjv what-cl.1sg.dat give.2sg
	‘He wrote to him what you should give to me’ (TC, 143)

No other Romanian language allows for the order of complementizers in (11). The test of subject placement on the template in (8) is, thus, unreliable in IR, given the unstable featural make-up associated with se/neca in selected clauses (i.e., [+/− operator]; [+/− fin]). Thus, a lexical subject may not be allowed in the CP field for independent reasons (e.g., operators force postverbal placement of subjects). Although there are instances that attest to the possibility of split Fin with ‘that’ in IR subjunctive complements, as in (5), the variable status of se/neca is such that its cooccurrence with ‘that’ in contexts with subject obviation (i.e., with ForceP) may allow for either Force or Fin reanalysis of this item.

The data discussed in this section indicate the following direction of development for the subjunctive marker: Considering the distribution and properties of să in the 16th-century Old Romanian texts, the reanalysis it underwent up to the 20th century is subject to micro-variation, as follows:

(i)

In standard DR, să has been completely reanalyzed as a remerged Fin complementizer dedicated to subjunctive clauses. Conditional să was productive in Old Romanian before the 16th century but showed a decrease later, being replaced with de, and was eliminated from conditional CPs in DR;

(ii)

In MR, să follows the same direction of reanalysis as in DR, but at a slower pace since there is still evidence of a split Fin. Also, the conditional use of să in Old Romanian is still present in MR, but it has strong competition from aku ‘if’ (Mišeska-Tomić 2006, p. 557);

(iii)

In AR, să seems to be at the same stage of reanalysis as in MR insofar as split/remerged Fin is concerned in subjunctive complements. However, the use of să in conditionals is very productive and unopposed; in fact, să has spread as a conditional marker on verbs (Vrabie 2000);

(iv)

In IR, să is lexically replaced by neca in subjunctive clauses (a process still ongoing) but remains productive in conditionals. IR strongly preserves the specialization of să for conditionals to the extent of eliminating it from subjunctive clauses. The subjunctive replacement by neca indicates that the stage of reanalysis as a Fin complementizer (i.e., either remerged or split) is comparable to that seen in MR and AR. At the same time, neca also displays instances of direct merge in Force in selected subjunctive clauses, unlike in the other modern Romanian languages where să never undergoes first merge higher than Fin.

In sum, the diachronic development contrasts DR with IR insofar as DR să is specialized for subjunctive CPs, whereas IR să is specialized for conditionals. In AR and MR să retains its earlier fluidity, preserving both the subjunctive and the conditional statuses from Old Romanian.

4. The Periphrastic Future across Romanian Languages

The periphrastic future is a Balkan Sprachbund property; so its presence in Romanian languages is unexceptional. This section looks at the syntactic mechanism that allows a subjunctive verb to derive this construction.

4.1. Old Romanian

The periphrastic future consists of an auxiliary of the ‘will’ type followed by a subjunctive verb string. In Old Romanian, the subjunctive is systematically present after the auxiliaries are (inflected), va (inflected up to the 18th century, then uninflected; Mandić 2010), or o (uninflected), as shown in (12).8

(12)	a.	Ci	să	ştii,		logofete,	că	nu	va	să	se
		but	sbjv	know.2sg		judge.voc	that	not	will	sbjv	cl.refl.3
		aleagă		nimic		din	judeţ.			(DRH, 285)
		achieve.3sg.sbjv		nothing		from	judgment
		‘But you should know, your Honor, that nothing will come out of this judgment.’
	b.	să	fac	ispisoc		domnesc	şi	o	să	scriu
		sbjv	make.1sg	decree		royal	and	will	sbjv	write.1sg
		la	târgu	(DRH, 296)
		to	town
		‘Let me issue a royal decree and I will write to the town.’
	c.	Să	se	ştie		ocina		Neanciului			ot Albeşti,
		sbjv	cl.refl.3	know.3sg		land.the		Neanciu.gen			from Albesti
		pre	unde	are	să	aibă		a	ţineare		Neanciul
		on	where	will	sbjv	have.3sg. sbjv		to	hold		Neanciu
		cu	feciorii	lui	ocină	în Neagreasca		şi		în Jilişte (DRH, 565)
		with	sons.the	his	land	in Negreasca		and		In Jiliste
		‘Let it be known about the land of Neanciu from Albesti, where Neanciu and his sons will have to hold land in Negreasca and in Jiliste.’

Etymologically, va comes from the verb vrea, ‘want’ (SOR 2016), by grammaticalization from a lexical to a functional category. The Old Romanian texts attest to this transition as intra-language variation, as in (13).

(13)	a.	cela	ce	nu-ş			va	să-ş
		the.one	who	not-cl.refl.3.dat			will	sbjv-cl.refl.3.dat
		împarţă		avuţiia		lui		(CEv, 345)
		share.3sbjv		wealth.the		his
		‘The one who doesn’t want to share his riches.’
	b.	Domnul	dentru	ea	va	să	se	nască	zise (CEv, 554)
		Lord.the	from	her	will	sbjv	cl.refl.3.acc	be.born	said
		‘He said that the Lord will be born from her.’

In (13a), va stands for the volition lexical verb ‘want’, whereas in (13b), va encodes the futurity of the event expressed by the lexical verb nască, ‘be.born’. A volition reading cannot arise in (13b) because the verb lacks agentivity (the child to be born cannot be a volitional subject, so it must be theta-marked by the unaccusative nască, ‘be.born’). According to philological studies, va has a strong epistemic modality in its initial stages of grammaticalization (Niculescu 2011), which lessens in time. Crucially, this item preserves the irrealis modality.

The alternate o, used as in (12b), is phonologically the most reduced future auxiliary. It is considered to be another result of the grammaticalization of vrea, ‘want’ (SOR 2016), which replaced va in modern DR with a similar distribution.

The option are, as in (12c), comes from the grammaticalization of the lexical verb avea, ‘have’, in contexts in which it has a deontic modality reading (SOR 2016). The difference between the future and the deontic modality reading is still ambiguous in Old Romanian. For example, in (14a), are has a future reading, while the deontic reading arises from the subjunctive aibă that follows it. However, in (14b), are is ambiguous between denoting futurity or necessity.

(14)	a.	pre	unde	are		să	aibă		a	ţineare	Neanciul
		on	where	have.fut		sbjv	have.3sg. sbjv		to	hold	Neanciu
		cu	feciorii	lui	ocină	în Neagreasca		şi	în Jilişte (DRH,565)
		with	sons.the	his	land	in Negreasca		and	In Jiliste
		‘where Neanciu and his sons will have to hold land in Negreasca and in Jiliste.’
	b.	Că	acest	orb,	nimenea		nu	avea	să derepteaze		pre el (CEv, 496)
		for	this	blind.man nobody			not	have.perf	sbjv guide.3		dom him
		‘as for this blind man, nobody had to guide/will have guided him’.

In sum, the examples from (12) to (14) show that the periphrastic future in Old Romanian displays the following traits:

• When negation is present, as in (12a) and (14b), it precedes va, o ‘want’/ are ‘have’, and takes scope over both the future and the subjunctive verbs;
• These constructions have a future tense reading, indicating that Fin is [+finite], inflectionally encoded on the ‘will’ verb;
• Since the construction is [+finite], să cannot merge as a Fin complementizer, or it would mark Fin as [-finite]; consequently, in these constructions, să merges lower, as an inflectional head, so is reduced to a [mood] marker.

Considering the timeline for the emergence of this construction in Old Romanian, its inheritance by DR is predictable, while its occurrence in some of the other Romanian languages is likely due to other factors. In particular, IR preserved the analytical future and did not develop the periphrastic one (Sârbu and Frăţilă 1998), but the construction occurs in AR and MR. Insofar as AR and MR are concerned, the inheritance of the periphrastic future from Old Romanian is questionable, and its occurrence is due, more likely, to the subsequent contact with Balkan languages that display the same template for expressing futurity. Thus, Mišeska-Tomić (2006) points out the similarities between MR and Macedonian in attributing the future reading only to negative ari with subjunctive complements. Also, Brâncuş (2007, p. 172) shows the similarities between Albanese Tosk and the AR variety of Fărşeroţi insofar as both dialects eliminate the subjunctive marker from the periphrastic future.

4.2. The Modern Romanian Languages

Similar to other Balkan languages, Romanian languages developed the periphrastic future on the basis of two properties: (i) the reanalysis of ‘want’ and/or ‘have’ as ‘will’ verbs with a reduced form (e.g., invariable); and (ii) the embedding of să subjunctive strings under this ‘will’ verb. This section provides an overview of the data.

DR uses the analytic future in the literary register but the periphrastic future, as in (15), in colloquial registers. For the latter, o and are constructions are productive, whereas va has disappeared with the subjunctive (or it occurs in the archaic register).

(15)		Mama	o/are	să	plece	devreme.	DR
		mother	will/have.fut	sbjv	leave.3sbjv	early
		‘Mother will leave early.’

As already mentioned, in MR, only the invariant ari version is available, and its future reading is restricted to negative contexts, as in (16a, 16b). In the absence of negation, ari yields deontic obligation readings, as in (16c).

(16)	a.	Nu ari	si	vin.	MR
		not have.fut	sbjv	come.1sg
		‘I won’t come.’
	b.	Nu ari	s-la	leai.
		not have.fut	sbjv-cl.3sg.acc	take.2sg
		‘You won’t take it.’ 548.
	c.	Ari s-la lei!
		have sbjv-cl.3Sg.m.acc take.2sg
		‘You have to take it!’ (MT547).

Otherwise, the future is expressed by root subjunctives, as in (17) and (18). Arapi (2016) argues that this option has developed in MR under language contact with Buzuku (Albanian), where this prospective future is productive.

(17)	şi	ea	să	rudească	un il’ĭŭ	şa cum	aţi ţeri buricu	MR
	and	she	sbjv	produce.3.sbjv	a son	so as	her asks heart
	‘and she will give birth to a boy as her heart desires’.						(Arapi 2016)

(18)	La ficioru si-ăi	trimet	pari	moini.
	To boy.the sbjv-cl.3sg.dat	send.1sg	money tomorrow
	‘To the boy I will send money tomorrow.’ (Mišeska-Tomić 2006, p. 297).

AR uses the periphrastic future with both va and are in invariable forms. As with MR, in the absence of negation, are yields deontic obligation readings, while under negation, the reading is ambiguous between future event and injunction. Importantly, are futures allow for a ‘that’ complementizer with the subjunctive complement, as in (19). Notably, these forms do not appear in our AR corpus, which may signal that they are unproductive.

(19)	Nore	(tă)	s-lu		ljai.	AR
	not.have	that	sbjv-cl.3sg.m.acc		take.2sg
	‘You won’t take it.’ (Mišeska-Tomić 2006, p. 566).

On the other hand, va futures are productive in the corpus, but their use differs from one variety of AR to another. For example, va combines with să-subjunctives to yield a future reading in the Grămosteni variety, as in (20a), but with a bare subjunctive (deletion of să) in the Fărşeroţi variety, as in (20b). For the latter variety, va with să-subjunctives yields only deontic readings (Nevaci 2012).

(20)	a.	Dzăk	ca	va		s-neg		AR	Grămosteni
		say.1sg	that	will		sbjv-go.1sg
		‘I say that I will go.’ (MT 562)
	b.	Va	ai. //	Va	s-ai.			AR	Fărşeroţi
		will	have.2sg	will	sbjv-have.2sg
		‘you will have.’ // ‘You must have.’

In sum, the periphrastic future is a later innovation of Old Romanian, so it is expected that Romanian languages that split from the Old Romanian speaker community long before the 16th century may not have inherited this construction, while Daco-Romanian did, due to the language continuity. However, the presence of subjunctive complementation in these languages provided the means of either developing the periphrastic future as a language-internal construct or as transfer through language contact from other Balkan languages. The next section proposes a formal explanation of how this derivation could be implemented.

5. Formal Analysis of the Periphrastic Future in Modern Romanian Languages

In Section 5.1, we show that the periphrastic future is monoclausal in Romanian languages, and we engage with both verb restructuring and serial verb construction analyses (SVC). We propose that an SVC analysis is better equipped to handle the data and show sample derivations for DR in Section 5.2.

5.1. The Monoclausal Property

The first question to be addressed concerns the degree of grammaticalization of the verb form that stands for ‘will’, as this will further decide whether the construction is biclausal or monoclausal. Specifically, is the ‘will’ element a control verb on a par with ‘can’ in (5), or is it an auxiliary on a par with ‘have’ in present perfect paradigms?

The constructions with the complementizer ‘that’, as in the AR example in (18), receive a straightforward analysis as being biclausal since ‘that’ signals the presence of a split Fin. Hence, these are constructions with obligatory control, similar to the ‘can’ constructions in (5).

Conversely, two observations indicate that the other constructions are monoclausal: (i) negation systematically precedes the highest verbal head while scoping over both verbs; and (ii) the ‘will’ head behaves like a modal auxiliary rather than a matrix verb. For example, in MR (16a, 16b), the subjunctive verb cannot be interpreted positively, and it is also not possible to include the negation on the subjunctive verb (Hill and Mišeska-Tomić 2009) or to have negation present twice. The same applies to AR and DR, as shown in (21), with negation in bold.

(21)	a.	Cum nu va	s-ducî? A.34 //	*Cum va s-nu ducî? //	*Cum nu va s-nu ducî?9	AR
		how not will sbjv-go	how	will sbjv-not go	how not will sbjv-not go
		‘How come s/he will not go?’
	b.	Cum n-o să plece? //		*Cum o să nu plece? //	*Cum n-o să nu plece?	DR
		how not-will sbjv go		how will sbjv not go	how not-will sbjv not go
		‘How come s/he won’t go?’

Crucially, a single NegP indicates a single TP, a monoclausal structure.

The second observation concerns the status of the ‘will’ form, which only forms a constituent with the subjunctive. Specifically, in DR, poate ‘can’ may stand alone in answer to a question, whereas o/are may not, as shown in (22).

(22)	a.	Poate	să plece?—	Da, poate.
		can.3sg	sbjv go.3	yes can.3
		‘Can s/he go?’—‘Yes, s/he can.’
	b.	O	să plece?—	*Da, o.
		will	sbjv go.3	yes will
		‘Will s/he go?’—Intended: ‘Yes, s/he will.’
	c.	Are să plece?—		*Da, are.
		will sbjv go.3		yes will
		‘Will s/he go?’—Intended: ‘Yes, s/he will.’

The ‘will’ forms in (22) behave as modal auxiliaries that cannot stand by themselves (see also Mandić 2010): they are monolithic with the subjunctive. Auxiliaries are functional versus lexical categories, so they cannot trigger biclausal structures.10

In fact, these auxiliaries are clitics on the subjunctive verb: nothing can intervene between these elements and the subjunctive marker. Non-clitic auxiliaries (e.g., in English) can be separated from the verb by subjects or adverbs. This is not the case with ‘will’ auxiliaries in Romanian languages, which is unexceptional since the auxiliaries used in present perfect and analytic futures are also clitics (see Dobrovie-Sorin 1990, 1994, for DR) in these languages.

In view of the properties above, one could be tempted to treat these monoclausal structures as instances of verb restructuring (Cinque 2004; Kayne 1991; Rizzi 1982; Roberts 1997; 2010, a.o.). Further support in this direction comes from another well-known property of restructuring discussed by Wurmbrand (2001) for Germanic: the availability of long passives, as in the DR data in (23).

(23)	a.	Emilia		are/o	să	scrie	o carte.	DR
		Emilia		have/will.fut	sbjv	write.3sbjv	a book
		‘Emilia will write a book.’
	b.	O carte	are/o	să	fie	scrisă	(curând).	DR
		a book	have/will.fut	sbjv	be.pass	write.3sg.f.prt	soon
		‘A book is going to be written (soon).’

However, long passives are a necessary but not a sufficient diagnostic for verb restructuring, as this process is sensitive to verb semantics in Romance (see Cinque 2001 for Italian). The key diagnostic for verb restructuring is clitic climbing, which is absent in the Romanian periphrastic future. In examples that contain clitic pronouns, as in MR (16b), the clitic follows the subjunctive marker să and does not climb to the ‘will’ head. This is also systematically the case in AR and DR, as in (24).

(24)	a.	Cred	că (*l)-o/am				să-l	invit.	DR
		think.1sg	that him-will/have.fut				sbjv-him	invite.1sg
		‘I think I’ll invite him.’
	b.	tini		nu (*l’i)	va	l’i	dai	A.34	AR
		you.dat		not you.dat-it	will	you.dat-it	give.3.sbjv
		‘S/he won’t give it to you.’

Another approach takes the serial verb perspective. Aikhenvald (2006, p. 22) defines SVC as containing two verb roots that have the meaning of one word. These constructions can be of two types: symmetric (i.e., each verb has full lexical features associated with sequential activities); and asymmetric (one of the verb slots is restricted to a certain class of grammaticalized verbs). Following Aikhenvald (2006), the periphrastic future discussed in this paper qualifies for asymmetric SVC.

The general consensus in the literature is that SVCs are monoclausal, without any marker of clausal connection (e.g., no complementizers; Riis 1854, p. 103). Following Lovestrand (2021), the main criteria for establishing the monoclausal status of SVCs are as follows: (i) location and scope of negation; (ii) shared TAM field; (iii) shared argument structure; (iv) single event conceptualization. Periphrastic future constructions meet all these criteria. The location and scope of negation are discussed for the examples in (21): the single negation is high and scopes over both verbs.

As for the TAM properties, the subjunctive is devoid of any tense feature. Although subjunctives are associated with the [-finite] feature elsewhere, here, the periphrastic future has a [+finite] feature associated with the ‘will’ auxiliary. Accordingly, să lacks the property that would qualify it as a Fin element, and its function is reduced to marking the inflectional [mood] in the TP field. When it comes to aspect, this can only be marked once (recall (2c) for DR repeated here as (25a)). Equally important is the fact that (2d) for DR, repeated here as (25b), shows that there is no Asp field independently projected in the subjunctive, so the DR canonical perfective marker fi ‘be’ is ruled out. Crucially, (25) further signals the sharing of a single inflectional field.

(25)	a.	O/	are/avea		să	devină	scriitor.	DR
		will	have.fut/have.fut.perf		sbjv	become.3sbjv	writer
		‘He will/will have become a writer.’
	b.	*O/are		să	fi	devenit	scriitor
		will/have.fut		sbjv	be	become	writer
		Intended: ‘He will have become a writer.’

Furthermore, stripping să of its complementizer features makes it possible to delete it—a phenomenon seen in AR, as shown in (20b). In sum, the criteria for SVC proposed in Riis (1854) are met: there is no clausal linker between the two verbal heads.

In the periphrastic future, the argument structure is determined by the properties of the subjunctive verb. The ‘will’ auxiliary has functional but not lexical features, so it is not involved in argument selection. Thus, the construction also meets the criterion of single-event conceptualization and qualifies as a macro-event. Lovestrand (2021) points out that construction has the macro-event property if temporal operations such as time adverbials, temporal clauses, and tenses necessarily have scope over all subevents encoded by the construction. This is the case for the periphrastic future in Romanian languages, as shown in (26).

(26)	a.	Chiruta	nu-şi			şti	că	tora	va-l’	tal’u	AR
		old.the	not-refl.dat			know	that	now	will-her.dat	cut.1.sbjv
		caplu				(from Vrabie 2000)
		head
		‘The wretched woman doesn’t know that now I’ll cut off her head.’
	b.	Bine	că		acum	o	să-l		văd.	DR
		well	that		now	will	sbjv-him		see.1
		‘It is good that now I’ll see him.’
	c.	Nu	ari	si		vin	moine.			MR
		not	will	sbjv		come	tomorrow
		‘I will not come tomorrow.’

The adverbs in (26) scope over both verbal forms, with no possibility of separate readings with each verb.

In sum, in the absence of clitic climbing with the periphrastic future in Romanian languages, we adopt an SVC analysis, where the higher verbal head is merged as a T [future] head selecting a low subjunctive MoodP with phi-features capable of hosting clitics.

5.2. Syntactic Derivations

Avram (1999, p. 43) mentions the well-known fact ‘that future forms represent highly modalized means of describing possible courses of affairs.’ Consequently, the availability of să subjunctives with future denotation should come as no surprise considering the intrinsic [modal] nature of să. However, as previously noted, its intrinsic [-finite] specification prevents să from merging as a Fin head in the periphrastic future. Consequently, we propose that nothing merges in the Fin head in these constructions and that the [+finite, modal] features of Fin are satisfied compositionally as a result of long-distance Agree operations established with two distinct inflectional heads: T, on the one hand, and mood, on the other.

Along these lines, the periphrastic future in DR has the CP structure shown in (27a) and the IP structure shown in (27b). This analysis adopts Cinque’s (1999) proposal of two T heads in the inflectional clause hierarchy and Ledgeway’s (2018) proposal that a low position for clitics is exploited in Romance languages. Corbeanu and Hill (this volume) identify this low clitic position to be the second T head in Romanian languages (i.e., higher than ‘in-situ’ postverbal subjects).

(27)	a.	CP:	Force > …		Fin		> IP
			[+finite]		[+finite, modal]
		că	Ø
		that
	b.	IP:	(Neg) >	T1	> Mood > T2		> <v> > <V>
				[fut, φ1]	[sbjv]	[φ2],verb
			(nu)	o/are	să	(cl)	V
			not	will/have.fut	sbjv	(dat-acc), verb.sbjv

The Fin head in (27a) checks its features compositionally by probing twice: T₁ for finiteness and mood for [modal]. The phi-features are transferred from Fin to each T head. More precisely, Fin [+finite] is checked via long-distance Agree against T₁ [fut]. Since reduced o (< ‘will’) is too semantically bleached to also check the [modal] feature of Fin, while are (< ‘have’) lost its modality in this context, Fin [modal] is checked by the subjunctive mood marker să, also via long-distance Agree. None of these auxiliaries can move to Fin since they are clitics, so obligatorily hosted by the lexical verb moved to T₂ (i.e., <…> indicate copies of moved lexical verb). This also explains why negation precedes both auxiliaries.

Since these clauses are finite, phi-features are available and transferred to the probed domains, as also indicated by the DR data, where the modal are shows subject-verb agreement, as in (28).

(28)	a.	Eu o/am	să plec.
		I will/have.1sg.fut	sbjv go.1sg
		‘I will go.’
	b.	Ea o/are	să plece.
		She will/have.3sg.fut	sbjv go.3sg
		‘She will go.’

Linearly, in the periphrastic future, pronominal clitics do not raise above the lower T₂ arguably because [mood] is also independently probed for by Fin discharging another φ domain (i.e., φ₂). Consequently, the low IP position for clitics (cf. Ledgeway 2018) is exploited by the object clitics.11

Lastly, a special note is required for the DR form avea, which denotes the future perfect, as in (25a). While this form is an auxiliary on par with o/are, it is not a clitic. Thus, it can be separated from the subjunctive string, as in (29a), whereas this is not possible with o/are (29b).

(29)	a.	(Toţi)	aveau	(toţi)	să (*toţi) plece (toţi).
		all	will.have	all	sbjv all go.3.sbjv
		‘All (of them) will have left.’
	b.	(Toţi)	or/au	(*toţi)	să (*toţi) plece (toţi).
		all	will/have.3pl.fut	all	sbjv all go.3.sbjv
		‘All (of them) will leave.’

Nonetheless, constructions with avea, which occur only in DR, are still monoclausal: on a par with o/are futures, periphrastic future with avea allows for long passives, as in (30a), and negation is high, and scopes over both the auxiliary and the subjunctive, as in (30b).

(30)	a.	O carte		avea	să	fie		scrisă			(curând).
		a book		have.fut.perf	sbjv	be.pass		write.3sg.f.prt			soon
		‘A book was going to be written (soon).’
	b.	(Toţi)		nu	aveau	(toţi)	să	(*nu)	(*toţi)	plece	(toţi).
		All	not	have.fut.perf		all	sbjv	not	all	leave	all
		‘Not all (of them) would have left.’

The perfective reading of avea arises, on the one hand, from the relation between the Reichenbach (1947) event time (ET) and reference time (RT) being expressed by the past participle (cf. Giorgi and Pianesi 1989), while, on the other hand, the relation between speech time (ST) and RT is expressed by the T₁ head hosting are ‘have.fut’. Hence, formally, we maintain the structure in (27), enhanced as in (31) to include aspectual properties on the lower T₂ head, and account for (29b) as a matter having to do with the non-clitic nature of avea: in particular, its non-clitic status does not require T₁-mood adjacency. Lastly, the relationship between T₁ and T₂ is also one of long-distance Agree.

(31)	a.	CP:	Force > …			Fin	> IP
			[+finite]			[+finite, modal]
			că			Ø
			that
	b.	IP:	(Neg) >	T1 >	Mood >	T2	> <v> > <V>
				[ST/RT:fut, φ1]	[sbjv]	[ET/TR:Perf, φ2]
			(nu)	avea	să	(cl)	V
			not	have.pst	sbjv	(dat-acc),verb.sbjv …

Note that Avram (1999) shows that while both avea ‘have’ and fi ‘be’ can mark perfectivity, these are not interchangeable in DR. In particular, the author suggests that avea is marked for realis contexts, while fi is marked for irrealis. Since the future is, by its very nature, irrealis, fi might be expected across the board. This is, indeed, the case with the analytic, Romance type, ‘will’ + infinitive future in DR, as seen in (32).

(32)	a.	Va	veni.
		will	come.inf.
		‘S/he will come.’
	b.	Va	fi	venit.
		will	be.inf	come.pst.prt.
		‘S/he will have come.’

So, what we see is that both analytic and periphrastic futures may express perfectivity in a monoclausal structure. The encoding of perfectivity is, however, distinct, requiring fi in the analytic future but surfacing directly on the modal auxiliary in the periphrastic future. In light of the structures proposed in (31b), it means that the perfectivity feature is on T₂ with the analytic future but on T₁ in the avea periphrastic future. The relevant point is that perfectivity is mapped only once, unlike what happens in Macedonian biclausal structures (2a, 2b). The exclusion of T₂ as a head for perfectivity with are futures has to be related to the fact that T₂ is not available for association with perfectivity, as this head has the phi-features responsible for hosting pronominal clitics and the subjunctive verb.

The periphrastic future in AR and MR are sub-cases of the DR structures shown in (27), as long as the structure is monoclausal. Thus, the AR biclausal constructions with nore, as in (19), do not qualify for this analysis but for an analysis with obligatory control verbs. On the other hand, the AR constructions with the modal auxiliary va, such as seen in (20), conform to the template in (27). The micro-variation here is that the subjunctive marker is optional. When s- is present, as in (20a), the checking operations proceed exactly as in (27) for DR. When the subjunctive marker is absent, as in (20b), the verb moves to T2 to check [mood] and [modal] in Fin through a long-distance Agree.

On par with AR, MR periphrastic future conforms to the pattern in (27), the variation arising from the pairing of lexical items with the relevant functional features. In particular, the negative constructions in (16a, 16b) versus (16c) indicate that the subjunctive marker is unable to check [modal] in Fin at long distance, so a modal item, such as the negation, is obligatory to take over this task. Alternatively, the subjunctive marker can check all the relevant features when it is in Fin, as in (17), where the modal auxiliary is excluded. The variation here is that MR si/să is strongly associated with conditional CP and, hence, with the [+finite] feature, whereas the same association disappeared in DR and is weak in AR. The future tense interpretation arises from the context, through the principle of compositionality, in such constructions.

In sum, subjunctive complements could be used for periphrastic future derivations because of the following: (i) the subjunctive is used for the encoding of irrealis modality; (ii) the subjunctive marker has been associated interchangeably with +/− finiteness and modality, being also paired with a specific grammatical mood; and (iii) the subjunctive marker can be used to check all the afore-mentioned features or only some of them, depending on whether other elements may take over some of the checking function compositionally. In most periphrastic futures, the modal auxiliary takes over the checking of finiteness, so the subjunctive marker is confined to mood and modality. Formally, the structures in (27) equally underly the periphrastic future in AR, DR, and MR, although each Romanian language variety developed this construction independently. Thus, micro-variation is expected, but it concerns only the options for feature checking, not the featural make-up of the syntactic derivation.

6. Conclusions

This paper discusses subjunctives in Romanian languages in connection to Old Romanian and the Balkan Sprachbund. It traces the diachrony of selected subjunctives in DR, AR, MR, and IR and explores the micro-parametric variation in their CP fields, as well as the potential for să subjunctives to derive periphrastic futures in these languages.

As part of the Balkan Sprachbund, Romanian languages have a preverbal morpheme that marks subjunctives, in addition to various degrees of inflectional morphology on the verb. Importantly, unlike in other Balkan languages, but on par with Romance, subjunctives also allow for a complementizer of the ‘that’ type in certain contexts. The subjunctive CP in Romanian languages projects fully (i.e., a ForceP domain) with subject obviation but only partially (i.e., a FinP domain) in control contexts, following the Balkan Sprachbund template resulting from the loss of infinitival complements. However, unlike other Balkan languages, Romanian languages allow for a split Fin at various historical stages. While standard DR has completely remerged this split Fin, DR să currently being able to satisfy both [finite] and [modal] features of Fin, this process is currently incomplete in AR, MR, and IR.

With respect to futurity, Romanian languages developed the periphrastic future with subjunctive morphology, as in other Balkan languages, in addition to allowing for the Romance analytic future with the infinitive in some cases (i.e., Old Romanian, DR). We argue that this was the result of two factors: (i) lexical ‘want’ and/or ‘have’ undergoing semantic bleaching and grammaticizing into T heads denoting [future]; and (ii) embedding of să subjunctives under these reanalyzed ‘will’-type auxiliaries. We show that these Romanian periphrastic futures have a monoclausal, as opposed to biclausal, structure that is best captured under a serial verb construction of the asymmetric type. The [+finite, modal] Fin specification in these derivations is satisfied compositionally via long-distance Agree, with two distinct inflectional heads: ‘will’ T [future] checks [+finite], while să in mood checks [mood] and [modal]. The splitting of Fin feature checking is forced by the intrinsic [-finite] property of să, which is blocked from merging as a Fin complementizer in these derivations (hence, the monoclausal status of the periphrastic future).

In periphrastic futures, micro-variation arises, first, from the degree of grammaticalization undergone by the modal ‘will’/’have’ as an element signaling future events: it could be a verb with obligatory control (i.e., in AR) yielding biclausal structures; or a modal auxiliary (clitic or non-clitic), yielding monoclausal structures, which is the default option in AR, DR, and MR. Although the underlying configuration is the same for these constructions in the three languages (i.e., identical list and distribution of formal features), micro-variation arises in the implementation of feature checking.