The auxiliary of the Romanian conditional : semantic and functional arguments concerning the reconstruction of a disputed grammaticalization process

This paper aims at demonstrating the explanatory advantages of the old hypothesis concerning the origins of the auxiliary of the Romanian analytic conditional (aș + infinitive) as deriving from the imperfect tense form of the verb (a) vrea ‘(to) want’ < *volere (< VELLE). The grammaticalization process, reconstructed through the comparison with the other Romance languages and by relating it to typical directions of the linguistic change, presupposes intermediary semantic phases (the future-in-the-past value, the hypothetical value which is mostly counter-factual), whose traces may be found in the first Romanian (translated) texts, but which have been generally considered a consequence of the simple loan translation from the language source. The uses of the conditional with a reduced auxiliary (aș, ai, etc. + infinitive) are related to those (co-occurring in the old texts) of the conditional with a recognisable auxiliary (vrea ‘wanted’ + infinitive), for which the value specific to the first grammaticalization stage is still strong—that of a future-in-the past. The text also puts forth an explanation for the atypical forms within the paradigm of the conditional auxiliary—aș(i), ar(ă)—through the overlap between the forms of the imperfect and the simple past of the verb (a) vrea ‘(to) want’.


Introduction
The origin of the auxiliary used in the formation of the paradigm of the conditional mood in Romanian constitutes the topic of a long standing debate.The main hypotheses are those which presuppose that at the origin of the auxiliary paradigm lies (in various modal-temporal states) (a) the verb habere; (b) the verb *volere; besides these, there also exist compromise solutions (e.g., the idea of a contamination between the two paradigms) or stands acknowledging lack of decision.In support of the various explanations, parallelisms with the Romance conditional or connections with other periphrases of the conditional in the old language, with the forms of the conditional in Istro-Romanian and the dialect of Banat have been mentioned.The various theories have sometimes provided additional etymological explanations for the atypical forms aș (1st person singular) and ar (3sg=3pl), though not sufficiently convincing.
We intend to browse through the main arguments brought forth by the supporters of each of the hypotheses and to reinterpret them by shifting the focus from those explanations based primarily on the evolution of forms to those which take into account the regularities of the grammaticalization process as a whole, with the involvement of the semantic and functional factors.This type of approach favours the hypothesis according to which the auxilary developed from the verb a vrea 'to want' , a hypothesis which enjoys the advantage of a tight correlation between the presupposed changes and the semantic and functional peculiarities of the conditional in general and of the Romanian modal-temporal system in particular.
In the earliest available texts, originating from the 16 th c., the conditional auxiliary is attested (in anteposition) within the forms: 1sg aș/ așu/ ași 1 ; 2sg ai; 3sg=pl ară/are 2 ; 1pl am; 2pl ați.A segmentation problem appears with the forms following the infinitival form of the verb which will be discussed later in the paper: the forms are preceded by the sequence -re-which could be either the ending of the infinitive or the initial segment of the auxiliary 3 .
The construction seems old and strongly grammaticalized: it exclusively contains the infinitive without a (which indicates its stabilization to be anterior to the association of the preposition a to the infinitive), and the specific forms (aș, ară) never appear by themselves in other combinations, but are specialized for conditional marking.The other forms within the auxiliary paradigm are also functionally specialized (for the collocation with the infinitive and for the conditional value of the periphrasis), even if they enter homonymy relations with other verbal forms: the form 2sg ai is identical to that of the present indicative of the verb a avea 'to have' and to that of the auxiliary for the compound perfect auxiliary also originating from a avea (for the same grammatical person), the 3sg variant are is identical to the same person of the present of the verb a avea (to have), and 1pl am and 2pl ați are homonymous to the auxiliary for the compound perfect.
The form for the 3sg is the only one which changes in time, from the time of the first attestations; ară, which is dominant in the 16 th c., persists into the 17 th c. (in mi~1620, Prav.1652(in mi~1620, Prav. , în sa.1683(in mi~1620, Prav. , în bb.1688, etc.), etc.), just like are (Prav.1646(Prav. , bb.1688, etc.), etc.), but the two variants start being replaced by ar, which had previously been occasionally used 4 .Towards the end of the 18 th century, the form ar gets generalised.Gheție (1994, p. 28) considers that ar is an innovation that begins from the Southern area and extends into Moldavia and then later into the trans-Carpathian areas, where are and ară have been recorded by dialectal investigations even during the 20 th century (see, for instance, Papahagi, 1925, p. LXVIII).The ending of the 1 st person sg.varies both in the language during the previous centuries as well as in presentday Romanian, being interpretable as a palatalized pronunciation of the final consonant ș 5 vs. a nonpalatalized one.

The hypotheses concerning the origin of the auxiliary
The hypothesis regarding its origin from *volere and the one taking habere as the source have had numerous supporters in time; the latter has also given rise to many variants depending on the selected modaltemporal paradigms, often combined, through which the forms of the auxiliary have been explained.Among the various supporters of both hypotheses there are some differences concerning the relative chronology of the transformations.Densusianu (1975, p. 572) and Lombard (1954Lombard ( -1955, p. 962-963) , p. 962-963) have found it preferable not to adopt any of the main theories on the origin of the conditional auxiliary 6 . 1 The form așu is first attested in ph.1500-10; ași appears in cv.1563-83, occasionally in po.1582, etc.; aș is the dominant form in po.1582.Several examples in the texts are reproduced by Densusianu (1975, p. 569).The frequency of the forms may be verified by resorting to the platfom Corpus electronic al textelor românești vechi [Electronic corpus of old Romanian texts] [online].
2 The form ară is very frequent in ph.1500-10, cc 2 .1581,ct.1560-1, po.1582, etc.A variant of it, aru, is attested in a document dating from 1593 (Frâncu, 2009, p. 121).In order to identify the forms within the texts, see the indication in the preceding footnote. 3The segmentations belong to the editors, all these forms containing a postposed auxiliary being written without a break: vreareaș, vreareați, vreareară, etc.  4 Frâncu (2009, p. 122) considers that the short form ar had not yet appeared during the old period of literary Romanian (until 1640), contesting the authenticity of several previous attestations.Nevertheless, there exist some attestations of ar in the 16 th century (e.g., in po.1582, see Arvinte, 2007, p. 317) and from the first decades of the 17 th century (e.g., in a.1620, 16 v ).The form ar becomes quite recurrent in Prav.1652, but it is still rare in Prav.1646, bb.1688 (Arvinte, 2004, p. LI), etc.  5 In the grammar of Diaconovici Loga (1822) there are inconsistencies even in the conjugation tables where both aș and ași are mentioned (see, for instance, p. 104 and p. 110). 6Densusianu's statement is often quoted:"le conditionnel formé avec ași reste encore une énigme de la morphologie roumaine" [the conditional formed with the aid of ași still remains a mystery of the Romanian morphology] (1975, p. 572).
The modal-temporal forms considered as a point of departure in the development towards the status of an auxiliary8 are the present (Titova, 1959), the pluperfect of the indicative or the imperfect of the subjunctive (Tiktin, 1904), the perfect indicative (Bugeanu, 1970;Elson, 1992) and often a combination of paradigms-the imperfect of the indicative, the imperfect of the subjunctive, the past perfect of the subjunctive, etc.These explanations firstly take into account the form and have no interest in a coherent motivation of the semantic development (from a avea 'to have' with various temporal values to the optative and conditional).The already mentioned points of formal relatedness have influenced the preference for the explanation through habere: the coincidence of some forms with the auxiliary in the compund perfect (ai, am, ați), the relatedness and identity with the form of the lexical verb a avea (ar/are).The habere theory has also rested on the comparison with Romance, favouring the auxiliary which acted as a source for the forms of the conditional in the other Romance languages.This is, however, a punctual and partial comparison which does not take into account the lack of similarity in the formation of the future.The habere hypothesis presupposed clefting of the construction with the postposed auxiliary wherein the main verb would have the long infinitive (fire-ar); Hasdeu (1892Hasdeu ( , p. 1946) considered that the reanalysis of the sequence of type fire-ar and the different clefting-fi-rear-might have led to the emergence of the auxiliary forms in Istro-Romanian and in the dialect of Banat (with the -re from the long infinitive having become their initial segment): rear fi, an explanation which is later resumed by numerous supporters of the habere hypothesis (Tiktin, 1904, a.o.).
The main arguments in favour of this explanation come from the comparison with Istro-Romanian, with the dialect of Banat and with the periphrases in older stages of Romanian which contain unmodified forms of the verb a vrea 'to want' .The *volere theory also rests on the comparison with Romance, retaining the peculiarity of evincing the same auxiliary for the future and the conditional: habere in the Western Romance languages, *volere in Romanian.The current form of the auxiliary would then come as a consequence of the apheresis of the initial consonants of the verb (am < vream, ați < vreați, etc.), and probably as a result of a reanalysis.The frequently post-verbal position of the auxiliary in old Romanian allows the reconstruction of a process which probably started from the long infinitive and the auxiliary with the apherised initial v-.In fire-(v)reați a haplology would thus have taken place through the dropping of the reiterated syllable re.Arvinte (2004, p. LI) considers that the syllable fell out of the auxiliary-so that the auxiliary retains the same form with anteposition or postposition, and the postposed infinitive is long; fire, in fire-ați.One may assert-as Weigand (1896) did, followed by Meyer-Lübke (1900, p. 356), Morariu (1925, p. 57)-that the syllable has got astray from the form in the infinitive, so that the auxiliary is in its inversion form nearer to the origin (reaț), and the anteposed infinitive is identical to the postposed one, that is, short: fi, in fi-reați 10 .
Irrespective of how the simplification of the repeated syllable in older sequences of the language might have occurred, it is possible that the sequence of type fi-reați (verb-auxiliary), used alongside reați fi (auxiliary-verb), might have been reinterpreted as including the long infinitive (during a period of time when it was still employed as a variation to the abbreviated form).The new segmentation fire-ați could have influenced the anteposed form of the auxiliary, thus producing or at least accelerating the shortening process for ați.The forms of the auxiliary could thus be the result of an apheresis (of the initial consonant v-), but also of a reanalysis of the sequence with postposition.This hypothesis represents the perfect symmetric overthrow of the quite well-spread idea (v.supra, §2.1) that the initial form fire-ați would have been reanalysed as fi-reați, giving rise to the variants of the auxiliary in Istro-Romanian and in the dialect of Banat.

The mixed paradigm hypothesis
A combination between the forms of the two auxiliaries has also been advanced by several authors (mentioned by Elson, 1992, p. 363).Frâncu (2009, p. 121-122) considers that an overlap between paradigms took place so that the auxiliary originates from the imperfect of *volere in the case of the 3sg and 3pl persons, from the present of habere in what concerns the 2sg, 1pl, 2pl persons and from the imperfect of *volere or that of habere when it comes to the 1sg person.Again, the explanation is only determined by the possibility of some phonological development, with no semantic justification regarding the process of grammaticalization of the auxiliary with respect to such blending of verbs and tenses.

Semantic arguments and characteristics of the grammaticalization path
An important argument for the *volere theory is supplied by the grammaticalization model of the conditional in relation to the future.As Fleischman (1982) showed, the conditional in Romance started as a future-in-the-past.The two periphrases-of the simple future and of the future-in-the-past-have grammaticalized at the same time (although there exist discussions about the possible precedence of one over the other form) and in obvious interrelation, having an auxiliary developing from the same verb11 .

The future and the conditional
If we accept the *volere hypothesis, from a typological perspective, Romanian ranges with languages which use the same auxiliary and the same morpho-syntactic pattern for the future and the conditional ("velle serves both for the future and the conditional", Iordan, 1928, p. 197) and which derive the conditional from the future-in-the-past 12 .It is the dominant model in Romance, but it is quite well-spread in other languages from various families as well; one of the most studied cases (and the most interesting for a comparison with Romanian) is that of the modal will in English, whose present form is used for the future, and whose etymological past form (would) becomes the auxiliary of the conditional (Bybee, 1995;Iatridou, 2000, etc.).
In 16 th century Romanian, both the future of the type voi face, and the conditional of the type aș face were completely grammaticalized.The high frequency of the future construction voi face ('I will do' , with the verb 'will/want'), the infinitive without a and its (exclusively) temporal sense indicate it to be older than other periphrases for the future (more rare, built with the aid of the a-infinitive or with the subjunctive, and having-in the case of the periphrases with a avea 'to have'-, rather a modal meaning of necessity than a temporal one).The periphrasis voi + infinitive represents at the time the prototypical future; this is why it is quite probable that a future in the past taking it as basis (by switching the auxiliary from the present to the past) might have triggered the development towards the conditional value.The periphrasis of the type va face 'will do' (present + infinitive) enabled the development of a periphrasis vrea face 'would do' (imperfect + infinitive), just as are să facă 'has to do' will give birth to the future-in-the-past avea să facă 'had to do' (Zafiu, 2013, p. 40;Popescu, 2014).
The future with a avea 'to have' undergoes grammaticalization at a later stage, and it does so incompletely; the construction of the verb a avea with the a-infinitive (are a fi 'has to be') is rather rare in the 16 th century, and only has a modal value, that of necessity (Călărașu, 1987, p. 187;Zamfir, 2007, p. 232-241).
The meanings which the construction aș + infinitive develops diachronically-of posteriority in the past and of possibility/hypothetical meaning-also count as an argument in favour of the *volere hypothesis.Titova (1959) supported the habere hypothesis by observing that aș + infinitive in the old texts had a meaning that would be nearer to the future than to the conditional, which seems to indicate a transformation of the construction-from a modal value of necessity ('has to do') to a future value first and then to a conditional one.The relatedness of the values of the future and the conditional and even the common contexts of occurrence (in the conditional construction) do not, however, presuppose equivalence and overlapping.There are no precedents to justify the simple diachronic transformation of a future into a conditional; on the other hand, the association of morphological markers for future with those for the past represents the typical source of a mood of the hypothetical condition-the conditional (James, 1982;Bybee, 1995;Iatridou, 2000).
In the analysis of old Romanian texts the value of the present or past conditional was established by the researcher in a rather arbitrary way, often through the mere replacement possibility with the current forms.For a better understanding of the values of the periphrases under discussion, we start from the observations in Iatridou (2000): the conditional is either counterfactual or non-counterfactual, the counterfactual interpretation being an implicature.The perfect conditional is counterfactual; the present conditional may be conterfactual but it is not necessarily so, depending on the meaning of the verbs and the context.Counterfactuality as a semantic feature does not thus overlap with the morphological opposition between the perfect and the present conditional.
The transformations which interest us and which lead to the fixing of certain modal values of a modaltemporal periphrasis have been preceded by the grammaticalization of the modal-temporal periphrasis starting from a free combination of words, having a modal value 13 .It is natural to presuppose (also by comparing developments in other languages) that the periphrasis *volere + infinitive initially expressed will, intention, from where the switch to expressing posteriority is easily reached (Bybee et al., 1994, p. 243-279;Heine & Kuteva, 2002, p. 310-311).The future, through its semantic component of not (yet) actualised facts still continues to have double and ambiguous status, with the open possibility of considering it both as a mood as well as a tense14 .
The future periphrasis (va + infinitive) could be used (through the modification of the tense auxiliary) as future in the past (vrea + infinitive): voi face (intention) → voi face (future) (I) will do 'I want to do' → (I) will do vrea(ș) 15 face (intention in the past) → vrea(ș) face (future in the past) (I) would do 'I wanted to do' → (I) would do 'I was about to do' The periphrasis for the future in the past, which has the characteristic to morphologically mark both the past as well as the future, constitutes the typical source for the (counterfactual) hypothesis in many languages, in the conditional construction and in desirative sentences (Iatridou, 2000).The hypothetical value extends afterwards to the present in sentences which can be counterfactual as well as noncounterfactual.With the rise of a specialised construction for the counterfactual hypothesis referring to the past (past conditional), the first form of the conditional restricts its meaning turning into a present conditional.
Hence, a periphrasis of the type vrea(ș) face '(I)would do' undergoes the following path of semantic change in time: (I) 'I had to do' (future in the past) → (II) 'I would have done' (hypothetical-counterfactual) → (III) 'I would do' (hypothetical-counterfactual or non-counterfactual) In the second stage, the periphrasis may be equivalent both with a past conditional and with a present conditional with a counterfactual interpretation.In the third stage, its value as a present conditional becomes fixed.The temporal value from the first stage may persist in certain contexts.
Such a change regarding the specific value of the Romanian conditional has been proposed by Ivănescu (2000), by analogy with the developments in other Romance languages, in Greek, etc.: "The initial meaning of the Western Romance form, which currently functions as unreal present, was that of unreal past.This «I would have sung» meaning must have been also held by the Romanian present irrealis" (p.163).

Traces of the grammaticalization process in the old texts
In order to verify the grammaticalization path described above, one needs to establish first whether in the old texts the periphrasis ar + infinitive had a temporal value of future in the past or not.In spite of what has been often claimed (Coene & Tasmowski, 2006;Zamfir, 2007, etc.), the value of the future in the past (posteriority in the past) seems to have persisted in the 16 th century, with the periphrasis subordinated to a declarative verb in the past (1) or in a narrative context, competing with the simple past ( 2 15 The 1 st person form vrea is attested in the 16 th century as the auxiliary of the conditional; vreaș is a reconstruction starting from the current forms aș (Daco-Romanian) and reaș / ręš (the dialect of Banat, Istro-Romanian).
(  Densusianu (1975, p. 572;cf. Zamfir, 2007, p. 326), where the conditional is a narrative tense with a perfect value.It is very probable that this is a linguistic loan translation from Hungarian, as it has been argued, but this indicates the persistence of a certain temporal value of the periphrasis, which makes it compatible with the contexts of the past: ( This narrative usage also appears in other old texts, generally associated with posteriority in the past; the biblical sequence in (5-6) is generally translated by means of past tenses (in the Greek original the first is an aorist, the second an imperfect) 16 ; the imperfect in nt.1648 and bb.1688 corresponds to the conditional in ct.1560-1: (5) At a later stage, the temporal value disappears so that it was asserted that the Romanian conditional is a true mood, illustrating the Eastern conditional (Thieroff, 2010), and not primarily a tense which also develops modal values (an opinion which has often been advanced regarding the French conditional).The lack of temporal value (of a future in the past) in present-day Romanian has been explained through the absence of the SOT ('sequence of tenses') parametric option (D'Hulst et al., 2004).The negative specification of 16 Cf.http://biblehub.com/interlinear/mark/6-56.htm.
the SOT parameter actually explains why the temporal value, which must have existed in the first stage of the grammaticalization of the periphrasis, disappeared.
The value derived from that of posteriority in the past is that of possibility, hypothesis17 .One would suppose that in an intermediary stage of the grammaticalization process, the periphrasis referred to past deeds (as long as the auxiliary was perceived as temporally marked), with counterfactual implicatures.The 16 th century texts do not seem to contain any uses of the periphrasis aș face which would illustrate this particular value (equivalent to a past conditional); the value is however occasionally realised as vrea face (v.infra).
The values of the present conditional illustrating the final stage of the grammaticalization process are already present in the 16 th century texts, being visible in the conditional construction (in the presence of the specific conjunctions să or de 'if ') as hypothesis ( 7), as an attenuator in other types of sentences (8) or as a marker of possibility ( 9 The periphrasis aș + infinitive (no longer) has the value of a perfect conditional in the 16 th century, because this value is rendered through the periphrasis of the verb to be and the participle-aș fi făcut 'I would have done ' . As Iatridou (2000) shows, the perfective marker imposes the temporal interpretation of a verbal construction and the doubly marked temporality may distribute thus ensuring the interpretation of counterfactual in the past 18 .One of the markers (aș) functions modally, and the other (the participle)temporally.
The periphrasis with a perfect value is already very well represented in the 16 th century texts, which may attest its age.The temporal-aspectual model characteristic to Romanian in the 16 th -17 th centuries-(a) basic form; (b) progressive periphrasis with a fi 'to be' + gerund; (c) perfective periphrasis with a fi + participle (Frâncu, 1983(Frâncu, -1984;;Zafiu, 2016, p. 32)-presupposes a distribution of the temporal and aspectual values.The basic periphrasis-ară putea (11), ară trebui (13) and the periphrasis with the gerund-ară fi fiind (11, 12)-have the value of a present conditional 19 ; the periphrasis with the participle-ară fi pus 'would have put' (12); ară fi postit, ară fi ținut 'would have held ' (13)  not=have fasted they then we toțe-am fost căzuți den dulceața raiului (cc 2 .1581,46) all=have been fall.pplefrom goodness.defHeaven.def.gen'And if they had fasted, and if they had refrained themselves from eating, we would not need to fast now; but since they did not fast then, we all fell from the goodness of the Heaven' In the examples above, the asymmetric distribution of the periphrases for the conditional simply reflect the semantic necessities of the text.In protasis it refers to the present or to the past and in apodosis it refers to the present or to the future.

The competing periphrases
In the 16 th century texts there are several periphrases whose conditional value is built with the aid of the auxiliary a vrea 'to will/want' in an identical form with that of the predicative verb; many of them circulate during the next centuries, some of them being retained at a regional level (Frâncu, 2009, p. 122-123, 317-318).The constructions built by means of an auxiliary and the infinitive are two, one with the auxiliary in the imperfect (vrea face 'I would do') and the other one with the auxiliary in the compound past (am vrut face 'I would do') 20 ; the main progressive construction is made up of the imperfect of the verb a vrea, the infinitive of a fi 'to be' and the gerund of the lexical verb (vrea fi făcînd 'I would be doing').According to the temporal-aspectual system in the old language, the picture is completed with constructions with the past of the verb a vrea, the infinitive of a fi and the participle of the lexical verb-vrea fi făcut, am vrut fi făcut 'I would have done' (Densusianu, 1975, p. 571)-which structurally and semantically correspond to the past conditional of the type aș fi făcut 'I would have done'; the constructions above may give rise to some doubly-compound forms, used rather rarely.

Controversies
The existence of periphrases with the imperfect of a avea has been considered both an argument in favour of the *volere hypothesis (proving that a variant of the auxiliary enters the formation of a conditional), and an argument against the same hypothesis.On the one hand, it has been argued that it was strange that only forms which illustrate the initial and the final stage of a grammaticalization process resisted and circulated simultaneously, but not those characteristic to its intermediary stages (Tiktin, 1904;Titova, 1959).On the other hand, it was claimed that the periphrasis only had a past value (an unaccomplished possibility in the past), while the present conditional with aș has a present or a future value (a possibility that can be realised in the present or in the future or not).
Both objections can be discarded.
(a) The periphrasis vrea + infinitive may be a case of reconstruction with the same elements, as it sometimes happens in the history of Romanian, even in the case of the verb a vrea (veri 'you will/want' circulates simultaneously with ori, vare 'he will/wants' with oare, and the parallel forms are attested for quite a long time 21 ).The idea was put forth by Philippide ( 1894), who considered it as a case of repetition of a pattern-"it is as if the language would have wanted to repeat the old process" (p.69); cf.Ivănescu (2000, p. 355).On the other hand, it would not be impossible that the periphrasis vrea + infinitive be a mere residue, a variant with certain morphological peculiarities, kept on a par with the grammaticalization of the reduced form.If, as we will argue in the following sections (infra, §6), aș comes from a hybrid paradigm of a vrea, related to the simple past of the auxiliary, the simultaneous existence of the periphrases using the regular paradigm of the imperfect is all the more possible.(b) In most cases, the periphrasis vrea + infinitive has the value of a present conditional.The quite well spread view according to which its main value is that of a past conditional (Densusianu, 1975, p. 571;Rosetti, 1978, p. 565;Frâncu, 2009, p. 123;Zamfir, 2007, p. 346) comes from the union of two different uses of the periphrasis in the language during the previous centuries-that of future in the past and of present conditional-but also from those cases where this seems to be equivalent to the perfect.From this point of view, the pattern vrea face evinces an obvious similarity with the pattern ar face: the difference consists in the fact that ar face, which gets grammaticalized at an earlier stage, has fewer uses as a future in the past, while vrea face has numerous uses of future in the past and still retains some of the uses of the counterfactual (past) conditional.

The periphrases vrea + infinitive; vrea + infinitive of fi + participle
The periphrasis containing the auxiliary a vrea in the imperfect and the lexical verb in the infinitive (vrea face 'would do') primarily has a future in the past value (the initial value of the Romance conditional and the one deductible from the form by way of transposing the future auxiliary into a past tense).This value frequently gets actualised in contexts in which the narrative tense is the perfect (in aspectual combination with the imperfect): 20 See Densusianu (1975, p. 571), Zamfir (2007, p. 346-355).The second periphrasis uses a compound form as an auxiliary which has its own auxiliary (which indicates a high degree of grammaticalization of the compound perfect). 21 In the examples above, the periphrasis may be equated to the current forms of the future in the past (avea să se întoarcă 'would come back/return' , avea să poată 'would be able to' , avea să intre 'would enter' , avea să stea 'would settle'); paraphrasing through the perfect conditional s-ar fi întors 'would have returned' is possible in example (15), in the hypothetical context of a question.
The examples in which the periphrasis is equivalent to the past conditional, which are extensively discussed by Zamfir (2007, p. 346-349), seem to activate the temporal value (of imperfect) of the auxiliary, which allows the rendition of the counterfactual hypothesis with reference to the past; the examples thus illustrate an intermediary grammaticalization stage which had already been superseded by the pattern aș + infinitive (see supra): The periphrasis vrea lăuda 'would praise' persists in the following centuries and is recorded in the grammar of Diaconovici Loga (1822) from Banat.He includes several series of forms in the conjugation tables (which are common for subjunctive and optative); he labels them "present"-aș lăuda 'I would praise' , "imperfect"-vream lăuda, "perfect"-aș fi lăudat and "pluperfect"-aș fi fost lăudat (p.96-130).The classification does not necessarily represent a confirmation of the past value of the periphrasis, because it may have been influenced by the tense of the auxiliary.In the 19 th century the periphrasis was still attested in all regions and especially in Transylvania (Dragomirescu, 2015, p. 209).Pamfil & Dănilă (2009) signal the current use of the construction in Crișana and Maramureș-with a present conditional value.
The gradual specialization of the present conditional value also finds an explanation, just as for ar + infinitive, in the existence and systematic use (at least in some texts, such as cc 2 .1581) of periphrases that contain a participle explicitly marking the counterfactual hypothetical in the past: (cc 2 .1581,90) cl.refl.acc.3plaux.3pl be.inf change.mind.pple'Had the disciples not known (this), maybe they would have worried and would have changed their mind' Although they were well-represented in the 16 th -17 th centuries, the periphrases with a vrea in the imperfect became obsolete; very few regional traces have been retained to the present day.Their disappearance might be related to the fact that they simply doubled the values of the standard paradigms-aș face 'would do' and aș fi făcut 'would have done' .

The periphrasis am vrut + infinitive
The periphrasis with the verb a vrea in the compound past, although more rarely used, has the values of future in the past and of counterfactual hypothesis referring to the past (that is, of perfect conditional): ( 23 The difference with respect to the imperfect auxiliary-the absence of the value of present conditionalmay arise from perfective marking (through the participle).

The periphrasis vrea + să-subjunctive
From the future built with va (3sg person of a vrea, with impersonal meaning) + subjunctive (example 26)23 one may form future in the past by using the imperfect of the verb a vrea (27,28) The Aromanian periphrasis, made up of the invariable form vrea (the imperfect 3sg of the verb a vrea) + the present subjunctive (vrea s-aflu 'I will find out') (Saramandu, 1984, p. 459;Nevaci, 2006, p. 145-146;Nevaci & Todi, 2009), represents the same construction pattern.The main difference between the perihrasis vrea să facă/vrea s-facu 'I will do' in old Romanian and in Aromanian and the type vrea face consists in the replacement of the infinitive through the subjunctive, as a consequence of the Balkanic tendency to eliminate the infinitive, which has generalized in Aromanian.

The similar paradigms in the dialect of Banat and in Istro-Romanian
In support of the *volere hypothesis, Weigand (1896) has brought forth the arguments regarding the similarities with the periphrases in Istro-Romanian and the dialect of Banat, where the connection between the auxiliary and the verb a vrea is obvious.
The forms of the Istro-Romanian auxiliary may be very easily explained through the apheresis of the initial v-, and appear as an intermediary link between a paradigm of the verb a vrea 'want' and the even shorter forms of the auxiliary in Daco-Romanian.
The dialect of Banat has similar forms: 1sg reaș -2sg reai -3sg rea -1pl ream -2pl reați -3pl rea (Neagoe, 1984, p. 264) The forms in the dialect of Banat are attested since the end of the 18 th century by Paul Iorgovici, who-in a book dedicated to the literary language-constantly uses the present and the perfect conditional: (29) Multe exempluri se reau putea aduce many examples cl.refl.pass.3sgaux.3pl can.inf bring.inf(Iorgovici.1799, 137) 'Many examples could be brought' (30) de le ream fi cunoscut, ream zice if cl.acc.f.pl aux.1pl be.inf know.ppleaux.1pl say.inf(Iorgovici.1799,147) 'Had we known them, we would have said' A difference between the forms used by Iorgovici and the ones recorded later in the dialect is the -u ending for 3pl-an innovation of the dialect of Banat involving the forms of the imperfect, attested since the 16 th century (Gheție, 1994, p. 124) and taken over in the literary language.The presence of the ending would indicate the fact that the auxiliary was still associated to the imperfect wherefrom it had sprung 24 .
The forms in Istro-Romanian and the dialect of Banat have the value of a non-counterfactual hypothetical in the first place and correspond very well, both formally and semantically, to the present conditional of the type aș + infinitive (kept almost unchanged since the 16 th century until the present day).
The connexion of this series of forms with that of the auxiliary aș, ai, are, etc. has been explained by the opponents of the *volere theory as late influence arrived at through contamination between the auxiliary vrea (the pattern discussed in §4.1) and the aș-series.It is, however, plausible that Istro-Romanian and the dialect in Banat might have simply kept intermediary phases of the shortening of the auxiliary.The only forms that cannot be explained through the presence/absence of the re-segment are those in the 3sg and pl: ară / are / ar vs. rea / reau.
The perfect conditional is a periphrasis which contains an extra element marked as 'perfect': the dialect of Banat features the same structure as the one used for the standard perfect conditional (am fi cunoscut '(we) would have known´): ream fi cunoscut [example (30), supra].The situations signalled in Neagoe (1984, p. 264) as archaic periphrases of the perfect conditional-of the type vream cînta (1sg)-seem to be examples of the future in the past in which the auxiliary has received the more recent endings of the imperfect.In Istro-Romanian the perfect is formed differently: with the participle of a fi, and the lexical verb in the infinitive (ręš fost cânta, Kovačec, 1984, p. 577).

The morphological hypothesis
Two of the auxiliary forms have posed difficulties when it comes to explaining their origin: 1sg aș and 3sg and pl ară/are/ar.The other forms can easily be accounted for by starting from the imperfect volebam, which, through phonological development, gave rise to the series (attested in the 16 th century): With respect to aș, Philippide (1894, p. 70) has proposed an explanation in the particle și < sic, which would have been added at the end of the verb with a deictic and intensifying role.Adopted by others, this idea has nevertheless failed to convince in the absence of similar developments suggesting a regular pattern.Rosetti (1978) supposed that aș had originated from the past perfect of the conjunctive in Latin of the verb habere-habuissem.
A more credible explanation is an account regarding the appearance of the ending -ș by analogy with the 1sg ending of some strong etymological simple past forms (ziș(i) 'I said' , duș(i) 'I carried' , merș(i) 'I went' , spuș(i) 'I said' , etc.).Tiktin (1904) has hypothesised that this analogic -ș in the paradigm derives from the verb habere; Bugeanu (1970) considered that the analogic ending was added for differentiation reasons to some forms which also derived from the simple past of habere.Frâncu (2009, p. 121) suggests that the analogic ending was added to the imperfect of the mixed paradigm volebam/habebam.
For ară/are/ar, the phonetic evolution has been reconstructed in several ways.The supporters of the habere hypothesis (with the present form) have considered that the ară form reflects the strong pronunciation of r, while are and ar are the natural results of phonetic evolution26 .In the *volere hypothesis, it is assumed that the forms of 3sg and 3pl are arrived at through regular eliminations of sounds, through the modification in the pronunciation of the sonorant r or through metathesis: (a) *volebat > reare > are > ar (Arvinte, 2004, p. LI) (b) volebat > *voare > vare > are; ară through velarizing e after r; ar by dropping the final vowel (Frâncu, 2009, p. 121-122) The auxiliary paradigm may, however, be accounted for by way of the contamination between the paradigms of the imperfect and the simple past of *volere/vrea27 .As the process of forming the conditional dates probably later than common Romanian as the same construction cannot be found in the dialects spoken to the South of the Danube (Caragiu-Marioțeanu, 1969, p. 272), there is no need to seek recourse in the remote Latin etymon, as the explanation could start either from the reconstructed variants for old Romanian or from the variants which are attested in the first texts.Romanian would be closer to Italian in this respect, where the conditional auxiliary comes from the Latin perfect (Meyer-Lübke, 1895;Rohlfs, 1968).The contamination could be formally accounted for through the necessity of marking person (the imperfect having too many syncretisms) and it would be semantically acceptable within the perspective of cross-linguistic comparison due to the fact that both the imperfect as well as the perfect (i.e.simple past) could become a source of the counterfactual meaning.The fact that it is only the periphrases for the simple past which are absent from the series where the form of the verb a vrea is completely preserved (see supra, §4.) would support this explanation, if we suppose that in the speakers' consciousness aș + infinitive was still connected to the simple past (and this is why it could be used as a narrative tense in the old translations).If it were so, we would no longer speak of vrea + infinitive sequence in terms of a residue or a reconstruction, but in terms of a paradigm (a vrea in the past + infinitive) which is only related to ar + infinitive.
In the 16 th century, the simple past forms of a vrea were (Zamfir, 2007, p. 94): vrui, vruși, vru, vrum, vrut, vrură In order to obtain the result attested in the 16 th century as the paradigm of the conditional auxiliary, we need to assume an analogic ending for the 1st person sg., a form vruș(i), homonymous to the current 2sg form, as well as an extension of the ending ră from 3pl to 3sg.Both extensions rely on syncretisms existing in various areas of the Romanian verbal system (1"2sg, 3sg"pl).Hence, at the origin of the current forms of the auxiliary and of the variants in Istro-Romanian and the dialect of Banat may lay the contamination of the paradigm of the imperfect with the one for the simple past in 1sg, 3sg and 3pl: The contamination between the forms of the simple past and of the imperfect also occurred in some Italian dialects; the facts are mentioned by Meyer-Lübke (1895, p. 402) 28 and described by Rohlfs (1968, p. 344-345).
The variants in which it is just the initial v-which gets removed correspond to the forms existing in the inversions in old Romanian as well as in those retained in Istro-Romanian and the dialect of Banat.One example dating from the 16 th century (and extensively discussed by Zamfir, 2007, p. 333, who considers it "a strange form") may be interpreted as an instance of retention of a longer form of the auxiliary in the context of the homonymy between the initial dropped consonant and the form of the pronominal clitic 29 : (31) de-ați fi fost den lume fiind, lumea if=aux.2plbe.inf be.pple of world be.gerworld.def 28"Dans les patois du Nord comme du Sud, les uns employaient jadis, les autres connaissent encore une flexion mélangée, c-à-d.qu'ils ont fait choix de l'imparf.pour les 1 re et 3 e p. sing.e 3 e plur., du parfait pour les autres formes: de là -ia, -esti, -ia, -emmo, -este, -iano" [In the Northern patois as well as in the Southern ones, some of them used a long time ago, while others still evince a mixed inflection, i.e., they selected the imperfect for 1 st and 3 rd person singular and 3 rd person plural, the perfect for the other forms: hence -ia, -esti, -ia, -emmo, -este, -iano] (Meyer-Lübke, 1895, p. 402).Surprisingly, the pattern of the blending of forms is identical with the one we propose for Romanian, but has the inverse distribution: 1sg, 3sg and 3pl person forms would come from the perfect. 29The editors transcribe the sequence vreară (the image is reproduced in Appendix A), but the context clearly requires the identification of the initial v-as belonging to the 2pl clitic.In the evangelical translation from ct.1560-1, the conditional construction includes the pronominal clitic and the conditional periphrasis without the participle (the present conditional), with reversed auxiliary: "De ați fi fostŭ den lume, lumea amu a sa iubi-v-ară" [If you had been from this world, this world would love you as its own] (ct.1560-1, 221 v ).The form occurs twice: in the evangelic text (which contains, as known, an older level of language than in the homiletic commentary) and in its fragmentary resumption, in explanations.Identical repetition renders this isolated attestation a little less contestable such that it may not be reduced to a mere writing mistake.

Conclusions
In order to account for the formation of the Romanian conditional, we propose the following path: similarly to the grammaticalization of the simple future formed with the aid of the infinitive and the present of the verb a vrea, another grammaticalization process took place involving a periphrasis of the future in the past, built with the infinitive together with a mixed paradigm obtained by way of contamination between the forms of simple past and imperfect of the verb a vrea (vreaș, vreai, vreară, vream, vreați, vreară).The future in the past (the association between uncertain posteriority and past) lays, as in many other languages, at the heart of the development of a hypothetical counterfactual meaning.Often used in postposition (during a period of time when the future and the conditional auxiliaries of the Western Romance languages got fixed in this position to the point that they even became inflexions), the auxiliary forms underwent an abbreviation process through phonetic simplification and as a consequence of the reinterpretation of the segment re-as belonging to the infinitive.The short forms aș, ai, ară, am, ați, ară partially lost their connection with the verb a vrea, which accelerated their specialization as markers of the conditional 30 .However, they still preserved a temporal meaning for a while (in the 16 th century), sometimes occurring in the contexts of the simple past as future in the past.The hypothetical value split in a hypothetical about present and future (aș face 'I would do') and a hypothetical referring to the past, due to the emergence of a periphrasis with the participle: aș face 'I would do' vs. aș fi făcut 'I would have done' .
Other similar constructions containing the verb a vrea in the imperfect and the compound past have had a slower pace of grammaticalization (or the process itself started at a later time).In the 16 th century the periphrasis built with the imperfect of the verb a vrea (vrea, vreai, vrea, vream, vreați, vrea + infinitive) is especially widespread.Its main values are those of future in the past and of hypothesis referring to the present.Besides this, a periphrasis with a doubly-marked temporality, specialized for the counterfactual past hypothetical (vrea fi făcut 'would have done') was created.The variants for the future in the past and the conditional are retained: aș face as a standard form, reaș face in Istro-Romanian, the dialect of Banat and in other areas (in isolation), vrea face in some Northern patois.
The explanation resting on the past forms of the verb *volere/vrea is the only one that can lend coherence to the data, due to the fact that it integrates them within a unitary and visible grammaticalization process manifested in many other languages.The semantic objections which have been raised against it are irrelevant because they do not take into account the typical formation process of a mood expressing the hypothetical condition.In this process, the passage from the temporal-modal value to the modal oneof counterfactual and non-counterfactual hypothetical meaning-is normal and frequently attested.The retention of some residues from the old temporal value or the use of these forms in hypothetical contexts 30 One may draw an analogy with the future of the type voi + infinitive: this future starts alternating with the syntactically identical periphrasis with an auxiliary undergoing apheresis: oi + infinitive.The modal and evidential values ("the presumptive mood") occur in both forms but the form oi + infinitive gets specialized for evidentiality in time.A similar process may have taken place with the periphrasis vreaș + infinitive, at the time when it started co-varying with the periphrasis undergoing apheresis *(e)aș + infinitive: the latter, given it obscured etymological connections, entered an accelerated process of specialization for the conditional value.Both forms continued to circulate, however. of present vs. past do not contradict their relatedness, because they belong to the same process, at different stages.The objections regarding the form may be rejected by presupposing a contamination between the paradigm of the simple past and of the imperfect.The periphrases with a vrea in the language of the first texts and in Aromanian confirm the productivity of the pattern, and the forms in the dialect of Banat and in Istro-Romanian are simply variants of a common grammaticalization process with internal variations and different rhythms.Some of these variants were abandoned in time on account of the equivalence and the competition between them.Națională, București, 1930. ct.1560-1 = Coresi, Tetraevanghelul tipărit de Coresi, Brașov 1560-1561, comparat cu Evangheliarul lui Radu de la Mănicești. 1574, ed. Fl. Dimitrescu, Editura Academiei, București, 1963. cv.1563-83 = Codicele Voronețean, ed. M. Costinescu, Editura Academiei, București, 1981. Diaconovici Loga.1822= C. Diaconovici Loga, Gramatica românească (Buda, 1822) de învățătură, 1646, ed. A. Rădulescu, Editura Academiei, București, 1961. Prav.1652= Îndreptarea legii, 1652, ed. A. Rădulescu, Editura Academiei, București, 1962. sa.1683= Ioan Zoba din Vinț, Sicriul de aur, ed. A. Goția, Editura Minerva, București, 1984.