1 Introduction
In this article, we compare the synchronic status of passive se (Pass
$_{\text{se}}$
) constructions in two Romance languages: Spanish and Romanian, examples of which are provided in (1a) and (1b), respectively.Footnote
[2]
A widely held intuition about Pass
$_{\text{se}}$
is that there is an implied external argument (in contrast to anticausative se (AntiC
$_{\text{se}}$
) constructions; see Haspelmath Reference Haspelmath1990, Mendikoetxea Reference Mendikoetxea and Siewierska2008, Schäfer Reference Schäfer2008, Koontz-Garboden Reference Koontz-Garboden2009, among others). In fact, MacDonald (published online 23 October Reference MacDonald2017) offers empirical support for the projection of an implicit external argument in Spec,Voice in Pass
$_{\text{se}}$
(and the lack thereof in AntiC
$_{\text{se}}$
) from an available inalienable possession interpretation of the sole overt DP when it is a body part, an interpretation only available when a c-commanding possessor is present in the syntax.Footnote
[3]
In this article, we apply the same diagnostic to Romanian, and expand on it. We see that both Romanian and Spanish Pass
$_{\text{se}}$
license an inalienable possession interpretation of the sole overt DP when it is a body part. They differ, however, in that Romanian allows ‘by’-phrases, while Spanish shows severe restrictions. We argue that we can account for the difference in the ‘by’-phrase patterns in terms of the properties of the projected implicit external arguments in Spec,Voice in Romanian and Spanish Pass
$_{\text{se}}$
. Following Legate’s (Reference Legate2014) discussion of implicit arguments (see also Landau Reference Landau2010), we claim that Romanian pro in Spec,Voice of Pass
$_{\text{se}}$
lacks a D(eterminer)-feature (i.e. pro) and Spanish pro in Spec,Voice of Pass
$_{\text{se}}$
has a D-feature 
Furthermore we make a related novel empirical observation about Old Spanish: At an earlier period, Old Spanish Pass
$_{\text{se}}$
allowed ‘by’-phrases at a time when body parts were also licensed in Pass
$_{\text{se}}$
constructions. Old Spanish, we observe, patterns with Modern Romanian, as illustrated in (2), where se heads Voice, following previous approaches, such as Cuervo (Reference Cuervo2003, Reference Cuervo2014), Kempchinsky (Reference Kempchinsky, J. Clancy Clements and Vance2004), Folli & Harley (Reference Folli, Harley, Kempchinsky and Slabakova2005), Basilico (Reference Basilico2010), Ordóñez & Treviño (Reference Ordóñez, Treviño and Ortiz-López2011), Armstrong (Reference Armstrong2013), among others.Footnote
[4]
As we discuss, these observations have consequences for the widely assumed diachronic path of Romance se constructions, where reflexive se (Refl
$_{\text{se}}$
) develops into AntiC
$_{\text{se}}$
, and AntiC
$_{\text{se}}$
into Pass
$_{\text{se}}$
(Geniušienė Reference Geniušienė1987, Haspelmath Reference Haspelmath1990, Cennamo Reference Cennamo1993, Portilla Reference Portilla2007).Footnote
[5]
The first consequence is that Pass
$_{\text{se}}$
needs to be divided into two separate stages. This conclusion comes from the observation that ungrammatical ‘by’-phrases in Modern Spanish Pass
$_{\text{se}}$
were once grammatical in Old Spanish Pass
$_{\text{se}}$
, a pattern already observed for Portuguese (Naro Reference Naro1976), Italian (Cinque Reference Cinque1988, Cennamo Reference Cennamo1993) and French (Heidinger & Schäfer Reference Heidinger, Schäfer, Fagard, Prevost, Combettes and Bertrand2010). Yet, despite these multiple cross-Romance observations, they have remained observations alone.
In this article, we offer an explanation for this pattern in terms of linguistic cycles (Roberts & Roussou Reference Roberts and Roussou2003; van Gelderen Reference Gelderen2004, Reference Gelderen2011 and references therein). We take Modern Romanian and Modern Spanish Pass
$_{\text{se}}$
to illustrate that ‘[d]ifferences between languages arise because they are in different stages of a particular cycle’ (van Gelderen Reference Gelderen2011: 29). We therefore contribute to the idea that linguistic change is cyclic, an idea which, as van Gelderen (Reference Gelderen2011: 3) points out, with the exception of the negative cycle, generative linguists have not seriously examined. Moreover, as we illustrate, the approach to the modified path in terms of a cycle provides a theoretical explanation for why some languages do not develop passive se constructions.
The structure of the article is as follows. Section 2 discusses the properties of Modern Spanish and Romanian Pass
$_{\text{se}}$
constructions. We see that they pattern together with respect to the main diagnostic for a projected null indefinite pro in Spec,Voice adopted from MacDonald (published online 23 October Reference MacDonald2017): the ability to license an inalienable possession interpretation of a body part DP. We also expand on this diagnostic by discussing body parts in adjunct clauses. We also see, nevertheless, that Romanian and Spanish differ with respect to licensing ‘by’-phrases: Romanian Pass
$_{\text{se}}$
licenses ‘by’-phrases, while Spanish Pass
$_{\text{se}}$
shows severe restrictions. In Section 3, we discuss the patterns in Old Spanish. In order to distinguish between Pass
$_{\text{se}}$
constructions and AntiC
$_{\text{se}}$
constructions in the historical data, we recap diagnostics from the literature. Moreover, we contribute to these diagnostics by offering an additional one related to body parts: a body part interpreted as inalienably possessed can only occur in Pass
$_{\text{se}}$
, whether it is the sole overt DP or whether it occurs in an adjunct clause, not in AntiC
$_{\text{se}}$
. We offer evidence that there was a time in the development of Spanish Pass
$_{\text{se}}$
that looked like Modern Romanian Pass
$_{\text{se}}$
, namely when body parts were interpreted as inalienably possessed and when there were no ‘by’-phrase restrictions. In Section 4, we propose that these ‘by’-phrase patterns relate to a difference in the internal properties of the two implicit arguments, following Legate (Reference Legate2014). The implicit projected pro in Spec,Voice in Romanian and Old Spanish Pass
$_{\text{se}}$
lacks a D-feature, while the implicit projected pro in Spec,Voice in Modern Spanish Pass
$_{\text{se}}$
has a D-feature. In Section 5, we offer an account of the development of a D-feature in terms of a subject cycle à la van Gelderen (Reference Gelderen2004, Reference Gelderen2011), and we discuss a prediction generated by the account regarding when a language will not develop a passive se construction. In Section 6, we briefly recap our main findings.
2 Modern Spanish and Romanian se constructions
In this section, we first discuss Spanish se constructions in general, paying special attention to Pass
$_{\text{se}}$
constructions. We then systematically compare and contrast Spanish Pass
$_{\text{se}}$
with Romanian Pass
$_{\text{se}}$
.
2.1 Modern Spanish se constructions
It is well known that Spanish has several se constructions (de Miguel Reference De Miguel Aparicio1992; Mendikoetxea Reference Mendikoetxea1992, Reference Mendikoetxea1999b, Reference Mendikoetxea and Siewierska2008; Sánchez-López Reference Sánchez-López2002 among others) that have been argued to have developed from ‘true’ reflexive se (Monge Reference Monge1954, Geniušienė Reference Geniušienė1987, Cennamo Reference Cennamo1993): AntiC
$_{\text{se}}$
, Pass
$_{\text{se}}$
and Impersonal se (Imp
$_{\text{se}}$
).Footnote
[6]
Examples are provided in (3a), (3b) and (3c), respectively. In (3c),
$a$
is the differential object marker (DOM).
The central difference between AntiC
$_{\text{se}}$
and Pass
$_{\text{se}}$
, on the one hand, and Imp
$_{\text{se}}$
, on the other, is that the sole overt DP in the former behaves like a grammatical subject, while in the latter it behaves like a grammatical object (Mendikoetxea Reference Mendikoetxea1992, Reference Mendikoetxea1999a, Reference Mendikoetxea and Siewierska2008; Sánchez-López Reference Sánchez-López2002 among others).Footnote
[7]
In this article, we focus on the properties of Pass
$_{\text{se}}$
(although see Section 5.2 for a brief dsicussion of Imp
$_{\text{se}}$
). In Section 3.1 we contrast the properties of Pass
$_{\text{se}}$
with AntiC
$_{\text{se}}$
as a way to arm ourselves with diagnostics to determine whether an intransitive se construction is Pass
$_{\text{se}}$
or AntiC
$_{\text{se}}$
in the historical data.
Spanish Pass
$_{\text{se}}$
has the following properties, indicating that the sole overt DP is a grammatical subject (see Mendikoetxea Reference Mendikoetxea1999a, Sánchez-López Reference Sánchez-López2002 and references therein for data and discussion): (i) the sole overt DP controls verbal agreement (4a), (ii) a direct object clitic corresponding to the sole overt DP is ungrammatical (4b), and (iii) the DOM
$a$
, limited to specific human direct objects, is not available (4c).
Additionally, there is an intuition that there is an implicit external argument in Pass
$_{\text{se}}$
constructions, which when agentive can license agent-oriented adverbs and rationale clauses (Mendikoetxea & Battye Reference Mendikoetxea and Battye1990, Mendikoetxea Reference Mendikoetxea1999a, Sánchez-López Reference Sánchez-López2002 among others), as illustrated in (5).
Of course, the implicit argument is not limited to agents, since it can receive a variety of different thematic roles, such as source or experiencer, as illustrated in (6a) and (6b), respectively; data from Mendikoetxea (Reference Mendikoetxea1999b: 1670).Footnote [8]
Example (6a) also illustrates that Pass
$_{\text{se}}$
can be formed with a verb whose non-Pass
$_{\text{se}}$
counterpart is ditransitive.
More recently, the intuition that there is an implicit external argument in Pass
$_{\text{se}}$
in Spanish has found syntactic support from the interpretation of body parts. MacDonald (published online 23 October Reference MacDonald2017) observes that when the sole overt DP in Pass
$_{\text{se}}$
is a body part, the body part can receive an inalienable possession interpretation, an illustration of which is in (7).
Given data like those in (7), MacDonald (published online 23 October Reference MacDonald2017) concludes that there is a syntactically projected implicit external argument in Pass
$_{\text{se}}$
(in contrast to AntiC
$_{\text{se}}$
; see Sections 3.1 and 5 below). Consider how this conclusion is reached.
As previously noted for other Romance languages (Guéron Reference Guéron, Guéron, Obenauer and Pollock1985, Reference Guéron, Everaert, van Riemsdijk, Goedemans and Hollebrandse2006; Nakamoto Reference Nakamoto2010), for a body part to be interpreted as inalienably possessed in Spanish, it must have a c-commanding possessor, as illustrated in (8), where coindexation indicates inalienable possession.
Second, as Guéron (Reference Guéron, Guéron, Obenauer and Pollock1985, Reference Guéron, Everaert, van Riemsdijk, Goedemans and Hollebrandse2006) observes for French, in Spanish, a pragmatically salient possessor cannot license an inalienable possession interpretation of a body part DP. Consider the contrast between (9a) and (9b) in a context where a father is talking to his daughter and answering her question about why she can run so fast.
Example (9a) is felicitous in this context, where the interlocutor (pro ‘you’) is the daughter, because it can express that the legs in question are inalienably possessed by the daughter. In contrast, (9b) is odd, precisely because the legs are not interpreted as inalienably possessed by the daughter.Footnote
[11]
In (9a), pro ‘you’ serves as the syntactically present possessor. In (9b), there is no such syntactically present possessor. A body part is interpreted as inalienably possessed only when the possessor is present in the syntax. Thus, since there is an inalienable possession interpretation of the sole overt DP in Pass
$_{\text{se}}$
in (7), the possessor must be syntactically present.
2.2 Romanian se constructions
Like Spanish, Romanian has AntiC
$_{\text{se}}$
and Pass
$_{\text{se}}$
constructions. However, unlike Spanish, Romanian lacks an impersonal se construction (an indication of which is the inability of dom
pe from appearing in Romanian se constructions, among others, as extensively discussed in Dobrovie-Sorin Reference Dobrovie-Sorin1998, Reference Dobrovie-Sorin, Everaert and van Riemsdijk2006; see (41b) below). Romanian Pass
$_{\text{se}}$
patterns with Spanish Pass
$_{\text{se}}$
in licensing agent-oriented adverbs and rationale clauses, as illustrated in (10). Example (10a) is from Dobrovie-Sorin (Reference Dobrovie-Sorin1998: 426) and (10b) from Cornilescu & Nicolae (Reference Cornilescu, Nicolae, Pană Dindelegan, Zafiu, Dragomirescu, Nicula, Nicolae and Esher2015: 329).
Additionally, as in Spanish, when the sole overt DP in Pass
$_{\text{se}}$
in Romanian is a body part, it can be interpreted as inalienably possessed, as illustrated in (11).
Observe in (12) that Romanian has the same construction as Spanish, where the subject can be interpreted as the possessor of a body part direct object. Data in (12a–b) are from Manoliu-Manea (Reference Manoliu-Manea, Chappell and McGregor1996: 725–726).
Just like in Spanish, the possessor must c-command the body part. In (13) Peter is not understood as opening his eyes, only his brother, as the coindexation indicates.
Importantly, pragmatics does not suffice to license an inalienable possession interpretation of a body part. Thus, in a context where a father answers his daughter’s question about why she can run so fast, (14a) is felicitous, while (14b) is not.
We conclude from this that body parts in Romanian pattern the same as in Spanish: they require a syntactically present c-commanding possessor for an inalienable possession interpretation.
It appears then that with respect to the licensing of agent-oriented adverbs, rationale clauses, and an inalienable possession interpretation of body parts in Pass
$_{\text{se}}$
constructions, Romanian and Spanish pattern the same.Footnote
[12]
Adopting the conclusions from MacDonald (published online 23 October Reference MacDonald2017) regarding Spanish, we assume that for both Spanish and Romanian Pass
$_{\text{se}}$
there is an implicit projected external argument, which we take to be pro in Spec,Voice, as illustrated in (15), where se heads Voice.
Given this conclusion, a question arises in light of one important difference between Romanian and Spanish Pass
$_{\text{se}}$
, related to the licensing of ‘by’-phrases. ‘By’-phrases are grammatical in Romanian Pass
$_{\text{se}}$
(Geniušienė Reference Geniušienė1987, Dragomirescu Reference Dragomirescu and Pană Dindelegan2013, Cornilescu & Nicolae Reference Cornilescu, Nicolae, Pană Dindelegan, Zafiu, Dragomirescu, Nicula, Nicolae and Esher2015), while there are severe restrictions on grammatical ‘by’-phrases in Pass
$_{\text{se}}$
in Spanish (see de Miguel Reference De Miguel Aparicio1992, Mendikoetxea Reference Mendikoetxea1999a, Sánchez-López Reference Sánchez-López2002).
Example (16) illustrates that ‘by’-phrases are grammatical in Romanian Pass
$_{\text{se}}$
; data in (16a) from Geniušienė (Reference Geniušienė1987: 267), and data in (16b) and (16c) from Cornilescu & Nicolae (Reference Cornilescu, Nicolae, Pană Dindelegan, Zafiu, Dragomirescu, Nicula, Nicolae and Esher2015: 323, 321, 327).Footnote
[13]
In Spanish, however, the situation is more complex. The discussion is still open regarding the exact conditions under which ‘by’-phrases are grammatical in Pass
$_{\text{se}}$
constructions in Spanish (although see discussions in Otero Reference Otero, Bordelois, Contreras and Zagona1986, de Miguel Reference De Miguel Aparicio1992, Mendikoetxea Reference Mendikoetxea1999a, Sánchez-López Reference Sánchez-López2002 and references therein). There may be a tendency, nevertheless, that generically interpreted DPs within the ‘by’-phrase, if grammatical at all, are grammatical.Footnote
[14]
What is clear, nevertheless, is that referential DPs are ungrammatical. This is illustrated by the contrasts in (17) and (18), data taken from Sánchez López (Reference Sánchez-López2002: 60), and expanded to include the pronouns ti/mi/ella.
Note also that the same patterns hold in all Pass
$_{\text{se}}$
constructions in Spanish, even when the external argument of the non-Pass
$_{\text{se}}$
is not an agent, illustrated by the experiencer ‘by’-phrase in (19) from Mendikoetxea (Reference Mendikoetxea1999b: 1683–1684), and in (20).
In Section 4, we argue that the variation can be accounted for in terms of different features of projected pro, where pro in Spec,Voice of Romanian lacks a D-feature and pro in Spec,Voice of Spanish has a D-feature. Before turning to the analysis, however, we will see that Old Spanish Pass
$_{\text{se}}$
patterns just like Modern Romanian Pass
$_{\text{se}}$
in licensing an inalienable possession interpretation of a body part at a time when referential ‘by’-phrases were grammatical.
3 Old Spanish
The main goal of this section is to illustrate that at a previous stage in its development, Pass
$_{\text{se}}$
in Spanish patterned with Modern Romanian along the lines just discussed in Section 2.2. We will see that there appear to be no obvious ‘by’-phrase restrictions in Old Spanish; specifically, referential ‘by’-phrases were grammatical, in contrast to Modern Spanish. Moreover, we see strong evidence that Pass
$_{\text{se}}$
also licensed an inalienable possession interpretation of body parts.
Before discussing these data directly, however, it is important to briefly discuss how to differentiate Pass
$_{\text{se}}$
from AntiC
$_{\text{se}}$
, since there are instances of formal ambiguity between the two (see Mendikoetxea Reference Mendikoetxea1999a among others). Thus in Section 3.1, we first mention previously proposed diagnostics for distinguishing Pass
$_{\text{se}}$
from AntiC
$_{\text{se}}$
. We then offer an additional diagnostic, based on interpreting body parts as inalienably possessed, something possible in Pass
$_{\text{se}}$
, but not in AntiC
$_{\text{se}}$
.
3.1 Distinguishing Passse from AntiCse
Anticausatives have an extensive literature (see Haspelmath Reference Haspelmath1990, Levin & Rappaport Hovav Reference Levin and Rappaport Hovav1995, Alexiadou, Anagnostopoulou & Schäfer Reference Alexiadou, Anagnostopoulou, Schäfer and Frascarelli2006, Schäfer Reference Schäfer2008, Koontz-Garboden Reference Koontz-Garboden2009, Cuervo Reference Cuervo2014, and references therein). In this section, we do not exhaustively review all properties of AntiC
$_{\text{se}}$
, only those that will help distinguish between AntiC
$_{\text{se}}$
and Pass
$_{\text{se}}$
, and those that will help us identify Pass
$_{\text{se}}$
in the Old Spanish texts.
First, AntiC
$_{\text{se}}$
is largely limited to transitive change of state verbs, such as abrir ‘open’, cerrar ‘close’, derretir ‘melt’, hundir ‘sink’, ahogar ‘drown’, acostumbrar ‘get used to’, romper ‘break’, etc., which do not require an agentive external argument in their transitive counterpart (Koontz-Garboden Reference Koontz-Garboden2009 and references therein). Additionally, ditransitive verbs do not to appear in the AntiC
$_{\text{se}}$
construction.Footnote
[15]
Second, por-phrases (i.e. ‘by’-phrases) that appear with AntiC
$_{\text{se}}$
are interpreted as Causes, as illustrated in (21) (see Alexiadou et al. Reference Alexiadou, Anagnostopoulou, Schäfer and Frascarelli2006, Schäfer Reference Schäfer2008, Koontz-Garboden Reference Koontz-Garboden2009).Footnote
[16]
Now, given that change of state verbs can also appear in a Pass
$_{\text{se}}$
construction, there are cases of formal ambiguity. This ambiguity, however, can be resolved by the presence of certain adjuncts. The presence of por sí solo ‘by itself’ (Mendikoetxea Reference Mendikoetxea1999a, Schäfer Reference Schäfer2008, Koontz-Garboden Reference Koontz-Garboden2009) or a Cause por-phrase picks out AntiC
$_{\text{se}}$
, while the presence of a rationale/purpose clause picks out Pass
$_{\text{se}}$
(Mendikoetxea Reference Mendikoetxea1999a). Thus, the se in (22a) is AntiC
$_{\text{se}}$
, while the se in (22b) is Pass
$_{\text{se}}$
, examples from Mendikoetxea (Reference Mendikoetxea1999b: 1643).
Consider another contrast recently observed by MacDonald (published online 23 October Reference MacDonald2017), regarding the interpretation of body part DPs in AntiC
$_{\text{se}}$
and Pass
$_{\text{se}}$
in Spanish. In contrast to Pass
$_{\text{se}}$
, when the sole overt DP in AntiC
$_{\text{se}}$
is a body part, it cannot be interpreted as inalienably possessed. It can, however, in the presence of a dative possessor. First consider AntiC
$_{\text{se}}$
constructions with a dative that can be interpreted as an inalienable possessor of a body part, as in (23).
Consider a context where sentence (23) can be stated felicitously. Your child is taking a nap on the couch and you are observing him when his mouth suddenly opens. When your child wakes, you can turn to him and say (23), where an inalienable possession interpretation between the interlocutor – your child – syntactically present as dative te ‘you’ and the mouth obtains. In contrast, (24) is infelicitous (indicated by #) in the same discourse context when there is no dative possessor.
The mouth is not interpreted as belonging to your child. It can be interpreted as a mouth as part of a doll, for instance, but not the mouth inalienably possessed by your child.Footnote [17]
Here we expand both on previous literature and on MacDonald (published online 23 October Reference MacDonald2017), by offering another diagnostic for distinguishing between formally ambiguous instances of AntiC
$_{\text{se}}$
and Pass
$_{\text{se}}$
related to the inalienable possession interpretation of body parts. Consider the data in (25), in which there is a combination of a body part and an adjunct that picks out either Pass
$_{\text{se}}$
or AntiC
$_{\text{se}}$
.
The example in (25a) illustrates that, in the presence of a rationale/purpose clause, a body part in an instrumental adjunct receives an inalienable possession interpretation. Recall that a rationale/purpose clause picks out Pass
$_{\text{se}}$
. In contrast, (25b) illustrates that, in the presence of por sí solo or a Cause por-phrase, a body part in an instrumental adjunct does not receive an inalienable possession interpretation. Recall that por sí solo or a Cause por-phrase pick out AntiC
$_{\text{se}}$
. Thus, body parts are licensed with Pass
$_{\text{se}}$
, but not with AntiC
$_{\text{se}}$
. These patterns are the expected ones.
Now, one might counter that the problem with (25b) does not have to do with inalienable possession, however, but with two semantically conflicting por-phrases: one that indicates that the window was opened by hand and one that indicates that it opened by itself/from the wind. We can control for this potential semantic clash by keeping the body part an argument and using a Cause adjunct PP. First consider (26) in the context of someone explaining what happened to Mary, who is holding her hand.
In (26) the por sí sola-phrase ensures that this is an AntiC
$_{\text{se}}$
construction. As (26a) illustrates, the body part noun mano ‘hand’ in the adjunct is licensed when the dative possessor le ‘her’ is present. As (26b) illustrates, when the dative is not present, the body part is not licensed on an inalienable possession interpretation. There is no semantic clash here. Consider another example in (27) in a context where a doctor is carrying out an experiment with sleeping individuals, administering different medication while they sleep to see what the consequent effects are.
The con-phrase introduces a Cause, ensuring that these se constructions are AntiC
$_{\text{se}}$
constructions. As (27a) illustrates in the presence of the dative les ‘them’, the sole overt body part DP can be interpreted as inalienably possessed. As (27b) illustrates, without dative les ‘them’, the body part is not licensed on an inalienable possession interpretation. It appears, then, that a body part can only receive an inalienable possession interpretation in AntiC
$_{\text{se}}$
in the presence of a dative, while in Pass
$_{\text{se}}$
an inalienable possession interpretation of the body part is licensed without a dative.
Not only does this discusson of the interaction between body parts and PP adjuncts that pick out different se constructions corroborate the conclusions from MacDonald (published online 23 October Reference MacDonald2017), it also allows us to add another test for disambiguating formally indistinguishable Pass
$_{\text{se}}$
and AntiC
$_{\text{se}}$
: the presence of a body part that is interpreted as inalienably possessed, whether it is the sole overt DP or whether it is in an adjunct. Pass
$_{\text{se}}$
licenses this interpretation; AntiC
$_{\text{se}}$
does not. This will prove an important diagnostic in the historical data.Footnote
[18]
3.2 Passse in Old Spanish
In this section we discuss the properties of Pass
$_{\text{se}}$
constructions in Old Spanish. The data were extracted manually from a corpus of texts and via electronic searches in the CORDE. The historical period coverd by the corpus is 1207 to 1550.Footnote
[19]
The texts that were used come from a variety of genres and registers.Footnote
[20]
We begin with body parts in Section 3.2.1 and then turn to ‘by’-phrases in Section 3.2.2.
3.2.1 Body parts in Old Spanish
In this section, we first illustrate that inalienably possessed body parts appear to have the same grammatical distribution in Old Spanish as in Modern Spanish, namely, that they required a syntactically present possessor. We then turn to their distribution in se constructions. We see occurrences of body parts with a dative ocurring with change of state verbs, much like we see in Modern Spanish. We also see instances of body parts both as the sole overt DP and in adjunct PPs in se constructions with change of state verbs. Given the discussion in the previous section, we infer that these are most likely instances of Pass
$_{\text{se}}$
. More importantly, we see cases where a body part appears in a non-change of state verb in an adjunct clause, which we conclude must be cases of Pass
$_{\text{se}}$
. We take these findings as evidence for the projection of an implicit external argument in Old Spanish Pass
$_{\text{se}}$
following the conclusions from Section 3.1 above. (The sources of the examples are listed in full towards the end of the article.)
First, observe cases where the subject is intepreted as the inalienable possessor of a body part direct object in (28).Footnote [21]
Of course, an inalienable possession interpretation is not limited to instances where the possessor and the body part are coarguments of the same predicates, as shown in (29).
In each case in (29), importantly, there is a syntactic possessor present to license the inalienable possession interpretation of the body part. In (29a) it is mançebos ‘boys’. In (29b) it is the null subject, which refers back to María, introduced in previous discourse (not shown here). In (29c) it is the first person singular null subject, and in (29d) it is homne ‘man’.
As we saw for modern Spanish, datives in Old Spanish can also license an inalienable possession interpretation of a body part, as illustrated in (30).
We also see datives serving as the possessor of a body part with change of state verbs with se, as in (31).
In addition to the change of state verbs in (31) where a dative appears to license the inalienable possession interpretation of a body part, there are instances of change of state verbs with se, without a dative clitic, and in which the sole overt DP is a body part, as illustrated in (32).
There are also instances of change of state verbs with se in which a body part appears in an adjunct and is interpreted as inalienably possessed, as illustrated in (33).
While the data in (32) and (33) could be taken as counterexamples to the generalization that body parts cannot appear with AntiC
$_{\text{se}}$
constructions in Old Spanish, we must recall, first, that a change of state verb with se could still be a Pass
$_{\text{se}}$
construction. Second, as discussed in Section 3.1, we concluded that a body part interpreted as inalienably possessed with a change of state verb with se was a diagnostic for distinguishing AntiC
$_{\text{se}}$
and Pass
$_{\text{se}}$
. Concretely, a body part can be interpreted as inalienably possessed only with Pass
$_{\text{se}}$
. Thus, it is possible that, in fact, the examples in (32) and (33) above are examples of Pass
$_{\text{se}}$
.
Now consider non-change of state verbs with se in (34).
In (34a), the adjunct clause de la boca ‘from the mouth’ indicates the source from which the words and reasons are thrown. The context indicates that this is in reference to talking. In (34b), the body parts clearly refer to parts of a human, since the discussion is about how to build a home. Finally, in (34c), a small book is described as enclosed within a hand, clearly the hand that belongs to the person who just pulled it out.
In our estimation, there is evidence that body parts are licensed in Pass
$_{\text{se}}$
constructions under the same conditions as in Modern Spanish. The examples of Pass
$_{\text{se}}$
and body parts that we have discussed span from 1252 to 1500. As we will see in Section 3.2.2, during the same period referential ‘by’-phrases were grammatical in Pass
$_{\text{se}}$
constructions.
3.2.2 ‘By’-phrases in Old Spanish
Both por and de were prepositions of agency in Old Spanish. Por overtakes de during the modern period, after the 16th century (Mendeloff Reference Mendeloff1964: 278). Por was also used with Causes. The Cause interpretation is illustrated below in (35) with the change of state verbs with se; data taken from Monge (Reference Monge1954: 44).
We also find por-phrases with non-change of state verbs, illustrated in (36).
Importantly the por-phrases that occur in these Pass
$_{\text{se}}$
constructions can be referential (e.g. vós ‘you’ in (36a), él ‘him’ in (36g), nos ‘us’ in (36h), cited by Kärde Reference Kärde1943: 90) in contrast to what we have seen for Modern Spanish in Section 2.1 above. Note that these ‘by’-phrases with se constructions occur as early as 1207, and as Monge (Reference Monge1954) observes as late as the 16th century. Some of his examples with referential ‘by’-phrases are in (37).
There is clear evidence that referential ‘by’-phrases were grammatical in Old Spanish. Yet, as we have seen above, they no longer are in Modern Spanish. This is consistent with patterns previously observed for Portuguese (Naro Reference Naro1976), Italian (Cinque Reference Cinque1988, Cennamo Reference Cennamo1993), and French (Heidinger & Schäfer Reference Heidinger, Schäfer, Fagard, Prevost, Combettes and Bertrand2010), where ‘by’-phrases were grammatical with Pass
$_{\text{se}}$
, but then became ungrammatical.
Thus, the ‘by’-phrase restrictions in Pass
$_{\text{se}}$
that hold in Modern Spanish did not hold in Old Spanish. The examples of referential ‘by’-phrases noted in this section span a period from around 1207 to 1550. In the previous section, we saw examples of inalienably possessed body parts in Pass
$_{\text{se}}$
from 1252 to 1500. These facts point to a stage of Spanish Pass
$_{\text{se}}$
that looked like Modern Romanian Pass
$_{\text{se}}$
. We conclude that there was a stage in the development of Spanish Pass
$_{\text{se}}$
that paralleled Modern Romanian Pass
$_{\text{se}}$
.
4 The proposal: A D(eterminer)-feature
We claim that the variation between Romanian Pass
$_{\text{se}}$
and Modern Spanish Pass
$_{\text{se}}$
can be accounted for in terms of different internal properties of the projected pro in Spec,Voice. Pro in Romanian lacks a D(eterminer)-feature and pro in Modern Spanish has a D-feature, following the analysis of Legate (Reference Legate2014) for a variety of languages.Footnote
[22]
We claim that the change from Old Spanish to Modern Spanish can also be accounted for in the same way: pro in Spec,Voice of Old Spanish Pass
$_{\text{se}}$
lacks a D-feature and pro in Spec,Voice of Modern Spanish Pass
$_{\text{se}}$
has a D-feature. This is illustrated in (38).Footnote
[23]
In this respect, we see both synchronic variation and diachronic change among implicit arguments parallel to differences between implicit arguments argued for independently in Landau (Reference Landau2010) and Legate (Reference Legate2014): the presence vs. absence of a D-feature.Footnote [24] As we claim in Section 5.2.1 below, the development of a D-feature on pro in Spec,Voice results from a subject cycle à la van Gelderen (Reference Gelderen2004, Reference Gelderen2011), reflecting a process she refers to as renewal.
Formally, we adopt Bruening’s (Reference Bruening2013) approach to ‘by’-phrases, which relies heavily on syntactic selection by feature checking. Informally, we follow Legate (Reference Legate2014), who discusses parallel patterns of ‘by’-phrase (in)compatibility in several languages and explains the patterns in terms of the internal make up of implicit arguments. Concretely, for what she terms impersonal constructions in Polish, Breton, and Irish she argues that the implicit external nominal expression in Spec,Voice has a D-feature.Footnote [25] Since the nominal expression has a D-feature, it is an argument, on the assumption that a determiner turns a predicate into a syntactic argument (see, for instance, Longobardi (Reference Longobardi1994), among others).Footnote [26] As an argument, the nominal expression can saturate the external argument position. As she shows, in the impersonal constructions in these languages, ‘by’-phrases are not allowed. In contrast, for what she terms grammatical object passives in Icelandic and Ukrainian, she argues that the implicit external nominal expression in Spec,Voice does not have a D-feature. Since it is not an argument, it does not saturate the external argument slot, and ‘by’-phrases are allowed. We believe that Legate’s analysis in terms of the internal properties of implicit arguments and their (in)ability to satisfy the argument structure properties of a predicate is the right approach. Nevertheless, we adopt the formal mechanisms from Bruening (Reference Bruening2013) to explain the patterns in these se constructions.
Bruening (Reference Bruening2013) proposes a feature checking account of argument selection in which a selectional feature is checked off when the appropriate category merges with it. A selectional feature no longer projects if it is checked, only non-checked selectional features project, and an unchecked feature leads to a crashed derivation.
Bruening assumes that Voice, the external argument introducing head, has two selectional features, one of category V and one of category N. Following his notation, Voice is Voice[S:V],[S:N] (see Adger Reference Adger2003). Thus, for the sentence in (39a), Bruening (Reference Bruening2013: 22) provides the structure in (39b), which illustrates selection by feature checking.
Upon merger, the senator checks the selectional N feature of V, which then does not project. Upon merger of Voice with V, the selectional V feature of Voice is checked and it does not project, and so on for the lobbyists. With respect to the passive, Bruening claims that there is a Pass projection that selects a Voice head with an unchecked N feature. A Voice head with an unchecked N feature is an unsaturated Voice projection for Bruening (Reference Bruening2013: 22). He assumes that Pass, when no by-phrase is present, saturates the external argument of Voice by existentially binding it. In checking terms, since Pass selects for a Voice head with an unchecked N feature, Pass checks said feature.
Bruening (Reference Bruening2013) also assumes that the by-phrase selects for an unsaturated Voice projection, namely a Voice projection with an unchecked N feature, just like the Pass head.Footnote [27] When the by-phrase combines with Voice[S:N], it does not check the selectional feature of Voice’s N feature, however; thus, Voice’s N feature projects. This happens, Bruening claims, because the by-phrase is an adjunct, and this reflects the intuition that the category with which an adjunct combines is the category that projects. The by-phrase’s selectional feature, in contrast, is checked by Voice[S:N]. Moreover, he assumes that Pass combines with the resulting Voice[S:N] projection and Pass checks the selectional feature of Voice, just like when no by-phrase is present in the passive.
We adopt this basic syntactic account, but follow Legate (Reference Legate2014), who refines it by arguing that only a nominal with a D-feature can saturate the external argument slot, not a nominal without a D-feature. Concretely, in the present context, in checking terms, we assume that Voice, apart from a selectional V feature, contains a selectional D-feature (see Müller Reference Müller2010, Alexiadou, Anagnostopoulou & Schäfer Reference Alexiadou, Anagnostopoulou and Schäfer2015). Thus, only 
can check Voice’s selectional feature; pro without a D-feature cannot. Consider how this accounts for the patterns.
When 
merges in Spec,Voice, Voice’s selectional D-feature is checked. A ‘by’-phrase cannot appear, because the ‘by’-phrase selects for an unsaturated Voice head, that is, a Voice[S:D]. In contrast, when pro merges in Spec,Voice, it does not check Voice’s selectional D-feature, because pro lacks a D-feature. In this case, the ‘by’-phrase can combine with Voice[S:D], since it is unsaturated. Recall, nevertheless, that the ‘by’-phrase does not check the selectional feature of Voice; the Pass head does this. In the present cases, however, there is no Pass head. Moreover, as Bruening assumes, a projection with an unchecked selectional feature is not an appropriate argument for any higher functional head. For instance, T selects for a projection with all its features checked off. This raises the question of what checks Voice’s selectional D-feature in Pass
$_{\text{se}}$
constructions when pro is in Spec,Voice.
The intuition we pursue is that se is responsible for checking Voice’s selectional D-feature when pro or the ‘by’-phrase cannot. This is consistent with the intuition in Pujalte & Saab (Reference Pujalte, Saab, Cuervo and Roberge2012) and Saab (Reference Saab2014), in which the presence of se saves a derivation with an unsaturated external argument position. Our account differs technically, however. As we discuss further below in Section 5.2, we assume that se in Pass
$_{\text{se}}$
constructions of the type in Romanian and Old Spanish still has a D-feature even though it heads Voice. This is not an uncommon situation in a subject cycle (van Gelderen Reference Gelderen2004, Reference Gelderen2011). We capitalize on this and on the idea that se establishes some syntactic relation with T (see Belletti Reference Belletti1982, D’Alessandro Reference D’Alessandro2007, Holmberg Reference Holmberg2010, Roberts Reference Roberts2010, among others); minimally it moves to adjoin to T. We suggest that se saves the derivation by carrying its D-feature with it to T. Thus, T with se has a D-feature 
and this configuration can check Voice’s unchecked D-feature when 
merges with Voice[S:D], saving the derivation.Footnote
[28]
Briefly consider two sets of noted differences between modern Spanish se constructions and modern Romanian se constructions, which have the potential of being explained in terms of the presence of a D-feature on pro in Spanish se constructions and its absence in Romanian se constructions. The first comes from the observation in Dobrovie-Sorin (Reference Dobrovie-Sorin1998: 424) that Romanian se can appear with the infinitive in (40a), which she notes cannot in Italian, and as we illustrate in (40b), cannot in Spanish.
A possible explanation of this contrast can be framed in terms of argument saturation. In Romanian, pro introduced in Spec,Voice in (40a) of the infinitive does not saturate the external arugment slot, in which case PRO can merge as an external argument. In Spanish, in contrast, since 
is merged in Spec,Voice in (40b), the external argument slot is saturated, leaving it impossible for PRO to merge.Footnote
[29]
The second set of data illustrates a contrast between Spanish and Romanian se constructions regarding the presence of the differential object marker (DOM). Spanish se can appear with it, Romanian se cannot, as illustrated in (41a) and (41b), respectively, from Dobrovie-Sorin (Reference Dobrovie-Sorin1998: 405).
Adopting an updated take on Burzio’s generalization (along the lines of Legate Reference Legate2014, although distinct), accusative Case manifested via DOM marking is available when there is a syntactic external argument. Only a 
counts as an argument and can license accusative Case; pro cannot.
5 The diachrony of se constructions: Linguistic cycles
In this section, we discuss the diachronic path of se constructions in more detail. The widely assumed path, not limited to Romance languages, nor to Indo-European languages, (see Geniušienė Reference Geniušienė1987, Haspelmath Reference Haspelmath1990, Cennamo Reference Cennamo1993, among others) is provided in (42).
We propose a modification of the path by dividing Pass
$_{\text{se}}$
into two stages, Pass
$_{\text{se1}}$
and Pass
$_{\text{se2}}$
, based on the ‘by’-phrase patterns discussed in Section 3 above and the analysis offered in Section 4. The modified path is illustrated in (43), for which we provide additional support in Section 5.1 below.
In this article, we focus only on the steps from Refl
$_{\text{se}}$
to Pass
$_{\text{se2}}$
. We illustrate how they appear to result from a subject cycle à la van Gelderen (Reference Gelderen2004, Reference Gelderen2011). By approaching the steps in terms of a cycle, we provide a context for understanding the developement of 
from pro. This relates directly to a crucial aspect of a cycle: renewal. Moreover, assuming that renewal underlies this development generates a prediction regarding the class of languages that will not develop Pass
$_{\text{se1}}$
from AntiC
$_{\text{se}}$
. We discuss this prediction in Section 5.2.3 and offer initial data that suggest that the prediction is on the right track.
5.1 Independent support for two Passse stages
One basis for the non-modified diachronic path of se constructions in (42) is the existence of languages that stop off at some point along the path. So, for instance, in Romance, Spanish, Italian and Portuguese are all known to have each of the constructions, while Romanian and French stop off at Pass
$_{\text{se}}$
; they lack Imp
$_{\text{se}}$
(see Dobrovie-Sorin Reference Dobrovie-Sorin1998, Reference Dobrovie-Sorin, Everaert and van Riemsdijk2006).Footnote
[30]
We are not aware of any Romance language that stops off at AntiC
$_{\text{se}}$
, but Geniušienė (Reference Geniušienė1987: 258) lists the following non-Romance languages, among others, that do: English, German, Nivkh, Khmer, Eskimo, and Fula.Footnote
[31]
Given our proposed modification, we would expect to find at least one language that has stopped off at Pass
$_{\text{se1}}$
and one language that has stopped off at Pass
$_{\text{se2}}$
stage. We claim that Romanian and French, respectively, are such languages.Footnote
[32]
First, if these languages have stopped off at one of the two Pass
$_{\text{se}}$
stages, then we would expect that neither of them would have developed Imp
$_{\text{se}}$
constructions. As mentioned above and as discussed extensively by Dobrovie-Sorin (Reference Dobrovie-Sorin1998, Reference Dobrovie-Sorin, Everaert and van Riemsdijk2006), this is the case for both Romanian and French.
Second, observe that French Pass
$_{\text{se}}$
allows an inalienable possession interpretation of the sole overt DP when it is a body part, as in (44a), and when the body part is in an adjunct, as in (44b).Footnote
[33]
In this respect, French and Romanian Pass
$_{\text{se}}$
pattern together. They differ, however, with respect to ‘by’-phrases. While Romanian allows ‘by’-phrases (examples in 16 above), French does not, as illustrated in (45), from Authier & Reed (Reference Authier and Reed1996: 514).Footnote
[34]
The widely assumed diachronic path of se constructions in (42) is too coarse to accomodate this difference. As Heidinger & Schäfer (Reference Heidinger, Schäfer, Fagard, Prevost, Combettes and Bertrand2010) show, there was a period when French did allow ‘by’-phrases in Pass
$_{\text{se}}$
; it no longer does. In fact, as mentioned above, this is a change that we have seen for Spanish and which has been documented for Portuguese (Naro Reference Naro1976) and Italian (Cinque Reference Cinque1988, Cennamo Reference Cennamo1993) as well.Footnote
[35]
The modified path offered in (43) better represents this change.
5.2 A subject cycle
In this section, we see that the steps from Refl
$_{\text{se}}$
to Pass
$_{\text{se2}}$
look like they result from a subject cycle á la van Gelderen (Reference Gelderen2004, Reference Gelderen2011). Concretely, in a subject cycle a pronoun is reanalyzed as part of the verb. As discussed by van Gelderen (Reference Gelderen2004, Reference Gelderen2011), there are two interrelated consequences of this. One is that the pronoun that is being reanalyzed as part of the verb loses features. Initially it loses its status as phrase, a full DP, at which point it is a D head. As a D head, it can become part of the verb. Finally, it loses its D-feature altogether. Ultimately the pronoun undergoes complete feature loss and disappears. The other interrelated consequence is that once the pronoun is no longer in the subject position as a result of being reanalyzed as part of the verb, this position is left open for what van Gelderen (Reference Gelderen2004, Reference Gelderen2011) calls renewal. Renewal refers to the filling up of the empty syntactic position with another pronoun. As the pronoun reanalyzed as part of the verb loses its features, a new pronoun in the empty position simultaneously contributes the features that are no longer provided by the reanalyzed pronoun. The two important aspects in the se constructions that we focus on here are se’s loss of features and the renewal of these features in terms of the development of pro to 
We see that as se loses its features, pro renews them simultaneously in parallel. We first discuss renewal.
5.2.1 Renewal in Spec,Voice
Recall the modified diachronic path repeated below in (46) in terms of how we analyze VoiceP for each se construction.
Example (46a) represents Refl
$_{\text{se}}$
constructions, widely assumed to be located at the beginning of the path of se constructions. We assume that pro in Spec,Voice is the definite referential null subject found in consistent null subject languages, represented as
$\text{pro}_{[\text{D}]}$
.Footnote
[36]
Recall from Section 3.1 that AntiC
$_{\text{se}}$
constructions do not license an inalienable possession interpretation of a body part in contrast to Pass
$_{\text{se1}}$
, Pass
$_{\text{se2}}$
, and, as illustrated in MacDonald (published online 23 October Reference MacDonald2017), Imp
$_{\text{se}}$
. Following this work, we take this to mean that there is no projected implicit external argument in AntiC
$_{\text{se}}$
constructions, while there is in Pass
$_{\text{se1}}$
and Pass
$_{\text{se2}}$
.Footnote
[37]
The lack of projected external argument is represented by the empty specifier in (46b). The projected implicit argument in Pass
$_{\text{se1}}$
and Pass
$_{\text{se2}}$
is represented as pro and 
respectively in (46c) and (46d).
We assume that the development of AntiC
$_{\text{se}}$
from Refl
$_{\text{se}}$
resulted from the reanalysis of se as a marker of intransitivity, following Monge (Reference Monge1954), Kemmer (Reference Kemmer1988), Cennamo (Reference Cennamo1999), and Pountain (Reference Pountain2000) (see also Faltz Reference Faltz1977, Reference Faltz2008, where it is inferred). The lack of pro in Spec,Voice and the single DP complement of V in (46b) reflect the intransitive structure underlying AntiC
$_{\text{se}}$
.
From AntiC
$_{\text{se}}$
to Pass
$_{\text{se1}}$
, we claim that renewal takes place. The empty Spec of Voice is renewed by pro giving rise to Pass
$_{\text{se1}}$
, a construction that licences ‘by’-phrases. We assume that the development from pro to 
also results from the renewal process, giving rise to Pass
$_{\text{se2}}$
, a construction that no longer licenses ‘by’-phrases. As we discuss below in Section 5.2.2, the gain of the D-feature from pro to 
goes hand in hand with the loss of the D-feature from 
to se, a situation that is common in cycles, as discussed in van Gelderen (Reference Gelderen2004, Reference Gelderen2011). At this point, Voice of Pass
$_{\text{se2}}$
looks likes Voice of Refl
$_{\text{se}}$
, in which there is a 
in its Spec. Pass
$_{\text{se2}}$
represents a return to the beginning of the cycle. The renewal taking place in Spec,Voice is represented below in (47) as a function of the diachronic path of se constructions.Footnote
[38]
5.2.2 Feature loss of se
Many have argued that the Old Spanish reflexive pronoun was a full phrasal DP (Rivero Reference Rivero1986, Reference Rivero, Campos and Martínez-Gil1991, Reference Rivero1997; Barbosa Reference Barbosa1993, Reference Barbosa, Halpern and Zwicky1996; Fontana Reference Fontana1993, Reference Fontana1997; Halpern & Fontana Reference Halpern and Fontana1994; Halpern Reference Halpern1995; Maddox Reference Maddox2016). At some point in the past, it was reanalyzed as a functional morpheme, on our account as the head of Voice (see also Cuervo Reference Cuervo2003, Reference Cuervo2014; Kempchinsky Reference Kempchinsky, J. Clancy Clements and Vance2004; Folli & Harley Reference Folli, Harley, Kempchinsky and Slabakova2005; Ordóñez & Treviño Reference Ordóñez, Treviño and Ortiz-López2011; Armstrong Reference Armstrong2013, among others, who assume that se heads a Voice projection), resulting in modern Spanish clitic se. As a Voice morpheme, nevertheless, in reflexive constructions, it still retains uninterpretable person features, since they vary as a function of the referent in Spec,Voice. Namely Refl
$_{\text{se}}$
surfaces as me when there is a first person singular referent and as te when there is a second person singular referent, etc. Observe for AntiC
$_{\text{se}}$
that se surfaces as me when there is a first person singular referent and as te when there is a second person singular referent, etc., as illustrated in (48), just like we see for Refl
$_{\text{se}}$
, suggesting no feature loss in the development from Refl
$_{\text{se}}$
to AntiC
$_{\text{se}}$
.
Recall that we assume that the development of AntiC
$_{\text{se}}$
from Refl
$_{\text{se}}$
resulted from the reanalysis of se as a marker of intransitivity. As a marker of intransitivity, se is not necessarily expected to lose features. The minimum expectation is that the resulting structure is intransitive. The intransitive status of AntiC
$_{\text{se}}$
is represented in our structure in (46b) by the lack of pro in Spec,Voice and the single DP complement of V.
We do see feature loss from AntiC
$_{\text{se}}$
to Pass
$_{\text{se1}}$
, nevertheless. Se no longer has uninterpretable person features. Se does, we assume, have an interpretable person feature, which restricts pro in its Spec to third person. (See Legate Reference Legate2014 for a way to capture this technically. See also footnote 3 above.) Moreover, we assume that in Pass
$_{\text{se1}}$
, se still has a D-feature, even though it heads Voice. This is not unexpected for a subject cycle, as discussed in van Gelderen (Reference Gelderen2004, Reference Gelderen2011), since not all features are all immediately lost when a pronoun is reanalyzed as part of a head. If we assume that se in Pass
$_{\text{se1}}$
still has a D-feature we are able to explain two related issues. The first brings us back to Section 4 and the checking of Voice’s selectional D-feature in Pass
$_{\text{se1}}$
. If se has a D-feature, se can check it, as we suggest above. The second relates to a question about why, if renewal is what gives rise to Pass
$_{\text{se1}}$
, can’t 
directly renew Spec,Voice, giving rise to Pass
$_{\text{se2}}$
directly from AntiC
$_{\text{se}}$
? Our answer relies on both feature loss of se and on renewal in Spec,Voice, two processes that occur simultaneously in parallel. Since the pronoun that is involved in renewal – here pro – contributes features that the pronoun reanalyzed as part of the verb can no longer contribute – here 
– the former will only renew the features that the latter is losing. So, if at Pass
$_{\text{se1}}$
stage, se still has a D-feature, then there is no need for 
to renew. It is only when se further loses this D-feature that pro can renew this D-feature. This, we claim, is the change from Pass
$_{\text{se1}}$
to Pass
$_{\text{se2}}$
, in which pro gains a D-feature giving rise to 
in Spec,Voice. From Pass
$_{\text{se1}}$
to Pass
$_{\text{se2}}$
, as part of the feature loss of the reanalyzed pronoun typical in a subject cycle, we speculate that se loses its D-feature.Footnote
[39]
Although our main focus of the diachronic path ends with Pass
$_{\text{se2}}$
, we consider here further feature loss in Imp
$_{\text{se}}$
constructions, since it lends general support to the claim that se loses features as we move to the right of the diachronic path in (43). Concretely, in Imp
$_{\text{se}}$
, there are cases where se loses its interpretable third person feature. This is evidenced by the lack of a requirement for a third person referent in Spec,Voice. The data in (49) below, from rural Brazilian Portuguese, taken from Nevins (Reference Nevins2007: 308), and in (50), from a variety of peninsular Spanish, taken from Benito Moreno (Reference Benito Moreno2015: 107), illustrate that in the presence of se, the external argument is not third person.
Nevins (Reference Nevins2007: 308 fn. 27) notes that the data in (49) suggest that se lacks a person feature.Footnote [40] The data in (50) suggest the same thing. We claim here that this is not an unexpected result if a subject cycle is underlying the diachronic path of se constructions in which se, as a pronoun reanalyzed as part of the verb, loses its features.
Example (51) illustrates the simultaneous processes of renewal in Spec,Voice and feature loss of se running in parallel as a function of the modified diachronic path of se constructions from Refl
$_{\text{se}}$
to Pass
$_{\text{se2}}$
. We illustrate only the D-features here.Footnote
[41]
As se loses its features, they are simultaneously renewed by pro in Spec,Voice. From this perspective, at least the portion of the modified diachronic path of se constructions from Refl
$_{\text{se}}$
to Pass
$_{\text{se2}}$
looks like a subject cycle à la van Gelderen (Reference Gelderen2004, Reference Gelderen2011). Consider a specific prediction generated from this subject cycle approach to se constructions.
5.2.3 A prediction from renewal
An integral part of any cycle is renewal. Renewal underlies the development of Pass
$_{\text{se1}}$
from AntiC
$_{\text{se}}$
on our account. Crucially, prior to renewal a syntactic position is left empty from the reanalysis of the pronoun as part of the verb. An expectation arises from this story. If a language uses a reflexive element in anticausatives, like Romance, but the reflexive element has not been reanalyzed as part of the verb and is an independent DP, we would expect renewal not to take place, because there would be no empty position to fill. On the present approach, this entails that in this language, no Pass
$_{\text{se}}$
construction would develop from AntiC
$_{\text{se}}$
. We suggest here that German is such a language.
Schäfer (Reference Schäfer2008) (see also Geniušienė Reference Geniušienė1987, Alexiadou et al. Reference Alexiadou, Anagnostopoulou and Schäfer2015, and references therein) argues that the reflexive element sich in German anticausatives is an independent DP, based on its free worder status, that it has (abstract) case, and that the auxiliary ‘have’, in contrast to ‘be’, surfaces just like it does with transitive predicates. Schäfer (Reference Schäfer2008) (see also Alexiadou et al. Reference Alexiadou, Anagnostopoulou and Schäfer2015) proposes the structure in (52) in which sich is located in Spec,Voice.
Importantly, German does not have Pass
$_{\text{se}}$
constructions (Florian Schäfer p.c.; see also Geniušienė Reference Geniušienė1987). If the development of Pass
$_{\text{se}}$
relies crucially on renewal and German anticausatives are analyzed as in (52), there is no empty position where renewal can take place, and no Pass
$_{\text{se}}$
will develop. These German facts fall out from the present proposal.
Consider another set of facts from Germanic that suggests that this explanation is on the right track. Languages such as Swedish and Norwegian have two reflexive elements, an independent DP (i.e. seg and sig, respectively) and an affixal form derived historically from a reflexive (i.e. -
$s$
) (Geniušienė Reference Geniušienė1987, Florian Schäfer p.c.). While one can find the independent DP reflexive in both ‘true’ reflexives and anticausatives, one does not find it in passives (or impersonals). Passives only occur with the affixal reflexive element (Geniušienė Reference Geniušienė1987).Footnote
[42]
As far as we are aware, there has been no previous explanation of these facts. On the present story, this situation is not unexpected. The independent DP cannot appear in passives, because there is no empty position for renewal to take place. In contrast, the affixal forms do not occupy a syntactic position, in which case, one is left open for renewal to take place. We take these patterns in Germanic as support for the proposal that the diachronic path of se constructions, at least from Refl
$_{\text{se}}$
to Pass
$_{\text{se2}}$
, results from a subject cycle à la van Gelderen (Reference Gelderen2004, Reference Gelderen2011).
6 Recap
In this article we have argued that Romanian and Spanish Pass
$_{\text{se}}$
constructions project an implicit external argument. They differ, however, in the internal properties of said implicit projected external argument; in Spanish it has a D(eterminer)-feature, while in Romanian it lacks a D-feature. Moreover, we saw that in Old Spanish the projected implicit argument in Pass
$_{\text{se}}$
patterned with Modern Romanian in not having a D-feature. We have argued that we can understand this change in terms of a linguistic cycle, one crucial part of which is renewal. This approach also offers an explanation for why some languages do not develop a Pass
$_{\text{se}}$
construction.
