Phase-edge properties and complementizer omission ∗

This paper deals with the diachrony of complementizer omission (C-omission) in some Italian clauses. C-omission is restricted to clauses that present [-realis] Mood in Old as well as in Modern Italian, and to some types of declarative clauses in Modern Florentine (Cocchi & Poletto, 2005). This phenomenon is instead much more pervasive in the Renaissance period (Wanner 1981, Scorretti 1991) and invests basically all types of subordinate clauses. The present study concentrates on C-omission in Renaissance Italian relative clauses, which is attested in both subject and non-subject extractions. There is a subject/non-subject asymmetry in the frequency of C-omission in relative clauses, which is analyzed as the result of the combination of an active vs. inactive distinction that characterizes both Old and Renaissance Italian, and the loss of V-to-C. The active vs. inactive distinction is attributed to the presence of a strong (*) feature on the low-phase head, v*, in both Old and Renaissance Italian, while the loss of (*) in CP determines the loss of V-to-C in Renaissance Italian only. The argument is corroborated by further comparative facts from Old Occitan and Old French, as well as by a contrast with Old Portuguese and Old Spanish.


Introduction
C-omission is a general term that conventionally indicates the possibility, in a grammar, to omit the subordinating element, being it a complementizer, a particle or some sort of pronoun.A language that to some extent permits C-omission is English.In English, complementizers can be dropped when introducing the declarative complements of so-called bridge-verbs (cf.Hooper and Thompson 1973, Bolinger 1972, Boskovič & Lasnik 2003, Staum 2005 and ref. therein, a.o.).
I heard about the fact *(that) Mary did it.(Boskovič & Lasnik 2003:534) The pair in (1) shows that C-omission in English complement clauses is generally accepted with bridge-verbs, but ungrammatical or degraded when the complement clause is (semi)-factive, (1b), or undergoes some syntactic operations, e.g.preposing/dislocation (cf.Boskovič & Lasnik 2003:527).It is however not the case that English C-omission is simply ruled out in island contexts, as other well known facts about relative clauses show (cf.Kayne 2010).
(2) a.I know the person (that) you met on the bus.b.
I know the person *(that) took the bus with you.
(2a) shows that the subordinating element can be omitted in non-subject extractions.This is a possibility that is typologically quite widespread (a.o. in Mainland Scandinavian, South-East Asian languages, Natchanan & Amara 2008).By contrast,(2b) shows that that-omission in subject extractions is ungrammatical, and provokes garden-path effects (McKoon & Ratcliff 2003).In general, subject relative clauses with an optional subordination marker have a typologically limited distribution (cf.Comrie & Kuteva 2005).
The study that is presented in this paper focuses on the diachrony of Italian, and related comparative facts.It is no new fact that Modern Italian, which is historically derived from the medieval variety spoken in the Tuscan area around Florence, has quite restricted C-omission.In Modern Italian C-omission is possible in declarative complements if the embedded verb is in subjunctive or another [-realis] mood (3a), whereas Comission is ungrammatical with indicative mood, even if the matrix predicate is a (5) a. Dice __ lo porta.
(Modern Florentine) say;3SG ACC;3SG take.IND;3SG 'He says he will bring it.' 1 A connection between subjunctive mood marking and C-omission has been also observed and analyzed in other Romance languages, see e.g.Schneider (1999Schneider ( , 2007) ) for Spanish.
2 Notice that C-omission is ungrammatical in semi-factive clauses such as those selected by (negated) 'know '-type (Cocchi & Poletto, 2005) A Modern Italian sentence equivalent to (5a) is ungrammatical.Cocchi & Poletto explain the difference between Modern Italian and Modern Florentine as a parametric difference concerning the contexts in which Alternative Checking (AC) can take place.Specifically, they argue that the illocution features encoded on the C projection ForceP (cf.Rizzi 1997) may either be checked by lexical insertion of a C-functional element, che, or via Agree with another functional element, which is located in the IP left-periphery (cf.(ii) above).In the latter case, che-omission is licensed via AC of the features encoded in ForceP, which is performed by such functional element.As Cocchi & Poletto (2005) themselves observe, this account is not straightforwardly applicable to the type of C-omission attested at an earlier stage of the grammar, namely in Renaissance Italian .In several Renaissance Italian corpora C-omission is attested in a much broader set of contexts, including non-finite sentences, purpose, factive, comparative and relative clauses (Wanner 1981, Scorretti 1991, see section 2.2).Specifically, relative clauses show a subject/object asymmetry in the frequency of C-omission (as it is illustrated in section 2.2), which cannot be directly explained under the AC proposal of Cocchi & Poletto.In this perspective, it is not clear why a non-subject relative-OP would perform AC more easily than a subject relative-OP.Furthermore, the AC account does not seem to properly account for Comission in other languages.For instance, English that-omission affects different types of subordinate clauses and cannot be directly related to AC of Force features.In this case, it is not clear what could constitute an alternative checker (cf.Jaeger 2005, 2010, Jaeger & Walter 2005, Levy & Jaeger 2007, a.o.).The investigation presented in this paper tries to identify the conditions allowing for C-omission in Renaissance Italian relative clauses.As mentioned above, C-omission in subject-relative clauses is typologically restricted, thus the question is what permits C-omission in subject relative clauses and, more generally, in various types of clauses in a grammar.This issue is explored from a diachronic perspective, by analyzing various changes that affect the Italian grammar from its initial stage (Old Italian), into its present stages (Modern Italian and Modern Florentine).I propose that the massive C-omission attested in Renaissance Italian corpora, but not at previous or later diachronic stages, depends on the interplay of various properties of functional phase heads.Specifically, C-omission in relative clauses is possible because a change affecting the CP phase, namely the loss of V-to-C, combines with a still productive property of the vP phase, namely an active vs. inactive distinction.
The paper is structured as follows: section 2 illustrates the relevant facts and changes concerning the possibility of C-omission in Old and Renaissance Italian; section 3 presents an argument for the hypothesis that massive C-omission in Renaissance Italian is due to the loss of V-to-C, and to the presence of an active vs. inactive distinction.Section 4 presents further facts from Old French and Old Occitan in support of the hypothesis that the loss of V-to-C, in combination with an active vs. inactive distinction, permits C-omission in relative clauses.Additional data from Old Spanish and Old Portuguese show how C-omission does not uniformly affect Old Romance varieties.Section 5 offers an account for the diachronic change from Old to Modern Italian in light of the proposed analysis and section 6 concludes the paper.

Data
In this section I present the relevant data concerning the lexicalization of C in Old Italian (2.1) and Renaissance Italian (2.2).Interestingly, the two diachronic stages differ with respect to both the presence of V-to-C and the possibility of Comission.

Old Italian
On a par with other Old Romance languages, Old Italian (around 1200-1350), from which Renaissance Italian, Modern Florentine, and Modern Italian descend, has a so-called 'V2 property', whereby the inflected verb always raises to the Cdomain in root clauses 3 (cf.Benincà 1984, 2006, Benincà & Poletto 2010, Poletto 2006, 2014, Roberts 1993, 2007, a.o.).This is visible, for instance, in (main) clauses where adverb preposing is followed by Aux-S word order, as in ( 6), whereas the Modern Italian word order would be Adv-S-Aux, S-Aux-Adv or an order in which the subject follows to the right of the past participle.(6) Primieramente avea ella fatta a llui ingiuria.(Old Italian) First had she done to him injury 'She had offended him for first.' (Brunetto Latini, Rettorica, 116) Benincà (1984Benincà ( , 2006) ) convincingly argues that V-to-C is limited to roots contexts, in Old Italian, whereas in subordinate clauses the inflected verb generally remains in the IP domain, as the frequent pronominal subject -finite V order shows.(7) Poniamo ch'io sapesse che tu vuoli rubare una buttega.Put.1PL that'I know.1SG;SBJV that you want.2SGrob a shop 'Let's assume that I know that you want to rob a shop.' (Trattati Morali di Albertano da Brescia, 12, 2, 1268) C-omission is highly restricted in Old Italian.Instead, the complementizer che (and its variants ch', ke, etc.) may be doubled, as in the example below (cf.become.COND crazy 'He found out that whoever ate eggplants for nine days in a row would become crazy.'(Novellino,35,208,2) C-doubling is a widespread phenomenon across Romance languages, and it is characterized by grammar-specific properties.For instance, some Northern Italian Dialects allow for C-doubling only with complement clauses in subjunctive mood (Paoli 2003 for Piedmontese and Ligurian), see ( 9) and ( 10).Spanish lexicalizes the same C-form in both positions (Demonte & Fernandez-Soriano 2009), whereas various Southern Italian Dialects lexicalize with different forms the two positions (Ledgeway 2003(Ledgeway , 2005 (Paoli 2003, 110:5) C-doubling cases attested in Romance share the property that some lexical material separates the two C occurrences, i.e. these are not adjacent. 4From a cartographic perspective (cf.Rizzi 1997, Haegeman 2006, a.o.), this fact seems to support the idea that whenever the Topic-Focus field is not activated, the (Sub/)Force-Finiteness system is collapsed into a single head, whereas activation of left-peripheral criterial positions provokes a split in the CP. 5 Rizzi (1997:312-313) argues, on the basis of English facts, that in case of split CP of a finite clause, Force must lexicalize, whereas Fin has no morphological realization.However, the facts in ( 8) and ( 9) show that this is not always the case, since the lower C-position may as well be lexicalized by a morphologically identical functional element, as is also proposed in Belletti (2009Belletti ( , 2012Belletti ( , 2013) ) for clefts.In other cases, only this position, between the two, is lexicalized, and the outcome is the string: HighC Ø -TopP/FocP XP -LowC che (Segre 1952, Vincent 2006, Meszler & Samu 2010).
In the above-mentioned Italo-Romance varieties the lexicalization of the lower C in the C-doubling cases seems to be somehow dependent on Mood marking.The same restriction applies to Old Italian: Franco (2009) observes that che-doubling is generally attested with embedded clauses that are marked with [-4 Cf.Saab (2011) on anti-adjacency effects of head reduplication.5 This is no new idea.An analysis of C-omission in Modern Spanish along these lines is proposed in Antonelli (2013).realis] Mood, i.e. if the finite verb is in the subjunctive, in the conditional or in the future indicative (with deontic or epistemic value).
A similar restriction is visible in Northwestern Italian dialects, where chedoubling is impossible if the embedded clause is not in the subjunctive (cf.Paoli 2003).
The John SCL says C the Mary C SCL not eats.IND of frogs 'John says that Mary does not eat frogs.' (Paoli 2003: 102-107, 1-3) Another option attested in Old Italian is one in which an element that is dislocated to the left periphery creates a potential context for C-recursion, but in this case the lower C-head is not lexicalized as che, cf. ( 11).
(11) Costuma era per lo reame di Francia che [l'uomo ch'era Custom was through the kingdom of France C the'man C'was.IND degno d'essere disonorato e giustiziato] sì andava… worth of'be.INF dishonored and executed si went.IND 'In the kingdom of France it was customary that the man worth being dishonored and executed went…' (Old Florentine,Novellino,27,192:1) In ( 11) the lower che is absent, and we find instead the particle sì, which is analyzed as a CP element located in FocusP (Poletto 2005) or in the lowest CP position (Ledgeway 2008).Crucially, the embedded clause in ( 11) is in the indicative mood and, differently from the doubling construction in (8), there is no lower che introducing it, only sì.From a first corpus search, lower che and sì appear indeed to be in complementary distribution. 6 With respect to Old Italian, the most relevant facts to the present discussion are the following: (i) C-doubling is a way to mark Mood (i.e.irrealis che…che vs. indicative che…(sì) constructions); (ii) radical C-omission is not attested.By 'radical C-omission', I refer to the possibility of omitting the subordinating element in all the available positions in which it can be lexicalized.If 'radical C-omission' were possible, the higher and the lower C, in C-doubling contexts, or the only C, in non-doubling contexts (i.e. when there is no dislocation) could be omitted, which is not attested in Old Italian.
To summarize, we have seen so far that Old Italian has both productive Vto-C in root clauses and requires lexicalization of C at least in one of the dedicated CP positions.I argue in section 3 that these two properties are the effect of the strength of CP-phase-edge features. 76 In the absence of sì, the lower C position remains empty, all other conditions being equal to those that apply in (11), cf.Vincent (2006), Franco (2009:202) and Meszler & Samu (2010) for data.

7
I have not specified what the C-positions mentioned above are.Following a cartographic perspective, I assume that che may lexicalize both the ForceP and FinP heads, which may be split (as in C-doubling clauses) or collapsed (in simple clauses, without dislocations).This would be compatible with the idea that Force and Fin operate in synergy (cf.Rizzi 1997), that is to say that the features that are

Reinassance Italian
In Renaissance Italian V-to-C is no longer productive: in root clauses the verb raises to the CP domain sporadically, non-systematically and only in a restricted number of contexts, for instance following preposed adverbials, adverbs or adjunct phrases. 8(12) [Considerate le difficultà le quali s'hanno a tenere uno stato Considered the difficulties the which se'have.3PL to keep.INF a state occupato di nuovo], potrebbe alcuno maravigliarsi… occupied of new could.3SGanyone be-surprised 'After considering the difficulties that there would be in occupying a state again, anyone could be surprised…' (P, IV, I) Moreover, Renaissance Italian displays radical C-omission (Wanner 1981, Scorretti 1991, Cocchi & Poletto 2005), in contrast with both Modern Italian and Modern Florentine, on the one hand (cf.section 1), and Old Italian, on the other (cf.section 2.1).The peculiarity of this phenomenon consists in the broad variety of syntactic contexts that it pervades, rather than in the frequency with which it is attested.The frequency of C-omission is arguably dependent on sociolinguistic factors, as the variation among texts belonging to different literary genres suggests.However, it is difficult to assess what ultimately determines a higher frequency of C-omission in a corpus rather than in another one, due to the limitation of available data and information.
With respect to its pervasiveness, C-omission in Renaissance Italian can be easily detected in several clause-types and syntactic contexts.Already Scorretti (1991) reports, along the lines of Wanner (1981), that C-omission is attested both in finite, ( 14)-( 16), and in non-finite clauses, the latter otherwise introduced by the complementizer forms di (of), as in (13), or per (for), which are used in control clauses, in Modern Italian. 9(13) Cercassi __ torli.tried.1SG;SBJVtake.INF.ACC;3PL 'I tried to take them.' (Mandragola; Cocchi & Poletto 2005:25) encoded on FinP are visible to ForceP, as these functional projections belong to the same phase edge.

8
A systematic study of the contexts in which residual V-to-C is attested in Renaissance Florentine still has to be done (cf.also Poletto 2014, Franco 2015).9 These complementizers allow for clitic climbing in Old Italian (Cardinaletti 2010, Franco et al. 2014), which suggests the possibility that they do not embed a full CP but rather they are functional heads.For a proposal concerning Old Italian clitic climbing see Kastelein (2012).For a discussion on the similarities between finite and non-finite clauses in Old Italian see Poletto (2014:98ff As the examples in ( 13)-( 16) show, C-omission in Renaissance Italian is not restricted as it is in Old or Modern Italian, or Modern Florentine, since neither [realis] marking nor the conditions given in (i)-(iii) of section 1 for Modern Florentine constrain its distribution.
A separate discussion must be reserved to C-omission in relative clauses, which is the main focus of this paper.In modern Romance languages C-omission is usually ungrammatical in relative clauses.The ungrammaticality regards symmetrically both subject and non-subject extractions, as the French sentences in (17) show (cf.(4) in section 1 for Italian, and section 4 for a comparison with Old French, Old Occitan and Old Ibero-Romance; cf.J'ai parlé avec l'homme *(qui) à été ici.I'have spoken with the'man who has been here 'I have spoken with the man who has been here.'Instead, some Romance languages, such as Old Occitan and Old French, display C-omission also in relative clauses, at a previous stage of the grammar that roughly corresponds to the Italian Renaissance period (Scorretti 1991 and ref. therein). 10Renaissance Italian equally displays C-omission both in subject (18) and non-subject extractions (19), with a significant asymmetry that indicates a reduced number of C-omissions in subject extractions.This can be seen in table 1, which shows data from three texts (cf.Sources, this paper).
(18) a. Che è faccenda ___ tocca a noi.that is issue touch.3SG to us 'That is an issue we have to deal with.' 10 There is still no systematic study comparing the period in which C-omission was productive also in relative clauses in the various Old Romance languages in which it is attested.There are no cases of headless relative clauses with C-omission, which follows from the expectations, as will become clear in sections 3, 5 and 6.
The table shows that attested cases of C-omission in relative clauses are at most 31% of all the relative clauses (see '% Total' row: 2% in P, 25% in AMS and 31% in VBV).This piece of data indicates that C-omission is not the preferred option in any of the examined corpora, despite the significant degree of variation among them.It is quite difficult to assess what is the reason for such variation, namely whether only diachronic or also diaphasic factors are at play.From a first analysis it seems that C-omission is more limited in texts belonging to a higher/more formal register (P is a political treatise) by contrast to texts of a more colloquial/informal style (AMS are letters, VBV biographies).Another factor that may potentially play a role in the frequency of C-omission is of diachronic nature: the most recent corpus among those analyzed (P) is the one with less C-omission, which might indicate that radical C-omission is disappearing already around 1500.Nonetheless, further data collection is needed in order to formulate an empirically grounded hypothesis. 12From a comparison between the numbers reported in the rows corresponding to the subject (Rel.S) and the object (Rel.O) relative clauses in each corpus, a subject/object asymmetry becomes immediately evident.Comission is much more frequent in object relative clauses (3 cases in P, 29 in AMS and 53 in VBV) than in subject relative clauses (no cases in P, 7 in AMS and 32 in VBV).The asymmetry can be quantified if we look at the percentage of C-omission in subject and object relative clauses with respect to the total of subject and object relative clauses.In P, 100% of the clauses that display Comission are object relative clauses, whereas in the other two texts the percent of object relative clauses on the total of clauses with C-omission is a little lower: 81% in AMS and 62% in VBV (see cyphers in bold in the REL -C column).In all three texts, far more than 50% of the clauses that display C-omission are object 12 It is worth mentioning that there is no syntactically parsed corpus for either Old or Renaissance Italian yet; so all the texts have to be manually parsed.
relative clauses.If we look at subject relative clauses, we see that their percent on the total of clauses with C-omission drops proportionally: only 38% in VBV, 19% in AMS and 0% in P. Conversely, subjects represent the most frequent type of items that are relativized in clauses that are introduced by a C element (66% in P; 56% in AMS and 70% in VBV).
The results can be summarized in the following observations: • C-omission is possible in relative clauses, and more frequent in object relative clauses, than in subject relative clauses.• C-omission is never the preferred option: only in VBV is there an equal number of C-less and C relative clauses, in the case of object extraction (53 C-less and 53 with C, see table 1).• Subject headless relative clauses are more frequent than object ones, and there is no case of C-omission.• Although C-omission is sporadic in headed relative clauses on the subject, it is attested.Notice that C-less subject relative clauses represent a potential processing ambiguity in a language like Renaissance Italian, that is, they may trigger gardenpath effects, cf.(18) above, (McKoon & Ratcliff 2003).The open issue at this point is how to explain the presence of C-omission in subject relative clause at all, in Renaissance Italian, against the typological scarcity of this type of sentences.

The analysis
In this section I first discuss the diachronic change underlying C-omission in declarative clauses in general (section 3.1) and then I offer an analysis of Comission in relative clauses that accounts for the subject/object asymmetry as a result of a specific property of the phase-edges, according to which arguments respond to an active vs. inactive distinction in Old and Renaissance Italian (section 3.2).The proposal further accounts for the frequency, and thus for the optionality, of C-omission as a result of parametric 13 change from a system with V-to-C to a system without V-to-C.

Diachronic change and C-omission
As mentioned in section 2.2, V-to-C, which characterizes the Old Italian system, becomes residual in the Renaissance period.I take this fact as a signal of an underlying variation that concerns the feature specification of the CP domain.More specifically, I assume that V-to-C is productive in grammars where the lowest C-head, Fin0 (Rizzi 1997), encodes a strong (*) feature (in the sense of Chomsky 1993, 1995, Lasnik 1999, Biberauer & Richards 2006, a.o., cf. below), and is thus Fin*.FinP encodes [finiteness], which permits the interpretation of temporal/locative coordinates and nominal deixis (cf.Bianchi 2003, Sigurðsson 2004, 2011).This means that time and location of the event/state expressed by the verb, as well as the person features of the arguments are anchored to the discourse context, and thus, interpreted, by checking [finiteness] on Fin*P.Along the lines 13 Parameters are to be understood as properties of functional heads, in the present perspective.
of Chomsky (1993Chomsky ( , 1995)), Lasnik (1999), andBiberauer &Richards (2006), I assume that in systems where a functional head bears a strong feature, checking requires overt Merge on that head. 14This explains why in Old Italian there is Vto-C and no C-omission:15 at this stage, the system has Fin*, which means that [finiteness] must be checked by overt Merge.This is done by V-to-C in root clauses, and by C-merger in subordinates, under the assumption that subordinating elements such as che in Italian may lexicalize both Fin and Force (cf.Belletti 2009Belletti , 2012Belletti , 2013, for , for  In Renaissance Florentine, the change from Fin* to Fin determines the loss of the requirement that the features encoded on Fin be checked via overt Merge.The transition from Fin* to Fin is visible in the loss of V-to-C, i.e. the loss of the trigger for feature-checking movement of finite V to the CP domain.In this sense, residual cases of V-to-C can be accounted for as cases of marked illocution, where the feature requiring checking via Merge is in fact not encoded on FinP, but on another (higher) functional head, e.g.Topic, Force, etc. (cf.Benincà & Poletto 2004, Poletto 2014, a.o.).
The Fin*>Fin change is a crucial factor for C-omission in Renaissance Italian (cf. the declaratives in ( 13)-( 15)).At this stage, C-omission in declarative clauses is not restricted to clauses marked with [-realis] Mood morphology, as is the case for Old and Modern Italian. 16Put differently, the change affecting Renaissance Italian has C-omission as a 'side-effect': once the overt Merge condition on Fin falls, V-to-C is lost and C-merger is no longer required.
Notice that C-omission is just optional at this stage, not obligatory.In other words, C-Merger is no longer required, but still grammatical and, as such, possible. 17Renaissance Italian can thus exploit two options: one with C-omission, 14 According to Lasnik (1999), strong features require overt Merge in order to avoid crash at PF.This optionality resembles in a sense the possibility of overt pronominal subjects in null subject languages, with the difference that, for null subject languages, the insertion of overt subject pro-s is pragmatically restricted, whereas the conditions for inserting an overt C in Renaissance Italian are not clear (i.e. it is not known whether C-insertion depends on sociolinguistic factors, e.g. by representing a more formal/accurate register, cf.section 2.2 above).An anonymous reviewer asks what are the conditions that allow or require C-insertion in Renaissance Italian, given the loss of (*) on Fin features.As far as I can tell from the data I analyzed, there is no specific requirement for C-insertion in declarative clauses.C-insertion is still the preferred option according to a written corpus analysis, but we do not know what was the frequency of C-omission at other register levels, and specifically in the spoken language.Written language changes more slowly which depends on the loss of the Merge requirement resulting from the Fin*>Fin shift.A second option, with C-Merge, simply results from a previous grammaticalization imposed by a Fin* system, but still possible under Fin.However, the innovative option, C-omission, never becomes more productive than C-insertion (see section 2.1 and table 2 below), probably partly due to the fact that C-omission undergoes some restrictions in certain syntactic contexts, such as relative clauses, which will be discussed in section 3.2ff.As a consequence, C-omission does not constitute relevant input for developing widespread C-omission in Modern Florentine and Italian (cf.section 5), and Cinsertion remains the default case.According to Cocchi & Poletto (2005), in the modern varieties C-omission is indeed limited to contexts in which AC can be performed (cf.section 1).I address the issue of the diachrony of C-omission after the Renaissance period more in detail in section 5, whereas in the following section I offer an account for Comission in relative clauses.

The active vs. inactive distinction and C-omission in relative clauses
A characteristic that remains more or less unchanged up to the whole Italian Renaissance period is a morphosyntactic distinction of arguments that is based on their semantics, thus on their theta role.Ledgeway (2012: 236) observes: "in the passage from classical Latin to Romance there is initially a notable decline in the nominative/accusative orientation of the nominal and verbal system, paralleled by a corresponding expansion in the range of the active/inactive alignment in the verbal and nominal domains".According to Ledgeway, the consequences of such "realignment" are observable at verbal, nominal and clausal levels, as is visible in the cases of past participle agreement, auxiliary selection, bare plural NPs, and word order (typically pre-and postverbal subjects, depending on the thematic role they cover; see Ledgeway 2012:335-339 and ref. therein).
As is also reported in La Fauci (1988), Formentin (1996), Parry (2005), and Ledgeway (2009: 963-7), Ledgeway (2012:308) further observes that the active vs. inactive distinction is also marked on the nominal morphology at the level of complementation.He argues that the Latin NOM vs. ACC distinction that is visible in the alternation between QUI 'who' vs. QUEM/QUOD 'whom/which' is "substituted", in early Romance, by the forms qui/chi for active arguments vs. than spoken language, and the Renaissance period in which radical changes are visible in the grammar covers a relatively short time elapse (before the prescriptivism that invests basically all written production in Italy, Bembo 1525).Thus we cannot know whether a more radical C-omission occurred in the spoken language, but never fully showed up in the written texts que/che, for inactive ones.However, the correspondence between semantic features, such as active or inactive, and morphology is in fact not so neat.In some cases qui/chi is used for non-active subjects, which indicates that the system is still sensitive to the NOM-ACC distinction (i.e.qui/chi may correspond to NOM inactive, see fn. 18 below).For this reason, I do not take these facts to mean that the Old Romance system features an alignment that would resemble that of ergative languages.Rather, I simply maintain that many Old Romance morphosyntactic phenomena (see Ledgeway 2012 for an overview) are explained under the hypothesis that (human/intentional) Agents (SA/A) often behave differently from non-active subjects, i.e. non-agentive subjects and objects (SO/O) (see also below and Benincà and Cinque 2010).Below I argue that this distinction depends on the feature specification of the highest head in the event domain, v.According to some analyses (Kratzer 1996, Alexiadou & Anagnostopoulou 1999, 2003, 2004, Alexiadou, Anagnostopoulou & Schäfer, 2006, a.o.), such head is Voice, and projects a specifier whenever an agentive external argument (EA) is merged, whereas non-active arguments are merged lower in the structure.My proposal just assumes that the head that assigns the Agent theta role is v.More specifically, in active constructions v encodes an [Agent] feature that is checked against merger of an agentive EA.This idea is also compatible with other analyses of the event domain (e.g.Folli & Harley 2005).The crucial point is that the distinction of Agent from other theta roles depends on the feature specification of the highest theta-assigning head (cf.with Ramchand 2008, where such head is labeled Initiator).
Le cosse che lo me respose no vojo a voi The things that he to.meanswered not want.1SGto you tener as [c]  (i) As a pronoun, when following a preposition.In this case it is always referred to a [-animate] antecedent, as bastone (=stick) in ( 24).
a stick with that refl pointed because was weak 'A stick with which he sustained himself because he was weak.' (Fiori e vite de' filosafi, 9, 4-5, in Benincà & Cinque 2010: 472, ( 6)) (ii) As a complementizer, when it does not follow any preposition and it introduces a relative clause on the subject or on the object.In this case, "che is insensitive to the semantic +human/-animate distinction, because it can also introduce a relative clause on a [+human] antecedent" (Benincà & Cinque 2010:473, my translation).
(25) Andò alli altri giovani che stavano Went to.the other youngsters that stayed a ricevere l' acqua piovana… to receive the water rainy 'He went up to the other youngsters that were staying under the rain…' (Novellino, 4, 16-17, in Benincà & Cinque 2010: 473) For this reason, Benincà & Cinque conclude that che, in a sentence like (25), in which it introduces a relative clause on an animate subject (altri giovani), is a complementizer following an abstract pronoun (2010:473).The existence of che as a complementizer (cf.(ii) above) means that the che that is employed in many relative clauses is in fact syncretic with the complementizer che that introduces declarative complement clauses.We have already seen in section 2.2 that declarative complementizers can be omitted in Renaissance Italian, and in section 3.1 I have argued that this is the result of the change from *Fin to Fin.C-omission in relative clauses in Renaissance Italian can thus be explained as a consequence of the syncretism between the declarative che, an instance of which is given in (7), repeated below for convenience, and the relative che, see (25) above, which are basically the same complementizer form with two different functions (cf.Hendery 2012 for a typological overview of the complementation strategies in relative clauses).
(26) Poniamo ch' io sapesse che tu vuoli Put.1PL that I know.1SG,SBJVthat you want.2SGrubare una buttega.rob a shop 'Let's assume that I know that you want to rob a shop.' (Trattati Morali di Albertano da Brescia, 12, 2, 1268) Put differently, because the declarative complementizer che can be omitted, the relative complementizer che can also be omitted. 19 This is expected to hold only for those cases in which che is a complementizer, not a pronoun.That is to say, prepositional relative clauses such as (24) should not allow for C-omission at any diachronic stage, which is confirmed by facts (sentences like "uno bastone (con)__ s'apogiava" are not attested).Moreover, C-omission should not affect the subordinating C-form of headless relative clauses, because in these clauses the C-form is in fact a pronoun (e.g.chi, cf.table 3), not a complementizer.This expectation is also borne out by facts, as is mentioned in section 2.2 (headless relative clauses with C-omission are unattested).
However, an explanation that is purely based on such syncretism cannot be the full story.According to what has just been proposed, che+abstract pro can introduce both subject and object relative clauses, thus che omission should be equally possible for both types of extraction, in this perspective.Instead, Renaissance Florentine C-omission in relative clauses displays a subject/object asymmetry (cf.table 2), which is so far left unexplained.
In order to account for this asymmetry, I have examined more attentively the theta role of the antecedent: the asymmetry apparently concerns subjects vs. objects, but, as is discussed at the beginning of this section, Old and Renaissance Italian are sensitive to which theta role the argument(s) is/are assigned in the event domain.Since this sensitivity is reflected in the morphology of relative pronouns in some varieties (cf.Old North Western dialects and Old Neapolitan), we can expect that a similar distinction be somehow marked also in coeval Tuscan varieties.We have seen above that this marking does not concern the morphology of the subordinating element, as this may be an invariable complementizer.I argue below that the active vs. inactive distinction is visible on the possibility of C-omission itself.

19
An anonymous reviewer points out that it is not clear that relative che spells out Fin, and is thus subject to C-omission on a par with declarative che, once the Fin*>Fin change occurs.I assume that the finite complementizer che is always merged in Fin, and when it appears on a higher C-head (e.g.Force, see Rizzi 1997), it is because Fin-to-Force movement has taken place (see Ledgeway 2005 on complementizer movement).If the Fin*>Fin change has taken place in the grammar, then Fin does not require merge, and if a complementizer is not merged in Fin, it cannot move to Force (or any other higher C head).This explanation holds both under the assumption of a 'rigid' left periphery that is always split, or of a structure that splits up only when the Focus/Topic field is activated (Rizzi 1997).
If we take a look back at table 1, we can see that the cases of C-omission in subject relative clauses are 7 in AMS and 32 in VBV (no cases in the P corpus).Interestingly, the extracted subject in these clauses has the following properties: • it is either [neuter] or [feminine] (e.g. an abstract/inanimate entity, as in ( 18) and (27a) or a feminine argument, as in (27b)); • it is non-agentive (e.g. the subject of a inactive predicate), as in ( 28). ( 27 )) The formulation of the properties (i) and (ii) above is thus intended to capture the distribution of C-omission in relative clauses (and more, generally, in extraction contexts). 20I argue that the active vs. inactive distinction is expressed by the properties of vP also in Renaissance Italian.When there is no agentive EA, v does not encode [Agent], whereas when v encodes such feature, an agentive EA argument is merged in its specifier.I take vP to be a phase only in the latter case, namely when v bears (*).v* assigns an Agent theta role and is phi-complete. 21 Notice that in Old Italian various syntactic phenomena involve movement of some lexical elements to the vP edge (widespread clitic climbing (WCC), object or XP-low scrambling, asymmetry in stylistic fronting) and VP-ellipsis is possible.These phenomena are argued to result from (*) on one of the functional heads that merge in a (split) vP (Franco et al. 2014, cf. Poletto 2014).Capitalizing 20 At this point it is still not possible to establish whether (i) and (ii) are hierarchically ranked.See the end of this section and section 3.3.

21
Following Chomsky (2001), (*) on v indicates that v* is phi-complete.According to Richards (2012:201) v* encodes [uPerson] and [uNumber], whereas v, phiincomplete, only [uNumber].Because of uninterpretable features on both v and v*, Richards argues that the notion of strength is not relevant to determine phasehood, and proposes a solution based on feature-inheritance.I assume that strength on v corresponds to phi-completeness, since an approach based on feature-inheritance does not work for OI, for reasons that I cannot discuss here, as they are beyond the scope of the paper.
on Poletto (2014), I assume that the left periphery of a vP phase in Old Italian is as follows.For Old Italian, Franco et al. (2014) and Franco (under review) propose that active structures are biphasic, and consist of a vP-phase, where the agentive EA is merged, and a CP-phase, where nominal deixis identifies the person features of the (agentive) subject, which are thus interpreted in relation to the discourse context, (cf.section 3.1, first paragraph).I have argued that in Old Italian (*) on v marks the vP as a full phase, which explains why only the material that is located on the vP edge, including the vP periphery, is accessible to probing operations from the higher phase (CP).This material is agentive EAs, but also fronted arguments/adjuncts, when another vP-peripheral head is (*) (cf.Poletto 2014, Franco et al. 2014).Conversely, the material that remains in the complement of v* is invisible to further probing, under the PIC. 22 In Renaissance Italian the vP periphery is no longer 'activated', as is witnessed by the fact that vP peripheral phenomena such as XP scrambling no longer take place (cf.Poletto 2014).The structural differences between Old and Renaissance Italian are given in figure 1 and 2 respectively.22 Which version of the PIC is taken to hold is irrelevant at this point.The two versions proposed by Chomsky (1995Chomsky ( , 2001) ) are given in (i), with (ib) the most frequently adopted, as it permits to account for a series of agreement phenomena that are unexplained under (ia), see Gallego (2010), Richards (2012), among many others.
, with H and Z the heads of phases: a.

Phase Impenetrability Condition 1 (strong version: PIC1)
In phase α with head H, the domain of H is not accessible to operations outside α; only H and its edge are accessible to such operations.b.

Phase Impenetrability Condition 2 (weak version: PIC2)
The domain of H is not accessible to operations at ZP; only H and its edge are accessible to such operations.One crucial difference is that, while the CP periphery may be activated with merger of topics or foci both in Old and Renaissance Italian, only in Old Italian can the vP periphery be activated in the same way.A further point of distinction between the two diachronic stages is that Fin loses (*) in Renaissance Italian, while v doesn't.Following Chomsky (2001), I take active vPs to be phases because they are phi-complete, i.e. v* encodes both [Person] and [Number] features, which are checked against merger of an EA.By contrast, inactive v does not encode [Person], but only [Number], and, being phi-incomplete, it is not a phase (see also fn.21 above).Notice that 3rd person is typically interpreted as [-Person].As a consequence, one would expect an active v* that probes for a 3rd Person (=[-Person]) EA to be equivalent to inactive v, which does not encode any [Person] feature.However, we have seen that this is not the case: agentive subject behave differently from inactive ones, at least when they are [masculine].As a generalization to these observations I conclude that in Old and Renaissance Italian vP is a phase if it respects both the following conditions.( 30 It has been proposed for Old Italian that prototypical Agents are masculine, whereas neuter or feminine subjects tend to pattern with inactive subjects, (see Parry 2007).The difference between masculine and feminine/neuter arguments feeds the expectation that feminine/neuter 3rd person subjects should behave like inactive arguments, because they lack [Person] and they are not prototypically agentive.This seems indeed to be the case, see property (i) above, and discussion in section 3.3.
I would like to propose that the active vs. inactive distinction of Renaissance Italian depends on the fact that active v has not lost its property of encoding strong features, i.e. active v is v*.This structural difference plays a role at the interface: the [+agent] EA in Specv*P is visible in the search space of the higher phase head, and it remains active for probing operations from above.
Capitalizing on Camacho's (2003) observation on the recoverability of subject features (see section 3.3), I propose that the (phi-)features of arguments that are merged in Specv*P (where v*P is a phase) must be recoverable by the end of the next phase (CP).Although the reasons behind the existence of such recoverability conditions are still unclear, it is a crosslinguistic fact that agentive EA arguments, and, more generally, subjects often pattern differently from objects, and sometimes from non-active subjects (see subject/object asymmetries in e.g.subject extractions).I suggest that these differences may depend on the fact that the features that are encoded on the lower phase head (v*), thus [Person] and [Agent], are relevant for deixis and, as such, they need to be recovered by the end of the main phase (CP), in order to be interpreted.In the specific case of subject extractions, it has been argued that [Person] features are directly interpreted on C (cf. Rizzi & Shlonsky 2007), I further suggest that agentivity may as well be subject to such interpretive requirements on C. 23 23 This intuition is based on the properties of subject/object asymmetries in Old and Renaissance Italian, and, specifically, on the fact that agentive subjects seem to have a special status with respect to extractability.Further investigations are needed in order to formalize this idea, but I will leave this future research.For a more detailed discussion see section 3.3.
As a consequence, the EA in Specv*P is subject to some recoverability conditions that do not affect inactive (i.e.non-agentive) arguments, which are merged in a structurally lower position.This difference contributes to distinguish active predicate structures from inactive ones, whereby only the first ones are (vP) phases, in line with Chomsky (2000), who argues that only transitive and unergative vPs, but not unaccusative and passive vPs are phases.Along these lines, the difference between phases and non-phases depends on the theta-role of the argument(s) that the verb selects.Specifically, only the presence of [+agent] arguments indicates that the vP is a phase. 24 Recall that in Renaissance Italian, Fin is no longer (*) (cf. Figure 2).In Renaissance Italian morphological visibility is nonetheless required on Fin for recoverability reasons (see section 3.3 below), whenever a morphologically prominent agentive EA is first merged in vP.The sense of 'morphologically prominent' will become clear below and in section 3.3.
In contrast to active predicates, inactive predicates do not take any agentive EA.Inactive predicates lack an impenetrable phase-edge (these vPs are not phases, cf.above).The minimal feature specification of v in inactive predicates (i.e.just [Number]) does not correlate with any morphological recoverability requirement imposed on the higher phase-edge.
Put differently, when the higher phase head, Fin, probes down, it is sensitive to the presence of [Person] and/or [Agent] features on v* as to a requirement of overt morphological realization that is necessary for recoverability.This means that the lexical material that is merged on the vP-phase edge must receive a morphological realization by the end of the higher phase, otherwise the reference of the extracted argument cannot be recovered, cf. the discussion in section 3.3.
For this reason, extraction of prototypically agentive EAs, but not of nonprominent (i.e.[-masculine; -human] and/or structurally lower) arguments, requires a lexicalization of FinP.In Renaissance Italian relative clauses, this asymmetry is visible in C-omission: the recoverability requirement that is imposed on FinP for the extraction of agentive EAs is formally satisfied in the morphology by merging an overt C-head (che), whereas this requirement does not hold for inactive subjects, thus C-omission can apply in this case, as is illustrated in (i), (ii), ( 27) and ( 28) above. 25In this sense, the 0-1 alternation that corresponds to the possibility vs. the impossibility of C-omission in Renaissance Italian mirrors the alternation that is attested between que/che and qui/chi forms in other Old Italo-Romance varieties (cf.above). 26Which morphological possibility 24 I agree with Gianollo (2010) in that verba dicendi, despite looking like unergatives, do not take a [+agent] EA.The subject is, in this case, a simple causer, and this is not sufficient to make the vP a phase.

25
There are basically two syntactic contexts for the extraction inactive arguments: either the structure is inactive, in which case recoverability is possible without imposing further morphological requirements on FinP (the closest phase edge), or the structure is active and a non-subject is extracted.In the latter case, a full vPphase is projected but the vP-phase edge constitutes an escape hatch for whmovement (cf.Grohmann 2003, Abels 2013 a.o.).a language adopts, as well as whether a language does or does not display such alternation, is a matter of variation.
In sum, in this subsection I have argued that Renaissance Italian undergoes a weakening that invests the properties of several functional heads that are merged in CP and vP peripheries, but while the CP phase-head Fin* loses (*), v* doesn't.The loss of (*) on Fin opens up the possibility of C-omission (i.e. of lack of morphological merger on Fin) in several clause types, a possibility which is however subject to an asymmetry in argument extractions.This asymmetry is explained by the requirement that arguments merged in a phase-edge position, i.e. agentive EAs in Specv*P, must be recoverable at the higher phase edge, i.e. on CP, before being sent to spell-out.This requirement imposes a morphological visibility condition on Fin, which must be spelled out in case of A/SA extraction.
In the next section I deal with an issue that concerns the relation between the C-omission properties (i) and (ii) above.A first hypothesis is that an argument is interpreted as [+agentive] if it is (lexically or morphologically) [+masculine; +human], and it is merged in Specv*P, where it arguably values and checks a [uAgent] feature.All other cases are interpreted as 'non prominent'.The latter should be cases in which the argument is [-masculine; -human], because it is feminine or neuter, and/or inanimate, and it is not merged in Specv*P, because of predicate structure.However, this restriction seems to be too strong to account for the facts, cf. ( 23).
A second hypothesis is that either property (i) or property (ii) must hold, however this automatically excludes the possibility that [+feminine; +human] arguments are agentive, and, as such, impose C-realization.Both these hypotheses, together with a third, related option, are discussed in greater detail in the next subsection.

A note on the recoverability of extracted arguments
In Renaissance Italian, the loss of (*) in CP has a number of syntactic consequences, among which the fact that null subjects are no longer licensed as either null topics or by morphological spell-out of the respective inflectional phifeatures on the verb, under V-to-C, as happens in Old Italian.In Renaissance Italian, the loss of a morphological realization requirement on CP coincides with the possibility for the subject [Person] features to be recovered via the inflection morphology on the verb in IP.Recall that nominal deixis, which formalizes the recoverability requirement for arguments at the phase-edge, is encoded in FinP (cf.section 3.1), which is the structural position onto which [Person] features are interpreted.
Following Camacho (2013:96ff.), who, in turn, capitalizes on Cole (2009), I assume that various mechanisms may contribute to the recoverability of subjects depending on the language.Specifically, Cole (2009) and Camacho (2013) propose a scale according to which "recoverability first resorts to morphological identification, then to identification by antecedent and finally by inserting an overt pronoun" (Camacho 2013:96).This means that some languages that do not resort to morphological identification may allow null subjects via contextual recovery by means of an antecedent, e.g. when an antecedent is given in the context or the subject is the topic, in which case a null subject may result from topic-drop.This seems to be what happens in certain root clauses in languages with V-to-Fin, such as Germanic (V2) languages and also Old Italian.Notice, however, that null subjects are not only attested in Old Italian root clauses, but also in subordinate clauses, i.e. also in absence of V-to-C.Specifically, Old Italian permits subject pro-drop with 3rd Person subjects in embedded clauses (Benincà 1984), which can be attributed to the fact that [3Person] is actually [-Person] for deixis and, as such, it does not require morphological visibility for interpretation on Fin*P. 27 Once the Italian system loses the morphological visibility requirement associated with (*) on Fin, which is satisfied by V-to-C, the recoverability of phi-features becomes possible via the morphological visibility of verbal inflectional features, which are no longer required to be spelled-out in CP.This is what happens in Renaissance Italian, which has no longer V-to-C because (*) is no longer associated with the Fin head. 28 Camacho (2013:97) observes that languages differ with respect to the type of morphological information that they require in order to identify a null subject.He specifically refers to a Minimal Morphological Threshold (MMT) as to "the minimal set of values overtly encoded in the morphology that a language requires to identify a null subject".Accordingly, he shows that some languages require only [Person], among the phi-features, whereas other languages also require [Number] and [Gender].
In line with Harley and Ritter (2002) and Béjar (2003), Camacho (2013) further assumes that nominal features are hierarchically ranked, as is illustrated in figure 3 below.For the present purposes I will just concentrate on the distinction that concerns 3rd Person referents, which are the arguments that are typically extracted in relative clauses.27 Some instances of null [1/2Pn] are equally attested, and these are all cases in which recoverability is guaranteed by the presence of an antecedent in the discourse, i.e. they are cases of topic-drop.This is just the result of a first analysis, but a more systematic study of the recoverability conditions for Old Italian null subjects is needed.I do not discuss the issue further at this point, since it is not directly pertinent to the analysis of C-omission in Renaissance Italian, but see also Vanelli (1987) and Poletto (2014:15, fn.15).

28
Renaissance Italian presents a higher rate of overt pronominal subjects than Modern Italian (cf.Franco, 2015).It seems, in this respect, that before morphological identification becomes the standard strategy of recoverability (cf.Modern Italian), Renaissance Italian still makes use of overt pronoun insertion.Again, the data available at this point are not sufficient for a full-fledged analysis of the recoverability of null subjects in Renaissance Italian, so this observation remains speculative.As was sketched at the end of section 3.2, a first hypothesis it that [masculine] is the semantically prominent 30 form for class, according to which we can derive the [masculine] vs. [feminine]/[neuter/-animate] opposition that Parry (2005) identifies in Old Italo-Romance varieties.This would in turn mean that the MMT for the identification of a [-Person] referent in these varieties includes gender features, or at least class features, as seems to be confirmed by agreement patterns, which are still visible in Modern Italian and several dialects of the Italian peninsula (cf.Bentley 2006 and ref. therein, a.o.).If [masculine] is indeed the prominent form, the recoverability of a [masculine] SA/A in Renaissance Italian will require morphological visibility at the highest phase (cf.section 3.2), whereas such visibility is not imposed to [-masculine] SA/As or SO/Os, since nonprominent features are recoverable without requiring overt morphology.This 29 Recall that Benincà & Cinque (2010) talk about a [+human]/[-animate] opposition for the lexicalization of C-forms, but such opposition is not specified in Harley & Ritter's hierarchy.This is not a big issue, since the only pronominal C-forms that lexicalize exclusively [+human] referents (and not [-animate] ones) are chi and cui in headless relatives and interrogatives.Clearly these clauses are not syntactic contexts where a [-human; +animate] referent, such as an animal, could be felicitous.30 I refer to prominence, instead of markedness, because the latter term is normally used in morphology to indicate the opposite distinction: masculine gender is typically unmarked in Romance morphology.I thank Imme Kuchenbrandt (p.c.) for this observation.means that, in case of extraction of a [masculine] SA/A, FinP must receive morphological realization, which is done by spelling out the Fin head with a Cform.Notice that this morphological realization of Fin is thus not imposed by the feature make-up of Fin (no (*) imposes lexical Merge after the Fin*>Fin shift in Renaissance Italian), but by a recoverability requirement on prominent arguments.
At this point a legitimate question is: why is verbal inflection not enough for recoverability, if null subjects are already licensed in absence of V-to-C in Renaissance Italian?A relevant observation in this respect is that C-omission and null-subject licensing via inflectional morphology coexist in the Italian system only for a relatively limited period of time.It is possible that at the time in which C-omission in relative clauses is productive, null subjects are not yet fully recoverable by means of the information on the verbal inflectional morphology, as is visible in the still high frequency of overt subjects pronouns in Renaissance Italian texts (cf.fn.28 above).This would in turn mean that verbal inflectional morphology alone is not yet sufficient to guarantee the recoverability of SA/As, and more specifically of extracted (prominent) subjects.
Notice that the morphological visibility requirement imposing the spell-out of Fin does not apply if the extracted argument is merged in a lower position in the thematic structure, i.e. it does not apply to SO/Os (cf.also 23 above).In (31) the subject is masculine but crucially not agentive.
(31) Uno giovane ___ istava con meser Gianozzo Maneti.A boy stayed with mister G. M. "A boy who was with Mr. Gianozzo Maneti."(VBV, 88, 1,92) Another important point is that verbal morphology, while expressing phi-and TMA features, does not directly lexicalize inner aspect features.This means that agentivity is not coded on verbal inflectional morphology, but can only be inferred from the verb semantics, with the sole exception of past participle agreement, which is triggered with raised Os or SO subjects.Agentivity is thus inferred from the lack of agreement on the past participle, but since no morphological marking permits an active vs. inactive distinction in non-perfective contexts, the visibility requirement on Fin in active contexts cannot be guaranteed solely by verb inflection.
Under a first hypothesis, referents bearing non-prominent features, such as [feminine] or [neuter/-animate], should in general not require morphological realization, by contrast to [masculine] SA/As.This possibility is based on the fact that a) [neuter/-animate] arguments are not agentive, since they cannot perform any intentional action; b) [feminine] arguments are frequently inactive (as is the case for abstract nouns, cf.faccenda, in (18a) above).
Albeit female protagonists are generally few (cf.Parry 2005:217), it is nonetheless plausible that [feminine] agentive arguments exist.In the analyzed Renaissance corpora, no instances of [feminine] SA/A for subject relative clauses were found.For this reason it is not possible to determine whether structuralsemantic properties, which is instantiated by merger of the agentive argument in vP, override morpho-semantic information, namely the different prominence between [masculine] and [feminine]/[-animate/neuter], in requiring a spell-out of FinP.Put differently, due to the lack of relevant data (i.e.behavior of [feminine] SA/As), a potential ranking of the conditions i) and ii) for C-omission (cf.section 3.2) seems doomed to remain undetermined for now.
A second hypothesis is that [feminine] SA/As pattern with [masculine] ones and thus require spell-out of the C-form in Fin.This would mean that what ultimately matters for C-omission is the structural active vs. inactive distinction, rather than a Gender-based distinction.By contrast, if [feminine] SA/As were interpreted as non-prominent and patterned with [feminine]/[neuter/inanimate] SO/Os (first hypothesis, above), this would mean that Gender also plays a role, and properties (i) and (ii) above would stand in an "either/or" relation.
A third option is that [feminine] is never agentive, i.e.
[feminine] SA/As are completely unattested.This third possibility is immediately disconfirmed by facts.Despite not being able to find relevant cases of [feminine] SA/As in subject extractions, I could find (not frequent 31 , but indeed existent) cases of [feminine] SA/A in simple clauses, as the active predicate in bold in ( 32)-( 36) indicates. 32 The examples ( 32)-( 36) show cases in which a female subject is agentive.Interestingly, this applies also to non-human subjects as in ( 36). ( 32 And you should know that the partridge makes her nest with thorns and small sticks, and moves often her baby birds to another place for fear of her male.'(Tesoro volg. 5,31) These facts confirm the hypothesis that female agentive subjects exist, which means that Specv*P may host [feminine] (and even [-human], cf. ( 36)) SA/As.These facts are however not sufficient to understand whether [feminine] and [masculine] SA/As pattern alike in extraction contexts, i.e. whether they are indeed subject to the same recoverability conditions.On the one hand, the possibility that [feminine] SA/As pattern with [feminine]/[neuter/inanimate] SO/Os for recoverability in extraction contexts is still open, which entails that Gender marking (cf.i) in section 3.2) is not subordinate to agentivity.On the other hand, an asymmetry between [masculine] SA/As -SO/Os and [feminine] SA/As -SO/Os with respect to extractions is unexpected, given that [feminine] SA/As should be also merged in Specv*P, according to their theta role.
Moreover, C-omission in examples like ( 27) and ( 28) shows that [masculine] alone is clearly not sufficient for imposing a morphologic visibility condition on Fin, because the extracted subjects in these examples are non-active (SOs) and, as such, the SO antecedents are recoverable without morphological spell-out of FinP.This amounts to say that, if Gender does play a role, it is anyway subordinate to agentivity.The same SA/A -SO/O asymmetry can be reasonably expected for [feminine] arguments.
A tentative conclusion, given the present state of knowledge, is thus that property (ii) overrides property (i), and the extent to which (i) is relevant requires further empirical investigation.This is summarized in (37) below.[SA/A]-[feminine] = ?c. [ From a broader perspective, I will just limit my observation to a well-documented fact, which is however not yet entirely clear from an explanatory viewpoint, namely to the fact that [+agentive] is commonly associated with [+masculine] in many Old Italo-Romance varieties (Parry 2005).This correlation finds empirical support in the morphology of the C-forms that is attested in texts across the Italian Peninsula.In various Old Italo-Romance varieties, che lexicalizes feminine (plural) antecedents, in contrast to chi (and analogous forms), which is used for masculine referents (Parry 2005:209).Notice that such a gender distinction seems to be absent in Old French, according to the descriptive literature.This remains to a large extent an unexplored field, for the moment.

Comparative facts
In this section I offer a comparative discussion of C-omission in Old French and Old Occitan relative clauses (section 4.1), showing that these languages share the same properties of Renaissance Italian, with respect to C-omission.Section 4.2 briefly illustrates the situation in Old Ibero-Romance (Old Spanish and Classical Portuguese), which instead does not allow for C-omission in relative clauses.I attribute this distinction to a different feature make-up of the C-form that is employed in these languages.

C-omission in Old French and Old Occitan relative clauses
Next to the characteristic features of active vs. inactive distinction, which pervades both the nominal, the verbal and the sentence domain, the old dialects of Northern Romània ("coinciding with the historical areas of Gallia transalpina (northern Gaul: langue d'oïl, southern Gaul: langue d'oc), Gallia cisalpina (northern Italian dialects), and Raetia (Raeto-Romance varieties)", Ledgeway (2012:314)) display a morphosyntax that reflects a bipartite case marking (NOM vs. ACC/OBLIQUE).On a par with Old North Western Italian varieties (cf.section 3.2), Old Occitan and Old French present various forms to introduce relative clauses: que, qui and cui (for obliques) (Bourciez 1967, Jensen 1986, 1990).Ledgeway (2012:336) argues that the qui/que distinction of Old Romance originated from the Latin NOM/ACC marking (cf.section 3.2), but began to be associated to an active vs. inactive opposition.Subject relative clauses are marked by qui when the subject is high in the animacy hierarchy, and typically displays agentivity features (it is human, dynamic, etc.).By contrast, que, even when it is used for human antecedents, generally denotes a non-controlled event or a state with a non-agentive subject.These general observations seem to partially contrast with the traditional descriptions of Old Occitan and Old French (Jensen 1986(Jensen :139ff, 1990)), according to which "both qui and que are used indifferently about persons or things" (Jensen 1986:141, cf. Jensen 1990:203).This is apparently not expected if the distribution of qui is determined by the animacy of the extracted argument.Jensen (1986) bases his observations on the attested usage of qui for inanimate antecedents, (38), already in old texts (pace Grafström 1968).
Une parole qui avant hier me fut dite.A word qui before yesterday to.me was said 'A word that was said to me the day before yesterday.'(Jensen 1990:203, Queste 53.33)Moreover, Jensen (1986Jensen ( , 1990) observes some differences between Old French and Old Occitan.There is a tendency, in Occitan, to use que regardless its syntactic function, thus also in subject extractions.
(39) Chascus hom que son gen cors ve.Every man que her beautiful body sees 'Every man who sees her beautiful body.' (Jensen 1986:140, Uc de Saint Circ III 19) However, among all the Old French and Old Occitan examples that Jensen provides, I could not find any case in which que is used for an SA/A antecedent, whereas qui is adopted both for agentive and non-agentive subjects (cf.(32) above). 33This intricate morphosyntactic situation seems to result from the partial overlap between an active vs. inactive opposition and a NOM/ACC marking, as Ledgeway ( 2012) also suggests for other Northern Romance languages (cf.above).I tentatively analyze these facts by assuming that Old French and Old Occitan present a (microparametrically different) mixed system.While qui preserves morphologically nominative case, que is unmarked for case and, as such, it is typically adopted for SO/O antecedents.This may account for the usage of qui also with inactive/non-human antecedents inasmuch as they are nominative, which gives rise to a mixed system.The distribution of qui thus parallels that of the over C-form in Renaissance Italian, which is possible with inactive antecedents as well.Notice moreover that Old French/Old Occitan relative que is syncretic with the generic subordinator que of declarative clauses and several adverbial clauses (Jensen 1990:477), on a par with Italian che.Jensen (1986:362;1990:497) analyzes C-omission in relative clauses of Old French and Old Occitan as parataxis.For both French and Occitan, Jensen (1986Jensen ( , 1990) ) observes that C-omission is most frequent in noun clauses, "which means that the conjunction most often omitted is the semantically insignificant que" (Jensen 1990:497).In Old Occitan, C-omission usually affects subject relative clauses, (40a), but is also attested in object relative clauses (40b).Similarly, Old French displays a subject-object asymmetry for C-omission that is apparently the reverse of what is attested in Renaissance Florentine.That is, "it is mostly the pronoun serving in subject function that may be omitted […, cf. (41a)].It is less common for the dative cui or the accusative que to be left unexpressed, [cf. (41b)]" (ibid.).A more attentive observation of the following examples reveals that this is not quite an appropriate picture.

33
The verb 'to see' in example (39) does not constitute a counterexample, since the thematic role of subjects of perception verbs in their 'basic' form (hear, see, etc., although probably not 'eardrop' or 'stare') is semantically more similar to that of experiencers rather than agents.The notion of agentivity seems to be tightly connected to that of intentionality, but a more detailed discussion of this point is outside the scope of this paper.As examples ( 40)-( 41) show, C-omission is "the norm" (Jensen 1990:498) when both the matrix and the relative clause are negated, which in fact results in an affirmative interpretation (e.g.(40b)= 'Love teaches everything'; (41a) = 'All his limbs hurt').Moreover, a closer look at the predicates of the relative clauses in these examples reveals that C-omission always follows an inactive antecedent, which is confirmed by further data reported in Jensen (1986Jensen ( ), (1990)), here omitted for space reasons.I can thus conclude that the active vs. inactive distinction also plays a role in C-omission in relative clauses of Old French and Old Occitan, being attested with inactive antecedents, on a par with C-omission in Renaissance Florentine.
As for the presence of expletive negation in clauses with C-omission, notice that these clauses receive a modal interpretation as typical instances of a [realis] situation or event, interpretation that results from anchoring the state/event to the discourse context (cf.section 3.1).The analysis of Belletti (2001) for the presence of expletive negation in Italian (and Old French) comparative clauses is thus directly applicable to the relative clauses in (40), (41) (also Old and Renaissance Italian display expletive negation in the same contexts).Belletti proposes that the presence of an expletive negative head (e.g.no, non, in Old Occitan, ne, in Old French) is licensed by a modal [-realis] OP, such as the one encoding subjunctive mood, which moves in the Spec of the NegP projection.The Spec-Head agreement with a [-realis] OP (rather than with a true Neg OP) yields the expletive [-realis] interpretation on the negation.It is also worth mentioning that C-omission in Old French and Old Occitan, let aside relative clauses, is most frequently attested with embedded subjunctive predicates or, regardless of mood, in complements of semifactives and propositional attitude verbs (Jensen 1990, Scorretti 1991).All these syntactic contexts seem to share the absence of embedded independent illocution.This explains why C-omission is possible: on the one hand, illocution on ForceP is unmarked, so it does not require overt spellout of illocution features.On the other hand, nominal and spatio-temporal deixis, which are encoded on FinP, can be recovered morphologically by means of functional elements (e.g.verbal morphology).Lack of illocution also characterizes relative clauses, in which C-omission is further subject to the condition that the antecedent be inactive, thus non prominent, for recoverability purposes, on a par with Renaissance Florentine (cf.discussion in section 3.3).
In conclusion of this subsection I make a few remarks on the setting of the V2 property in Old French and Old Occitan.Old French has also a Romance-type V2 (Vanelli, Renzi, Benincà 1985, Adams 1987, Roberts 1993, 2005 and ref. therein, a.o., cf.section 2.1 above for Old Italian), and null subjects are only possible in Old French if the subject is postverbal (Foulet 1928), namely if V-to-C occurs.This seems at odds with the properties of Renaissance Italian, which has lost V2.Notice, however, that C-omission, at least in Old French, occurs in typical non-V2 clauses, such as clauses with unmarked illocution (cf.above), where namely no V-to-C takes place.The absence of V-to-C in embedded contexts is related to the absence of (*) on the unique C head, where both Fin and Force features are conflated.Put differently, it seems that C-omission affects embedded clauses that lack illocutive force and thus that do not have V-to-C.Moreover, Vance et al. (2010) attribute the loss of V2 in Old French to the progressive increase of SV orders in main clauses that are preceded by a fronted adverbial clause.This happens already in the course of the 13th century, which seems to support the hypothesis that V-to-C and C-omission are in complementary distribution, because it is mostly around this time that C-omission is attested.By contrast, the evolution of V2 in Old Occitan cannot be detected in the same way, given the higher frequency of null subjects in this language (Lafont 1967, Vanelli, Renzi, Benincà 1985, Sitaridou 2005, Vance 1997, Vance et al. 2010).At this point the possible correlation between a V2 property (determined by (*) in CP, in the present analysis) and the productivity of C-omission in Old French and Old Occitan requires further investigations.More specifically, the micro-parametric differences with respect to Old Italian V2 have to be identified in order to understand what the possible correlation between absence/presence of V2 and C-omission can be, in Old French and Old Occitan.I leave this issue open for future research.

Old Spanish and Old Portuguese
Both Old Spanish and Old Portuguese display C-omission in declarative complements (Lleal 2000, Girón & José 2005:879;Scorretti 1991) However, all the cases Elvira (2005) illustrates are free relative clauses, in which qui is thus clearly a pronoun.This is confirmed by the fact that qui is also sometimes possible after a preposition (Elvira 2005:fn.17)and that "the compound relative el qui was also possible, with the same preference for the subject function" (Elvira 2005:465, translation mine).In the latter case, qui is associated to a definite article.The same pronominal property is shared by the relative form que, which is often associated to prepositions like de (of).The form que seems to be the most common one for introducing relative clauses both in Spanish and Portuguese already in the late Middle Ages, whereas qui is only attested in older texts (cf.Dulce de Faria (1988) and ref. therein).These facts suggest a difference in the encoding of the active vs. inactive distinction in Old Spanish and Old Portuguese, at least with respect to the 34 Thus C-omission seems to be restricted to declarative que, as in (i) and (42) above.(i) Pensaron ___ era alguno de los suyos.thought.3PLwas someone of the theirs 'They thought he was one of their men.' (Luna, 120b in Hanssen 1945:275) morphological requirements imposed at the CP level.This does not imply that the distinction was utterly absent from the Old Spanish and Old Portuguese systems, but arguably that it was marked with a different strategy than distinct Cmorphology (as in Northern Italian vernaculars) or C-omission (as in Renaissance Italian).In the present analysis, any strategy that permits the recovery of the antecedent features is, in principle, possible.The pronominal property of que in Old Ibero-Romance and the specification of the thematic role of the antecedent by means of e.g.prepositions are, in this perspective, a way to satisfy the recoverability requirement on a par with C-omission in Renaissance Italian.

Diachronic change in Italian
In this section I return to the Italian system and I try to provide a more complete picture of the diachronic changes that affect the conditions for C-omission, in light of what has been proposed above.
After the Renaissance period, Italian C-omission decreases drastically and eventually disappears, with the exception of few syntactic contexts (cf.section 1).I have argued that the reason why C-omission becomes productive in Renaissance Florentine is the change from (*) in CP and more specifically from Fin*, which requires overt spell-out, to Fin, which permits absence of an overt C-head.The problem is that widespread C-omission is not an option at later stages, which also do not have (*) in CP.That is, C-omission in (subject) relative clauses is ungrammatical in the modern counterparts.
From a diachronic perspective, there are two issues to be addressed: a) Why is C-omission possible in Renaissance Italian, whereas it is restricted to specific (modal) contexts in Modern Italian and Modern Florentine, given that these languages all have lost [*] on CP? b) More specifically, why is C-omission possible in relative clauses in Renaissance Florentine but it is no longer permitted in Modern Italian and Modern Florentine?
As an answer to a) I suggest that while Fin* requires a morphological realization (as in Old Italian), Fin does not, but of course nothing rules out a morphological spell-out, as for the overt C cases in Renaissance Italian.My intuition in this respect is that C-omission in Renaissance Italian is symptomatic of the ongoing Fin*>Fin change.The absence of (*) in CP offers the possibility for C-omission, but C-insertion becomes the default choice, on the basis of a higher input frequency of overt C contexts, in comparison to C-omission contexts.It is not clear whether other pragmatic or sociolinguistic factors also play a relevant role for the type of input generating diachronic change, but this is plausible, given that C-omission is permitted in a greater number of syntactic contexts in Modern Florentine (a dialect), with respect to Modern Italian.
Complementizers are thus merged only for subordinating purposes in Modern Italian and Modern Florentine (cf.Rizzi & Shlonsky 2007).Broadly speaking, complementizers can only be omitted if the interpretation of the clause as a subordinate is guaranteed, e.g. by means of morphological marking (for instance, with subjunctive morphology; with a sentential negation in the complement position of a predicate that selects a sentential complement, such as dire (="say"), cf.section 1 for Modern Florentine), or in absolutive constructions (cf.Poletto 1995, a.o.)). 35 The answer to b) follows straightforwardly from the analysis presented in section 3, according to which C-omission in relative clauses depends on the combination of two factors: i) an active vs. inactive distinction in the argument structure, whereby C-omission is possible in the presence of inactive and morphologically non-prominent (i.e.[-masculine]) antecedents, and ii) syncretism between the relative C-form and the declarative complementizer (ke, che, que), which are both unmarked for case.In Modern Italian and Modern Florentine, we assist at a fall of structural marking of active vs. inactive distinction, so factor i) no longer applies.As a consequence, the [Agent] feature encoded on the v head no longer requires a morphological spell-out.The loss of an active vs. inactive structural distinction thus provokes a change in the recoverability conditions of (extracted) arguments.My hypothesis is that due to the loss of the visibility requirement on v (the low phase-head) the identification of active vs. inactive relative-clause antecendents via a formal, structural marking on Fin, i.e. at the end of the phase, is no longer permitted.
Given the loss of (*) in both CP and vP in Modern Italian and Modern Florentine, the question now is why is C-omission not licensed in all argument extractions in these languages?I suggest that the obligatory spell-out of C in relative clauses generally depends on the obligatoriness of C as a complementizer in other types of subordinate clauses, at this stage (cf.above).That is, C-insertion is no longer related to an active vs. inactive distinction, and no morphosyntactic requirement imposes a marking of the NOM/ACC distinction at the CP-phase edge, given the properties of Modern Italian and Modern Florentine ('weak' v, 'weak' Fin).

Summary and conclusion
On the basis of the collected data, I have proposed that widespread C-omission in Renaissance Italian results from the coexistence of the following conditions: a) Loss of (*) in CP; b) Morphological recoverability of [Person] features of the arguments via longdistance agreement; c) Presence of an active vs. inactive distinction, where inactive is the unmarked option.
35 Poletto (1995) shows that in case of C-omission in Modern Italian the subjunctive verb raises to the CP domain, where she argues that it checks a [-realis] feature encoded on C. It seems that C-omission is not possible in contexts that have independent illocution, such as indicative complements of 'bridge verbs' (vs.prepositional attitude complements).If that is the case, C-insertion could be a sort of default Force marking.As the focus of this paper is mainly a diachronic analysis, I do not speculate further on the feature-checking mechanism permitting C-omission or imposing C-insertion in Modern Italian, which is a topic that deserves further investigations.See a.o.Llinàs-Grau & Fernández-Sánchez (2011) for a proposal, and references therein.
Condition a) is relevant for C-omission in all clauses; conditions b) and c) for omission in A-bar extractions of arguments.All three conditions are met in Renaissance Italian, by contrast to Old Italian (which still has (*) in CP and in which pro-drop is dependent on V-to-C*), or Modern Italian and Modern Florentine (in which the active vs. inactive distinction marking in CP is lost).
In Renaissance Italian, spelling out C is no longer required, but it is permitted by weak Fin.As a result of an active vs. inactive distinction, inactive arguments do not require C spell-out for recoverability, when extracted.This proposal predicts that the following restrictions should apply to C-omission in Renaissance Italian: (i) C-omission is unattested (=ungrammatical) in headless relative and interrogative clauses in which C is a pronoun and has a [+human, SA/A] value, cf.Table 3.; (ii) C-omission is unattested in headed relative clauses in which the extracted argument is [+human, SA/A], for the reasons discussed in sections 3.2, 3.3.
As it has been discussed in sections 2.2 and 3.2, predictions (i) and (ii) are borne out by facts.Nonetheless, further research needs to be done in order to understand what are the potential differences among Old Romance languages in relation to Comission.

(
29) [(Topic1…Topic2…Topic3… OP/Focus…) vP EA[Agent] v*… InAsp…[V…]] Phenomena like WCC, VP ellipsis, object/XP-low scrambling, and asymmetry in stylistic fronting in non-root clauses (cf.Franco 2009, under review) involve movement to some vP-peripheral projection in Old Italian (e.g.Topic or Focus above vP, see Franco et al. 2014), but these phenomena are no longer productive in Renaissance Italian.I take this change to indicate that (*) is progressively lost on the Topic/Focus/OP heads of the low left periphery, in a parallel fashion to what happens at the CP level (cf.Poletto 2014).Nonetheless Renaissance Italian still distinguishes active and inactive structures at a morphosyntactic level in a way that is clarified below.

Table 2 :
Troya 47.35)This variation indicates that the active vs. inactive distinction still takes place in a NOM/ACC system.Put differently, also nominative inactive subjects may correspond to a qui form (rather than a que form).This overlap would be unexpected if these Medieval varieties underwent a real ergativity-based realignme Differently from Old North Western vernaculars and Old Neapolitan in which the A/SA vs. O/SO alternation is morphologically marked with qui/chi vs. que/che respectively, Old and Renaissance Italian and other Tuscan varieties display no morphologically distinct C-forms for A/SAs of headed relative clauses, which would correspond to the form qui/chi of those varieties.Benincà & Cinque (2010) distinguish the various forms that are attested in Old and Renaissance Italian on the basis of the semantic features [+/-human], [+/animate], and observe that only the extremes are morphologically realized, namely C-forms realize either[+human]or [-animate].The various forms that are attested in Old and Renaissance Italian are thus chi, che and cui similarly to other Italic vernaculars, but they have a different feature specification, as is reported in table 3 below (cf.Benincà & Cinque 2010:437, table 12.1).

Table 3 :
Relative subordination forms in Old and Renaissance Italian Old/Renaissance Florentine Put differently, 3rd person EAs, which are [-Person], can still specify a phase head, v*, when they are agentive.Which types of arguments can be agentive may be subject to crosslinguistic variation.

Figure 3: Hierarchical ranking of nominal features, from Harley & Ritter (2002:8).
gives a representation of the hierarchy of the features of referential expressions.Crucially, 3rd Person is 'non-Person', which corresponds to the fact that [-Person] referents are not participants in the discourse.Béjar (2009:49)remarks that "3rd persons subcategorize in more ways than can be represented by the binary contrast between participants and nonparticipants".Specifically, a first opposition within a class is given by the marking of animacy, which has been argued to play a role in the morphosyntax of C-forms in Old Italo-Romance (cf. Parry 2005, Benincà & Cinque 2010, above).If we apply the identification of Cforms of Benincà & Cinque (2010) to figure 3, [animate] should also include [+human]. 29Following Harley & Ritter's hierarchy, referents that are [+animate] further divide into [feminine] and [masculine].
And when Moses was born, the mother gently closed him into a basket and threw him in a flowing river.'(Tesorovolg., ed.Gaiter, 40,1)On a corpus search (OVI Gattoweb corpus) for donzella (maiden) and fanciulla (girl) I obtained 696 occurrences but only one S A /A.Of course, this must be related to the fact that the occurrences represents any type of arguments (even non-arguments), not only subjects.OVI Gattoweb online corpus, for Old and (early) Renaissance Italian.Cf.Sources, this paper. 32Source: Faria 1988, Morel Pinto 1988, Sant'Anna Martins 1988her C-omission was possible in relative clauses during the medieval an, Menendez Pidal 1973, Lleal 2000, Elvira 2005I have consulted various grammars and critical editions of old texts (Dulce deFaria 1988, Morel Pinto 1988, Sant'Anna Martins 1988, Spina 1988, Hamini Boainain 1989 for  Portuguese; Bolaño e Isla 1971, Menendez Pidal 1973, Lleal 2000, Elvira 2005, Girón & José 2005 for Spanish) and excerpted the texts contained therein.I could not find any record or trace of C-omission in either object or subject relative clauses.34Inthisrespect it is worth pointing out that a distinction in the argument function seems to be lexicalized at the complementizer level only at earlier stages of the grammar.Specifically, "some texts of the XIII century present the difficult surviving of an old relative pronoun qui, with a statistically prevalent subject use"(Elvira 2005: 464, translation mine).By contrast, "in the XIV century qui was already obsolete" (Menendez Pidal 1973: 263, translation mine).