RRL, LXIV, 2, p. 149–162, Bucureşti, 2019
ION GIURGEA
1, MARIA AURELIA COTFAS
2Abstract: We analyze control by the implicit agent of passives in Romanian.
We show that, although some OC verb classes do not have OC in Romanian, there is evidence for control at least in the case of aspectual verbs taking infinitives or subjunctives. Control is only possible in Romanian with se-passives, because copular passives do not allow clausal subjects. We show that in control configurations, Romanian has the peculiarity that se must be repeated on the embedded verb. This cannot be an instance of voice agreement under restructuring, because the matrix verb does not agree with the theme and the embedded clause has a more developed structure (it can be a subjunctive). We explain the data by assuming control via matching between the external argument of the matrix and the embedded se-verbs, which are generated in an argumental position, SpecVoice.
Keywords: control, passive se, infinitive, subjunctive, Agree, Romanian.
1. INTRODUCTION
The aim of our paper is to discuss implicit control in Romanian, that is, instances where the implicit (external) argument of a passivized matrix verb controls into an infinitival or subjunctive embedded clause. The data in Romanian is challenging for several reasons: i) copular passives do not easily accept clausal subjects, therefore se-passives must be taken into account; ii) se-passives are incompatible with (active) control complements but iii) the constructions become grammatical once se is replicated on the embedded predicate. Hence the ‘double passive’ label we will be using throughout. The paper is organized as follows: Section 2 introduces implicit control cross-linguistically, as discussed by Landau (2015) and later on challenged by Pitteroff and Schäfer (2019). Section 3 moves on to discuss implicit predicative control in Romanian, with the analysis of our double passive control constructions in Section 4. A brief Section 5 gives supporting evidence for the existence of a null external argument in se-passives. We draw the main conclusions in Section 6.
1 The “Iorgu Iordan – Alexandru Rosetti” Institute of Linguistics of the Romanian Academy, Bucharest, [email protected].
2The University of Bucharest, [email protected].
2. STARTING POINT: IMPLICIT CONTROL ACROSS LANGUAGES
What ‘implicit control’ is taken to refer to are situations of the type in (1a), where the implicit argument of the (passivized) main clause predicate (here, a subject control verb like promise) controls the PRO subject of the infinitival argument. Importantly, this is only possible when the passive does not have a non-expletive subject, as can be seen in the ungrammatical (1b), featuring a nominal (derived) subject (Mary). To account for the ungrammaticality of (1b), it has been argued that grammar forces the derived subject to control PRO, which is semantically deviant (Visser 1973, van Urk 2013).
(1) a. It was promised (IAi/by Peteri) [PROi to do the shopping]
b. * Marywas promised [PRO to do the shopping]
Discussing implicit control cross-linguistically, Landau (2015) formulates a further restriction concerning the type of predicates that either allow or disallow it, claiming that implicit control is allowed with attitude verbs (which in his analysis yield logophoric control, corresponding to Partial Control in Landau’s previous work), but not with non- attitude predicates (which instantiate predicative control or Exhaustive Control).
According to Landau (2015), attitude verbs and non-attitude verbs differ in their instantiation of obligatory control relations. While with the latter we have a simple predication relation between the controller DP and a property-denoting FinP (the infinitival clause, where PRO has moved from Spec,TP to Spec,FinP), with the former there are two interpretational tiers: first, a predication relation between a variable (pro) in the Spec,CP of the proposition-denoting infinitival clause and FinP, and then variable binding between the controller and pro. Landau (2015) thus formulates the generalization in (2a), which he derives from a more general restriction concerning predication relations (2b). By consequence, predicative control (non-attitude) verbs disallow implicit control because the argument predicated of is not syntactically represented. Conversely, logophoric control (attitude) verbs do allow implicit control of the type in (1a) due to the presence of pro in Spec,CP.
(2) a. Landau’s Generalization: Predicative Control cannot be exerted by an implicit argument/controller (vs Logophoric control)
b. Condition on Syntactic Predication: The argument predicated of must be syntactically present3
More recently, Pitteroff and Schäfer (2019) have challenged the generalization in (2a), testing it against several languages. What they reveal is that while it is true that attitude verbs in all tested languages allow control by an implicit argument in passive configurations (3), it is not true that all languages disallow implicit control with non- attitude verbs. That is to say, some languages are shown to actually allow predicative control to be exerted by implicit arguments, against (2a). It seems, therefore, that languages
3 Supporting evidence comes from data involving secondary predicates (which cannot modify implicit objects) (i), as well as the implicit subject of passive constructions (ii):
(i) John ate *(the meat) raw.
(ii) The room was left (*angry)
like English, French, Russian or Hebrew abide by Landau’s Generalization (4), whereas languages like German, Norwegian, Icelandic or Dutch do not, cf. (5) below.
(3) a. It was decided [PRO to leave the country at once]. (En.) b. Il a été décidé de quitter le pays immédiatement. (Fr.) it has been decided to leave the country immediately
c. Bylo zaplanirovano /obeščano obnovit´ zdanie (Ru.) was.NSG planned.NSG /promissed.NSG renovate.INF building
d. Es wurde beschlossen, das Land zu verlassen (Ge.) it was decided the land to leave
(4) a. *It was managed/tried/dared/stopped to raise taxes (En.) b. *Il a été commencé à augmenter les impôts (Fr.) it has been started to raise the taxes
c.*Bylo načato tratit´ den´gi na bezpoleznje lekarstva (Ru.)
was begun.NSG spend.INF money on useless medicines
(5) a. Først da ble det stoppet å røyke (Nor.)
only then was it stopped to smoke ‘Only then people stopped smoking’
b. Die Liste wurde öffentlich ausgehängt und the list was openly posted and
es wurde begonnen, sie zu arbeiten (Ge.)
it was started it to work-off
‘The list was posted and people began to work it off.’
c. Er wordt geprobeerd (om) de deur open te maken (Dutch) there was tried for the door open to make
d. það var byriað að moka snjóinn (Icelandic) it was begun to shovel snow-the
To account for the different behaviour of non-attitude verbs in the two sets of languages illustrated in (4) and (5), the authors propose that English-type languages (4) do not allow passives with predicative control verbs because they do not have ‘truly impersonal’ passives: in (4), the expletive is a subject pronoun which needs a CP complement to associate with. Complements of predicative control verbs are smaller projections (FinPs only), with which the expletive cannot associate, therefore these predicates cannot passivize, accounting for the ungrammaticality of structures of the type
*It was begun. Conversely, German-type languages (5) do have ‘truly impersonal’ passives.
This is supported by the fact that unergative verbs in these languages can freely passivize (cf. (6a, b)), which shows that their expletives do not need to associate with a clause.
Hence, predicative implicit control is possible.
(6) a. Es wurde getanzt (Ger.) it was danced
b. I går ble det danset (Nor.) yesterday was it danced
c. * It was danced. (En.) d. *Il a été dansé. (Fr.)
3. THE PROBLEM: IMPLICIT CONTROL IN ROMANIAN
Romanian not being part of the range of languages analyzed by Pitteroff and Schäfer (2019), it would be interesting to see whether it abides by Landau’s generalization in (2a) above or whether, on a par with German-type languages illustrated under (5), it does not preclude implicit predicative control.
One important observation to start with is that in Romanian copular passives are not standardly used with clausal subjects: only se passives can occur in this configuration (see (7))4. Therefore, Pitteroff and Schäfer’s (2019) claims must be tested with se-passives in Romanian.
(7) a.?? A fost decis [ca toată lumea să vină la opt]
has been decided that everybody SBJV come at 8 o’clock b.* A fost promis [că toţi vor veni / să vină toţi]
has been promised that all will come/ SBJV come everybody a´. S-a decis [ca toată lumea să vină la opt]
REFL-has decided that everybody SBJV come at 8 o’clock
b´. S-a promis [că toată lumea va veni / să vină toată lumea]
REFL-has promised that everybody will come/ SBJV come everybody
As noticed already by Dobrovie-Sorin (1998), se-passives disallow control infinitives (8). The author includes this among the arguments against the existence of a ‘nominative’
se in Romanian:
(8) a.*S-a promis [a merge la cumpărături.]
REFL-has promised to go to shopping b. *S-a început [a curăţa străzile]
REFL-has begun to shovel streets-the
Be that as it may, a very interesting observation that can be made about examples of the type illustrated in (8) is that they become acceptable once passive/impersonal se is replicated on the embedded verb, as shown in (9). Thus, while the examples under (8) are ungrammatical, the ones with a replicated se in the infinitival clause (i.e., our ‘double passive’ constructions) become grammatical.
4 Examples of clausal subjects with copular passives can be found on the Internet. It is not always clear whether they are due to the influence of English (many texts are translated; authomatic translations can also be a reason) or reflect a different grammar. In the present paper, we are interested in the standard variety, which disallows such examples.
Among the few examples of copular passives with infinitive clauses we could find on Google, the active construction, predicted by Schäfer and Piteroff’s generalization, was attested (for a fost decis ‘it was decided’ we found 4 ex. of active infinitives, 2 ex. of se-passive infinitives and 2 ex. of double passives):
(i) În urma discuţiilor a fost decis a focaliza intervenţiile în … following discussions-the has been decided to focus interventions-the in (aids.md/aids/files/693/minutes-gfatm-round-x-23-june-2010-ro.doc)
No examples of clausal complements (of any sort) with a fost decis ‘it was decided’, a fost plănuit ‘it was planned’ or a fost promis ‘it was promised’ have been found in the Corola corpus.
(9) a. S-a promis [a se ţine seama de toate doleanţele]
REFL-has promised to REFL take account-the of all grievances ‘It was promised to take into account all the grievances’
b. S-a încercat [a se mări /să se mărească salariile]
REFL-has tried to REFL raise/SBJV REFL raise.3 salaries-the
One might argue that, due to the predominance of the subjunctive with such verbs in Modern Romanian, the phenomenon of control is weakened even with infinitives, so that (9) might not even involve control. Cotfas (2012) has shown that even implicative verbs, which involve obligatory control in other languages, may accept disjoint subjects in Romanian and thus obviate (obligatory) control readings.
However, we find the “double se” pattern (se-matrix V... se-embedded V) even with aspectual verbs, for which disjoint subjects are clearly impossible also in Romanian – irrespective of whether the complement is infinitive or subjunctive:
(10) Atunci s-a început [a se dilua laptele cu apa]
then REFL-has started to REFL dilute milk-the with water-the
(http://informatiicenzurate.ro/2015/01/24/alimentatia-naturala-in-copilarie/) (11) S-a început [să se discute despre asta]
REFL-has started SBJVREFL discuss.3 about this
(http://inliniedreapta.net/monitorul-neoficial/...)
Another observation is that there is a clear preference for se-passives (to the detriment of copular passives) in the embedded clause in such environments: although examples such as the one under (12) can be found, they are extremely rare:
(12) S-a început [a fi împărţită pe hălci]
REFL-has begun to be divided.FSG in pieces
‘(The factory) started to be divided in pieces’ (http://confluente.org/)
The scarcity of such examples is confirmed by the data in the available corpora. In the Corola corpus, we have searched for both începe + subjunctive, as well as începe + infinitive contexts – in the relevant distribution, i.e., (i) REFL.ACC + începe + să and (ii) REFL.ACC + începe + a. As far as (i) is concerned, we have found 18 examples in which the subjunctive has passive/impersonal se, but no example of copular embedded passives (no results for the query „se [drukola/base=începe] să fie”). As for the (ii) distribution, involving infinitive complements, 12 examples feature an embedded se and only one an embedded copular passive (which is actually the example quoted in (12)). In 5 examples the infinitive is active, on the pattern of (8) above (ungrammatical for us).
The same type of search was performed for Google. For (i) începe + subjunctive, we found 38 examples with embedded se (transitive verbs) and only 1 example of embedded copular passives of transitive verbs. There were also 44 examples of embedded intransitives, all featuring embedded se. As for (ii) începe + infinitive, there were 50 examples with an embedded se-passive (with transitive verbs) versus just 6 with copular embedded passives of transitive verbs. With intransitive predicates, we were able to find 26 relevant examples, all with embedded se. Other aspectual verbs behave similarly (continua
‘continue’, termina ‘finish’, înceta ‘cease’), insofar as they allow se-passives (in order to allow se-passives, they must have persons as external arguments and not contain another type of se – reflexive, anticausative – which excludes se apuca ‘start’, se opri ‘stop’).
Moreover, double se occurs if the embedded clause is an infinitive or subjunctive. With the supine, which is allowed after completion verbs, there is a different construction (voice restructuring, see section 4 below, ex. (24)).
Other verbs with obligatory control are ‘know’, ‘learn’ and ability ‘can’ (see Wurmbrand 1999 for arguments that deontic modals are not control verbs). Here are attested examples of double se with şti ‘know’ and învăţa ‘learn’:
(13) a. Încă o dovadă a manierei în care se ştie să se facă comerţ still a proof GEN manner-the.GEN in which REFL knows SBJV REFL do.3 trade în România. (www.ziare.com › Life Show › Magazin › obiceiuri)
in Romania
‘Another proof of the way in which people know how to trade in Romania’
b. Se învaţă să se asocieze două sau mai multe răspunsuri verbale.
REFL learns SBJV REFL associate.3 two or more answers verbal ‘One learns (how) to associate two or more verbal answers.’
(https://books.google.ro/books?isbn=973166291X)
With ‘can’, a potential example is (14); note that, besides the double se pattern, putea also allows a bare infinitive5; this is a restructuring configuration in which clitics, including se, are disallowed. We may assume that the restructuring construction involves raising:
(14) Nu se mai putea {să se înainteze / înainta } din cauza zăpezii not REFL more can.IMPF.3 SBJV REFL advance.3 / advance because snow-the.GEN
‘One couldn’t go further because of the snow.’
Since the aforementioned (aspectuals and ability) verbs disallow disjoint subjects, one must assume either a raising or a control account. Some analyses, which focus on subjunctive complements, follow a raising account (Alboiu 2007, Cotfas 2012) – actually, as nominative is licensed post-verbally in Romanian, there is no obligatory syntactic raising; by ‘raising’ we understand downwards Agree. However, raising is impossible with passive se (Dobrovie-Sorin 1998):
(15) *Nu se este niciodatǎ mulţumit / mulţumiţi. (Ro.) not SE is never satisfied.MSG/MPL
In the “double-se” configuration with aspectuals and ability verbs, the fact that raising is not involved is shown by the absence of agreement with the embedded Theme (in the following, we give examples with începe ‘begin’; the other aspectual and ability verbs taking subjunctives and a-infinitives behave the same):
5 Examples with se-passive putea are hard to identify because se putea may also represent an impersonal modal ‘be possible, be likely’.
(16) a. (*S-) au început să se discute aceste chestiuni la televizor.
(*REFL)-have.3PL begun SBJV REFL discuss these issues on TV b. (*S-) au început să fie discutate aceste chestiuni la televizor.
(REFL-)have.3PL begun SBJV be discussed these issues on TV
Consequently, our examples in (10)-(11) must rely on control. More convincing examples can be found in (17) and (18), all attested examples, where we have double passive (se) marking (on both matrix and embedded predicates), but there is no agreement in number and person between the matrix aspectual (which appears in the singular) and the embedded plural Theme argument:
(17) În 1994 s -a început să se efectueze lucrări de in 1994 REFL-has begun SBJV REFL carry-out.3 works of stabilizare şi restaurare a întregului monument.
stabilization and restoration GEN whole-the.GEN monument
‘In 1994 works of consolidation and restoration of the whole
monument were initiated’ (Corola-website/Science/308053_a_309382) (18) epocă fixă dela care se începe a se socoti anii
epoch fixed from which REFL begins to REFL count years-the
‘fixed epoch from which the counting of the years starts’
(Şăineanu, Dicţionar Universal, 6th edition, 1929, s.v. eră)
A quick internet search (on Google) has revealed no example of double se-passive constructions with plural a începe + subjunctive (s-au început să se). There are some examples with the infinitive in this distribution (s-au început a se), but these can be ruled out, since they are from Old Romanian, where au (which is a plural form in Modern Romanian) was number-neutral.
As for contexts with singular marking on the aspectual, both with subjunctive and infinitive complements, a subset of them do involve a plural embedded subject. More to the point, out of 25 examples of the type s-a început să se, 7 have a plural Theme. In the same vein, out of 42 examples of the type s-a început a se, 7 have a plural Theme (as also shown in (17), (18) above).
Taking into account the data discussed above, three questions may be raised. First of all, why is an active embedded verb impossible in our examples? (i.e., why does the se need to be replicated on the embedded predicate?, cf. * S-a început [a discuta]). Secondly, why is there a clear preference, in our ‘double passive constructions’, for an embedded se- passive rather than the copular passive? (see the discussion under (12) above, showing how rare this type is). Third, if the embedded verb is passive and the matrix verb is not a raising verb (see the lack of (plural) agreement and the se-marking on the matrix aspectual in (17)- (18)), how is the Theme of the embedded verb licensed, in the case of infinitives?
4. ANALYSIS
At first sight, the Romanian facts might lead one to think of the double passives encountered in control configurations in other languages ((19)-(21)) (Florian Schäfer, p.c.).
Note however that these examples involve personal passives in the matrix, which is not acceptable in Romanian (see (22)):
(19) el producto fue empezado a ser utilizado como conservante the product was begun.MSG to be used as preservative
(Spanish; Bosque and Gallego 2011)
(20) Slike ting forsøkes ofte å gjøres (Norwegian, Lødrup 2014) such things try.PRES.PASS often to do.INF.PASS
‘One often tries to do such things.’
(21) Pära tafan-ma-chägi ma-na’fanätuk ni lalahi siha
FUT 1PL.IR.IN-PASS-try NPL.RL.IN.PASS-hide OBL men PL
‘The men will try to hide all of us’ (Chamorro; Wurmbrand and Shimamura 2017) (22) a.?? Asemenea lucruri se încep să se facă /a se
such things REFL begin.3PL SBJV REFL do.3/ to REFL face tot mai des.
do ever more often
b.*Asemenea lucruri sunt începute greu a se face / such things are begun.FPL hard to REFL do / a fi făcute / să se facă... / să fie făcute.
to be done/ SBJV REFL do.3 / SBJV be.3 done
For the double passive construction in (19)-(21), Wurmbrand and Shimamura (2017) propose voice agreement, as a variant of the Voice restructuring configuration assumed to underlie long passives (long object movement). Voice restructuring refers to a configuration where there is passive marking on the matrix predicate only and the embedded verb has a Voice head with an unvalued voice feature. As such, there is no accusative case assignment in the embedded clause, and no embedded external argument (i.e., no PRO).
For double passives, the authors claim that the embedded verb has a Voice head with unvalued voice features, valued by upward (Reverse) Agree with the matrix Voice.
(23) A feature F: __ on α is valued by a feature F: val on β, iff i. β c-commands α AND
ii. α is accessible to β. [accessible: not spelled-out]
iii. α does not value {a feature of β}/{a feature F of β}. (Wurmbrand 2014) For Romanian, however, no restructuring (long passive) can be claimed in the relevant contexts, since restructuring predicts number agreement on the matrix verb. The absence of this number agreement indicates that the embedded theme is licensed in the embedded clause, which is not the case in the instances of voice agreement discussed in Wurmbrand and Shimamura (2017), who conclude that the embedded Voice has no case- licensing property. In our Romanian data, it is true that the theme is not licensed by the embedded Voice, but it is licensed in the embedded clause, by T, since it is nominative and the clause may show subject agreement (when it is a subjunctive, see (11), (13), (14)). In Voice restructuring configurations, including voice agreement, the embedded clause never represents a case-licensing domain (correlatively, there is no embedded T in Wurmbrand and Shimamura (2017)’s analysis). Therefore, voice agreement cannot be adopted for the Romanian double-se construction.
Romanian does actually have a long passive restructuring construction, but with a more reduced embedded clause, the supine, which does not allow clitics, negation or voice marking; this construction is possible termina ‘finish’:
(24) Voturile s-au terminat de numărat votes-the REFL-have(3PL)finished of count.SUP
Carmen Dobrovie-Sorin (p.c.) suggested that the Romanian constructions under analysis (double se-passives) actually involve a subject clause and hence no control. If this were the case, then (25b) would have the structure of (25a)(25), which has a DP subject:
(25) a. S-a început plantarea grâului REFL-has started planting-the wheat-the.GEN
b. S-a început [să se planteze / a se planta grâul]
REFL-has started SBJV REFL plant.3 / to REFL plant wheat-the
However, this does not predict a difference in acceptability between an embedded se-verb and an embedded copular passive – (25c) is predicted to have the same status as (25b), which is not the case (see the corpus data in section 3):
(25) c.?? S-a început [să fie plantat grâul / a fi plantat grâul].
REFL-has begun SBJV be.3 planted wheat-the /to be planted wheat-the
Note that (25a) also differs from (25b) in that it allows a participial (i.e., copular) passive (see (25a´) below), but this may be explained by assuming that passive participles require a Theme with ϕ-features.
(25) a´. A fost începută plantarea grâului has been begun planting-the wheat-the.GEN
‘Wheat planting was begun’
Capitalizing on all of the above, we would like to advance a new proposal, which we claim better captures the behaviour of the Romania data.
Having ruled out raising, voice restructuring, as well as a subject clause analysis, we conclude that we are in the presence of control constructions. This means that the embedded se-passive must project an argument position that can be controlled. We propose that se-passives (both the one in the matrix and the one in the embedded predicate) project an (arbitrary) null external argument in SpecVoice/SpecvP (cf. Giurgea 2015, 2016, 2019, MacDonald and Maddox 2018; see section 5), and the embedded null subject agrees in features with the matrix one6:
(26) [PRO+3 Arb [se [începe [FinP să/a T ...[PRO+3 Arb. [se ..DP]]]
Control
6 We continue to refer to se-impersonal as “se-passives” because the verb shows agreement with the theme (on further restrictions on the theme of se-passives, see Cornilescu 1998, Giurgea 2019):
(i) S-au adus cadouri.
REFL-have.3PL brought presents
But then, why is se obligatory on the embedded verb? Why can’t the null external argument control the PRO subject of an active verb, as actually happens with impersonal se in Italian and Spanish, where the counterpart of (8)b is grammatical (see (27))?
(27) Si è cominciato a pulire le strade (It.)
REFL is started to clean the streets
‘They started to clean the streets.’
(8) b. *S-a început [a curăţa străzile] / [să cureţe străzile] (Ro.)
REFL-has begun to shovel streets-the SBJV clean.3 streets-the
Our solution builds on another property by which Romanian differs from the other Romance languages, namely, the fact that the subjunctive has gradually been replacing the infinitive in control environments, to the point that nowadays, the infinitive is rather bookish in complement clauses (except for the bare infinitive with putea, which is not part of the double-se construction we discuss). We propose that, whereas in Italian and similar languages controlled PRO has unvalued person-number features, which are valued by Agree with the controller (see Landau 2000, 2004, 2015 for an Agree-based analysis of control), in Romanian control involves matching: the embedded subject is born with ϕ- features, which must match those of the matrix verb.
One might assume that se is the spell-out of a PRO carrying the features {+3 +Arb}
received via Agree from the controller (the matrix PRO, see (26)). However, this doesn’t account for the fact that the embedded PRO must be thematically related to the embedded verb – one would expect (28)a to be as good as (28)b:
(28) a. * S-a început să se fie {atent/atenţi / devreme la şcoală }.
REFL-has begun SBJV REFL be.3 attentive/-PL / early at school
b. S-a început {să se acorde atenţie.../ să se meargă devreme la şcoală}
REFL-has started SBJV REFL pay attention SBJV REFL go.3 early to school ‘People started to pay attention to.../ to go to school early.’
The ungrammaticality of (28)a shows that the {+3 +Arb} PRO can only occur under selection by a special Voice (this is one of Dobrovie-Sorin’s (1998) arguments against analyzing Romanian ‘impersonal se’ as an impersonal pronoun). Therefore, we assume that the {+3 +Arb} features are visible for syntax already when the embedded clause is built, which means that they cannot be unvalued features receiving a value only from the matrix.
This leads to the conclusion that the control relation may also be achieved via matching of two sets of valued features – a mechanism that can also be used for variable binding relations over an arbitrary distance, see the binding of mi ‘my’ inside an island in (29)7: (29) Numai eu mă supăr pe colegii care-mi critică articolele.
only I REFL get-angry.1SG at colleagues-the who-me.DAT criticize articles-the
‘Only I get angry at the colleagues who criticize my articles’ (possible interpretation:
‘The othersi don’t get angry at the colleagues who criticize theiri articles’)
7 The variable binding interpretation of personal pronouns has also been analyzed using unvalued features + Agree (Kratzer’s 2009 ‘minimal pronouns’, also adopted by Landau 2015). We cannot adopt such an account because of the contrast in (28). The data in (29) is also problematic for an Agree-based account.
A consequence of this proposal is that the non-controlled arbitrary subject of active infinitives (e.g. A cânta e o bucurie ‘to sing is a joy’) does not bear the formal features {+3 Arb} – otherwise, by our reasoning, se would have been required. We may assume that non-controlled PRO only bears a D-feature and, in the absence of a definiteness feature, the arbitrary interpretation ensues.
As for the second question in section 3, which concerns the overwhelming preference for embedded se-passives over copular passives, the argument goes as follows: the embedded verb cannot be a copular passive because this is a control configuration, and control requires a projected external argument; this is characteristic of se-passives, as opposed to copular (participial) passives.
Finally, coming to the third issue, i.e., the case of the embedded Theme argument of infinitival complements, let us notice that such arguments can be case-licensed in infinitival complements independently of control, cf. (30):
(30) Sper [a nu se înţelege greşit situaţia]
hope.1SG to not REFL understand wrongly situation-the
‘I hope the situation will not be misunderstood’
(forum.seopedia.ro › Chestiuni generale › Comunitatea Seopedia› Bar, lobby...) Unlike subjects of adjunct and subject infinitives, subjects of complement (object) infinitives are constrained, under conditions that have yet to be determined. Besides themes of se-passives, which are freely available, subjects are allowed if the verb is stative (see (31)), non-agentive unaccusative (see (32)); otherwise, examples become acceptable only if the subject is A´-moved out of the infinitival clause (see (33)):
(31) (..) sperând a fi benefică pentru ea mutarea în capitală hoping to be beneficial for her movement-the in capital-city ‘hoping that moving to the capital will be beneficial for her’
(www.sighet-online.ro) (32) a crea haos sperând a se naşte o nouă ordine
to create chaos hoping to REFL be-borne a new order (www.miscarearatiunii.ro/)
‘to create Chaos hoping that a new order will emerge’
(33) ce speram a da un suflu nou that hope.IMPF.1PL to give a breath fresh
‘... that we hoped will give a fresh breath’ (www.taekwondowtf.ro›National)
We leave this issue for further research.
5. INDEPENDENT EVIDENCE FOR THE PROJECTION OF THE EXTERNAL ARGUMENT (EA) IN A SPEC POSITION IN SE-PASSIVES
In this section, we aim to strengthen the claim that se-passives involve the projection of a (null) external argument in the Specifier projection of their Voice (or v) head.
As shown in Giurgea (2015, 2016, 2019), proof for the existence of this EA comes from the fact that it acts as an intervener for Agree between T and the Theme, which
explains why certain types of Themes are not licensed. The generalization (based on Cornilescu (1998)) is given in (34), alongside a relevant example in (35), which shows that some animate common nouns are allowed, but proper names, personal pronouns or animate definite DPs with possessors are not. The difference between the former and the latter is that the latter (el/Ion/maică-sa ‘he/Ion/his mother’), when used as objects, require differential object marking.
(34) DPs that, when objects, need the object marker pe and allow or require clitic doubling are excluded as Themes of se-passives
(35) S -a adus {prizonierul / *el/ *Ion/ *maică-sa} la judecată
REFL-has brought prisoner-the/ he/Ion / mother-his at judgment
Giurgea’s account, which we adopt here, is based on the intervention account for other types of person-case constraints (see Rezac 2011) and it is schematized in (36):
(36) a. The DPs which take DOM and allow clitic doubling have a Person feature b. The EA of se-passives also has a Person feature (see its obligatory human
interpretation) and is projected in SpecVoice, blocking person agreement, but allowing number agreement
c. DPs must be case-licensed by T via Agree in all their ϕ-features
On account of (36), DPs which take DOM and allow clitic doubling (and which therefore have a [Person] feature) cannot be case-licensed by T in se-passives, this being precluded by the presence of the EA in Spec,VoiceP, which also has a Person feature.
Moreover, MacDonald and Maddox (2018), Giurgea (2019) show that, unlike the EA of copular (participial) passives, the EA of se-passives can control definite inalienable possesses (the purpose clause in (37) confirms that we are dealing with passive se rather than anticausative se):
(37) Aici, pentru a pune o întrebare {se ridică/ # este ridicată} mâna.
here for to put a question SE raises is raised hand-the
‘Here, in order to ask a question, one raises one’s hand.’
If both se-passives and copular passives had a non-projected EA, the contrast in (37) would remain mysterious.
7. CONCLUSIONS
The aim of our paper was to test Landau’s (2015) generalization, as well as the challenges brought to it by Pitteroff and Schäfer (2019) against the Romanian data. Namely, we wanted to see whether implicit predicative control (i.e., with EC verbs) is possible or not in Romanian – since, according to Landau (2015), it should be impossible, but Pitteroff and Schäfer (2019) have shown this to be the case only in some languages, not in others.
Cross-linguistically, implicit control contexts involve a main clause control verb in the passive and an implicit EA in a clausal infinitival complement. For Romanian, due to
the diachronic weakening and even loss of infinitives in complement position, the relevant contexts may involve either infinitival or subjunctive complements. As far as the passive is concerned, Romanian has both copular (fi) as well as impersonal (se-) passives, and it has been independently shown that fi-passives are not felicitous with clausal complements, which are widely compatible with se-passives. Consequently, the type of constructions under analysis can only be tested with se-passives in Romanian. However, se-passives in Romanian are incompatible with active control infinitives. Interestingly, they become licit once the se is replicated on the embedded predicate. This is what we called ‘double passives’ in control configurations.
We have argued that this construction does not represent an instance of voice agreement (a sub-type of voice restructuring), because the matrix verb does not agree with the embedded theme, which is case-licensed in the embedded clause. This shows that the embedded clause has a T-layer (see also the use of subjunctive in free alternation with infinitives), which is incompatible with voice restructuring.
We have thus claimed that our constructions involve a control relation between a null embedded argument (PRO, controlee) and another null matrix argument (PRO, controller), which must match in phi-features. We have proposed that PRO in this case bears the features {+3Person +Arb} and se must appear on an embedded verb because such a PRO can only appear in the Spec of a special Voice headed by passive se. We are thus forced to assume that the controlled PRO in Romanian is not born with unvalued features, but bears features which must match with those of the controller (rather than be valued by them). We correlated this fact with the weakening of complement infinitives in Romanian, which are largely replaced nowadays by subjunctives.
These control configurations provide an additional argument for the projection of the EA in se-passives, which are half-way between bona fide passives and actives. That is, on the one hand, their Theme is nominative and T agrees with it, while se is not a nominative pronoun, but indicates a special Voice head. On the other hand, there is a null EA in an argument position which is involved in control and blocks nominative assignment for [+Person] Themes.
In conclusion, Romanian neither contradicts nor supports Pitteroff and Schäfer’s (2019) claims: if control by an ‘implicit’ argument is understood as control by a non-projected argument, then the generalization cannot be tested in Romanian, given that (i) se-passives do not involve control by a non-projected argument and (ii) in a true passive configuration (i.e. copular passives), clausal complements are disallowed for independent reasons.
REFERENCES
Bosque, I., Á. Gallego, 2011, “Spanish double passives and related structures”, Linguística. Revista de Estudos Linguísticos da Universidade do Porto, 6, 1, 9–50.
Cornilescu, A., 1998, “Remarks on the syntax and the interpretation of Romanian middle passive se sentences”, Revue roumaine de linguistique, 43, 317–342.
Cotfas, M. A., 2012, On the Syntax of the Romanian Subjunctive. Control and Obviation, doctoral dissertation, University of Bucharest.
Dobrovie-Sorin, C., 1998, “Impersonal Se Constructions in Romance and the Passivization of Unergatives”, Linguistic Inquiry, 29, 3, 399–438.
Giurgea, I., 2015, “On the Person Constraint on Romance se-passives”, talk given at the Workshop on Non-local dependencies in the nominal and verbal domain, Centro de Linguística da Universidade Nova de Lisboa, November 13.
Giurgea, I., 2016, “Intervention in Romanian se-passives”, talk given at the 42th Incontro di Grammatica Generativa, University of Salento, Lecce.
Giurgea, I., 2019, “On the Person constraint in Romanian se-passives”, in: L. Franco, M. Marchis Moreno, M. Reeve (eds), Local and Non-Local Dependencies in the Nominal and Verbal Domains, Language Science Press, 109–147.
Kratzer, A., 2009, “Making a Pronoun: Fake Indexicals as Windows into the Properties of Pronouns”, Linguistic Inquiry, 40, 187–237.
Landau, I., 2000, Elements of Control: Structure and Meaning in Infinitival Constructions, Dordrecht, Kluwer Academic Publishers.
Landau, I., 2004, “The Scale of Finiteness and the Calculus of Control”, Natural Language and Linguistic Theory, 22, 811–877.
Landau, I., 2015, A Two-Tiered Theory of Control, Cambridge, Mass., MIT Press.
Lødrup, H., 2014, “Long passives in Norwegian: Evidence for complex predicates”, Nordic Journal of Linguistics, 37, 3, 367–391.
MacDonald, J. E., M. L. Maddox, 2018, “Passive se in Romanian and Spanish. A subject cycle”, Journal of Linguistics, 54, 389–427.
Piteroff, M., F. Schäfer, 2019, “Implicit control cross linguistically”, Language, 95, 1, 136–184.
Wurmbrand, S., 1999, “Modal Verbs Must Be Raising Verbs”, in: S. Bird, A. Carnie, J. Haugen, P. Norquest (eds), WCCFL 18 Proceedings, Somerville, MA, Cascadilla Press, 599–612.
Wurmbrand, S., 2014, “The Merge Condition: A syntactic approach to selection”, in: P. Kosta, L. Schürcks, S. Franks, T. Radev-Bork (eds), Minimalism and Beyond: Radicalizing the interfaces, Amsterdam, Benjamins, 139–177.
Wurmbrand, S., K. Shimamura, 2017, “The features of the voice domain: actives, passives, and restructuring”, in: R. d’Alessandro, I. Franco, Á. Gallego (eds), The verbal domain, Oxford, Oxford University Press, 179–204.
