PCC Effects with Expletives and NonAssociate Postverbal Subjects in Bolognese

This paper contrasts a Bolognese postverbal subject construction and other grammars with the common Romance one (also in Bolognese) that has long­distance full agreement of the tensed verb and the Case­Licensed subject, with an expletive satisfying EPP. In the new Bolognese data, full agreement is absent, a special clitic occurs, and the postverbal subject is person­restricted. Lack of subject agreement also raises questions about its licensing. The Minimalist proposal is that grammars like Bolognese can specify a feature set on the expletive that checks EPP in this data, and that it is thus an independent second nominal in the domain of the sole agreement and Case­Licensing probe, T. This specified expletive is shown to explain all the properties of this data. For the person­restrictions and Case­ Licensing of the postverbal subject, it applies Cyclic/Multiple Agree, the elaboration of Agree underlying PCC­effects, to the two nominals. The analysis is extended to other grammars with similar but slightly differing data by simple manipulation of the feature­ set on the specified expletive and of the clitic inventory of the grammar. 2 Isogloss 2022, 8(2)/6 Edward J. Rubin

Many Romance grammars permit postverbal subjects (pvSs), and an extensive literature on them has developed. Generally, these postverbal subjects control agreement on T and are CaseLicensed by it, while an expletive pro occupies the preverbal subject position only to satisfy EPP (Rizzi 1982, Burzio 1986, Cardinaletti 1997b, 2004, Belletti 2005, Roberts 2010, among many others). We refer to such data, with examples provided at the end of the section, as common Romance postverbal subject constructions (cRpvS). In Bolognese, the GalloItalic grammar of Bologna, Italy, there are such data, but there also exists a distinct postverbal construction, which we will call a nonagreeing person restricted postverbal subject construction (naprpvS) for reasons quickly to be made clear and to distinguish it from cRpvS. In naprpvS, the subject of unergative (1) or unaccusative (2) verbs may appear in a postverbal position, but the relevant data exhibit empirical differences and theoretical problems that distinguish naprpvS from cRpvS. This section documents these issues of the Bolognese data with discussion of important aspects of its analysis. (All Bolognese data were collected in close work with native consultants.) (1) a. Ai  The first notable issue of this Bolognese naprpvS data in (12) is that it contains an invariable clitic on the tensed verb (glossed 1 as AI 2 ). We will demonstrate that AI is not * For helpful discussions and suggestions, I thank my colleagues in my department, members of the audience at the Going Romance 34, and two anonymous reviewers. I am thankful to the University of Bologna for support during my fieldwork in Fall 2019. I am extremely grateful to my consultants for their patience and generosity in spirit and time. Any problems in this paper are mine alone. 1 Other glosses/abbreviations: Person (π) first (1), second (2), and third (3); Number (#) singular (S) and plural (P); Gender Feminine (F) and Masculine (M); Subject Clitics (SCL), Ac cusative Clitics (ACL), and Dative Clitics (DCL); and postverbal subject (pvS). 2 A complete morphological analysis of Bolognese or just its clitics is beyond the scope of this work. AI is glossed as a single element in this work to focus on the main points of discussion, which can ignore AI's internal complexity. In brief, it is clear that AI contains two pieces. The first, [a], is homophonous with the preposition a 'at, to' and with three of the Bolognese SCLs, the 1S, the 1P, and the 2P (see example (8) in the text). These may sound like vocalic SCLs as discussed in Poletto (2000), but they generally don't show the properties of this type (nor does AI). A full demonstration is not possible here, but these three Bolognese SCLs behave generally identically to the other SCLs (and AI behaves differently from them). The other part of AI is identical to the locative i, but bleached of its meaning. Suggestively, the community has established different spelling rules for AI (the two letters must be written together) and for a SCL followed by the locative one of the Bolognese Subject Clitics (SCLs), of which the 3SF and 3PF are seen in (34), the correspondents to (12) with preverbal thematic subjects. Bolognese is what Roberts (2010: 106) calls "a 'fully redundant' nullsubject system," i.e. one which distinguishes all six personnumber combinations via verbal suffixes together with SCLs. Bolognese also distinguishes gender in the third person via the relevant SCLs. (Some of the extensive morphophonological variation in particular Bolognese SCLs can be seen in (34).) In (34), the same thematic subjects as in (12) appear in a preverbal position, and the tensed verbs necessarily occur both with π and # identical to those of the subject and with SCLs that are also dependent on the subject's π and # (and Gender, when π = 3). If there instead appeared AI in any example in (34) The second notable issue of such naprpvS data as (12) is that the tensed verb always shows 3S agreement, regardless of the number of the postverbal subject (pvS), which can be S, as in (1a,2a), or P, as in (1b,2b). The pvSs in such data are 3, like the verbs, which could suggest some notion of partial agreement, essentially that πonly agreement might be operative. However, further facts involving a pvS.1 or a pvS.2 in structures like (12), shown in (56) below, demonstrate that no such notion can be adopted without significant elaboration. As seen in (5a,6a), pvS.1/2s are illformed with the V.3S seen in (12), which would be (must be written separately). Additional syntactic differences between AI and SCLs, including the homophonous ones, are discussed below. Despite the homophony, they are not one and the same element: AI does not contain a SCL (nor the P a 'at, to'). expected if partial agreement (in π only) is required. However, a pvS.1S/P is also ill formed with a V.1S that matches it in π (5b,6b), and a pvS.2S/P is also illformed with a V.2S that matches it in π (5c,6c), both contrary to an expectation that agreement in π is all that is required. Something more, or something else, must be said: Any analysis must account for this third notable issue, that first or second person pvSs and tensed verbs are impossible in the naprpvS construction. PvS.1/2s are possible in Bolognese, just not in naprpvS, only in cRpvS, as we see next.
The final issue in understanding naprpvS is best seen in comparing it to cRpvS, which occurs in many Romance grammars, including Italian (7) and even Bolognese itself (8). We compare the properties discussed above first, then turn to this last issue. In the naprpvS data in (12), we first noted the special Bolognese clitic AI, and we see that that clitic (or one like it) is lacking in cRpvS data including (78). (If such a clitic occurred anywhere in Romance in cRpvS data, it most likely would have been investigated as intently as other aspects of the data. But see the discussion of Fiorentino in the following section.) The Bolognese cRpvS examples in (8) do include SCLs, though they are not always available in other Romance varieties, e.g. Italian, thus explaining their absence in (7). Following Roberts (2010) and many others, we take SCLs to relate to subject agreement. In (8), thus, each SCL φagrees with the cooccurring pvS. Agreement was also central to our second issue above, when we noted that the naprpvS data in (12) showed 3S on the tensed verb, regardless of the features of the pvS. In contrast, in the wellformed cRpvS data in (78), we see that the tensed verbs show agreement morphology that matches each pvS. In Italian (7d), it doesn't, but instead bears 3S like in (1), and the example is illformed. This is evidence of the distinct nature of the two pvS constructions, on the assumption that Italian has grammatical elements that allow only cRpvS while Bolognese has some additional element(s), allowing both cRpvS and naprpvS. Finally, the third issue above, that explaining naprpvS requires an account of the impossibility of first and second person pvSs, clearly does not apply to cRpvS. Both Italian and Bolognese allow such pvSs in cRpvS, as shown in (7c) and throughout (8).
To clarify the fourth important issue in explaining naprpvS, we build on the stan dard analyses of cRpvS as involving an expletive pro (expl) in a preverbal position like the one occupied by a preverbal subject in data such as (34) (Rizzi 1982, Burzio 1986, Cardinaletti 1997b, 2004, Belletti 2005, Roberts 2010, among many others). In such data, a preverbal subject satisfies EPP, determines the agreement on the tensed verb, and is CaseLicensed, all effected via Agree holding between T and the subject in its Tposition, with movement of the subject to its final preverbal position (Chomsky 2008, etc). In stan dard analyses of cRpvS, on the other hand, expl occupies a preverbal postion and "shares" with the pvS the three properties of the preverbal subjects effected by Agree: expl satisfies EPP while pvS determines agreement on the tensed verb and is CaseLicensed by T, via longdistance Agree. Lasnik (1995) called this situation "Case Transmission." Because the pvS in naprpvS cannot satisfy EPP for the same reasons as the pvS in cRpvS, according to standard Minimalist approaches, we conclude that there also exists an expl in a similar preverbal position in naprpvS just as there is in cRpvS.
In naprpvS, however, determination of agreement by pvS and CaseLicensing of pvS cannot simultaneously occur by means of Agree as they do in instances of Case Trans mission in cRpvS. This is related to the second issue above: the pvS in naprpvS does not in fact determine agreement, as it does in cRpvS. When data in other grammars similar to (12) is mentioned in the context of other discussions, it is usually suggested, e.g. Bel letti (2005: 19) and Roberts (2010: 113), that the expl is what determines the agreement, i.e. that Agree(T,expl) values uφ on T. But, if so, then there arise issues raised by Lasnik (1995) in another context (and in earlier but still relevant formulations of the mechanisms). If Agree(T,expl) values uφ on T, then the pvS is not part of the Agree relation with T, and is thus not CaseLicensed as required. If, contrary to such an idea, Agree(T,pvS) holds in all such data, then the pvS is CaseLicensed, but it should also determine agreement on the tensed verb, as it does in cRpvS. Standard notions of Agree do not separate these two results, and doing so would be inappropriate in the Italian and Bolognese cRpvS seen in (78). The fourth issue in understanding naprpvS, then, is to explain why the valuation of uφ on T and the CaseLicensing of the pvS by T are apparently separated.

Previous Work and Further Issues
In this section, we compare Bolognese naprpvS data to past analyses of similar data from other grammars. First, we look at Brandi and Cordin's (1989) analysis of Fiorentino, followed by modern proposals that essentially adopt the same ideas using more recent theoretical mechanisms. Then we turn to Tortora's (1999) analysis of Borgomanerese, where a specific proposal for πrestrictions is found. Brandi and Cordin (1989) discuss data similar to (12) in Fiorentino, e.g. (9): telefonato telephoned delle some ragazze . girls 'Some girls telephoned.' Brandi andCordin (1989: 1213) describe this data as involving a verb in "an unmarked neutral form (third person masculine singular)" and a "neutral impersonal clitic ... strongly reminiscent of the French clitic il." They posit that pro occupies the subject position, and "the impersonal clitic of Fiorentino represents the spelling out of AGR features. ... the absence of a similar expletive agreement clitic in Trentino may simply be attributed to a gap in the morphological paradigm." They add that "the preverbal subject is expletive pro, which, like other pleonastic elements, may be expected to bear neutral features (third per son, masculine, singular); thus subject clitics and verbal agreement will also appear with neutral features in the inversion constructions." Brandi and Cordin (1989) includes data that demonstrate that the Fiorentino clitic in (9) is identical to the Fiorentino SCL.3SM that appears on tensed verbs with a preverbal 3SM subject.
Comparing this to our observations about (12), we find that Fiorentino and Bolog nese differ with regard to the first noted issue, in that Fiorentino shows a SCL.3SM (with "neutral" features) in these inversion structures and Bolognese shows AI. AI is not identi cal to (a)l, the Bolognese SCL.3SM that also shows these "neutral" features, and is not a SCL at all, as shown below. The Fiorentino data do however show the same second noted issue of requiring 3S agreement on the tensed verb as observed in (12). In a footnote, Brandi and Cordin (1989) mention data exemplifying πrestrictions like those in (56), the third issue concerning naprpvS noted above. They ascribe these πrestrictions to a mismatch in features between the expl in subject position, which is "always third person" (pg. 138), and a coindexed non3 pvS. They do not mention the CaseLicensing of pvS nor how the coindexation between the expl and the pvS, a notion presumably adopted in parallel with its use in CaseTransmission data like (78), might make possible a separation between CaseLicensing and the determination of agreement in (9), but not in (78) (which is the fourth issue).
This analysis remains essentially standard, even in works adopting significantly updated theoretical mechanisms. For example, Belletti (2005: 19) says: Languages may vary as to the status of 'pro' in the nominative position of inver sion structures. ... In those cases where 'pro' has an expletive status, and it is thus assimilated to French il type expletive, verbal agreement would not obtain with the postverbal subject. A possibility which is well known to occur in several lan guages/dialects. In that paper there is no mention of πrestrictions, which are therefore unaddressed. It also does not address how or why the separation of CaseLicensing of pvS and determination of agreement might occur. This analysis is a literal footnote to a discussion of agreement patterns in data like (78), in which a Case Transmission account is established by a "big DP" hypothesis in which the pro in the preverbal nominative position and the pvS start together and share identical features. That pro does not have the quoted expletive status. " [A]ccording to this proposal, nominative assignment to the postverbal position comes as a direct consequence" (pg. 18), for the commonly discussed cRpvS data like (78).
In the above quoted account, however, the treatment of data like (12, 9) comes with the important noted change, that an expletive pro stands in place of the one identical to pvS. As a result, just as in the discussion of Brandi and Cordin (1989) above, the fourth important theoretical issue concerning Agree arises again. In data like (12,9), something blocks the sharing of features that holds in the other data between a pro and a pvS within a big DP, so that, here, it is the 'expletive' pro that determines agreement on the tensed verb when it moves to the preverbal subject position. Agree by T is presumably involved, and it can't find the pvS or there wouldn't be a difference in agreement patterns to explain. If Agree by T can't find pvS (as Belletti's explicit statement about verbal agreement not holding with the pvS implies), then an explanation of CaseLicensing of pvS is required. (Partitive Case (Belletti 1988, Lasnik 1995 can't help: These data include definite specific pvSs and unergative verbs.) Another similar example of the endurance of Brandi and Cordin's (1989) analysis is found in Roberts (2010: 113), which directly discusses their data in (9), saying that "the obvious account of this is that SpecTP contains a deleted expletive pronoun, with which the subject clitic and the verb agree." Again, if the obvious is the case, and thus Agree(T,expl) determines agreement on the SCL and the verb, then the pvS wouldn't be involved in Agree with T, so CaseLicensing of the pvS needs an explanation. If, on the other hand, Agree(T,pvS) does hold, then why doesn't pvS determine agreement and the subject clitic? If both hold, we need a clearer understanding of how this is possible.
Another general shortcoming of such characterizations of the data is that, in Bolog nese, there are actually expletive constructions distinct from those in (12) in which there is an SCL.3SM, comparable to the French expletive that is sometimes mentioned. These include weather verbs and clausal complement verbs: In data like (1012), the clitic (a)l is identical to the SCL.3SM that appears in data with a preverbal 3SM subject (e.g. Pèvel al = dscårr 'Pèvel is speaking'), both in form and behavior. Its form exhibits the 3SM features that are often considered default or neutral, and the natural state of an expletive, as has been mentioned above. A relevant crucial behavior is inversion, which applies generally to SCLs in interrogative clauses in Bolognese, as is common in Northern Italian grammars (e.g. dscårr = el? 'Is he speaking?'). These clitics also invert in interrogatives, evidence that they are indeed SCL.3SMs like any other: In contrast, the clitic AI in the naprpvS data of interest to us here doesn't show the same similarities to the SCL.3SM in form and behavior. It does not have the same form in terms of the features it displays, a first concern. And note: If expls are 3SM as a default or neutral state, and if Bolognese has the SCL.3SM that can display those features, as it does with the expls in (1012), and if there are expls both in (12) and in (1012), then why does the SCL.3SM not appear in (12) as it does in (1012)? Given the logic underlying the relation between expls and EPP, we cannot simply assume there is no expl in (12).
Another concern is that AI does not invert in interrogative clauses like Bolognese SCLs do (including those pronounced [a]: a cradd 'I believe', craddia? 'Do I believe?'). It instead remains proclitic in interrogative clauses, just as it is in declarative clauses (though there are typical intonational differences between the two): Because of the distinct forms and syntactic behaviors of AI and SCLs, we cannot simply claim that AI is a SCL (or contains one), as has been done for the data from Brandi and Cordin (1989). Like Bolognese complement clitics in declarative and interrogative clauses (T s i dè. 'You give them to us'; S i dèt? 'Do you give them to us?'), AI remains proclitic in interrogative clauses. Standard analyses hold that complement clitics attach below T, which is where SCLs cliticize, allowing SCLs but not the complement clitics that cliticize lower to invert when a tensed verb raises higher than T in an interrogative clause. We therefore hold that AI also cliticizes below T, thus avoiding inversion in the same way as the complement clitics do. 3 Finally, let us consider the proposal made in Tortora (1999) for data again similar to Bolognese (12) and (56), as seen in (14). (14) (12), though it shows no clitic in such data that is obviously similar to Bolognese's AI or Fiorentino's SCL. It does have an el ement ngh glossed as LOC, which is treated as "a F[rench]type expletive, since it can only occur as a structural subject" (pg. 404). This grammar does have a typical SCL.3SM which doubles preverbal 3SM structural subjects, but it doesn't double this structural sub ject. Borgomanerese either has a morphological gap in its clitic inventory similar to the one in Trentino, if ngh is in SpecT like French il, or ngh is in fact like Bolognese AI, if it is a clitic doubling an expl.3SM (see the discussion in the conclusions). Like Bolognese, Fiorentino, Trentino, and others, it shows 3S on the tensed verb whether the pvS is S or P. (14c) reveals πrestrictions like those in Bolognese (56), and finally, as in Bolognese, it uses cRpvS for those πs that are forbidden in this other pvS construction (14d). Tortora (1999) uses early Minimalist mechanisms (Chomsky 1995, Cardinaletti 1997a, and proposes that the expletive checks the NOM(inative), 3, and S features of Agr, deleting all but the NOM feature (a point to which we return below). It should be noted that Agr is split in this account, with # above π, and NOM located in π. With NOM un deleted, the pvS can be, in our terms, CaseLicensed, via raising of its formal features at LF to Agr π . There, the formal features of the pvS are checked against NOM and 3 (as the expletive was), matching successfully whether the pvS is S (14a) or P (14b). For data like (9), Tortora (1999: 404) says that, "the person feature of a first/second isubject [pvS], on the other hand, would not match Agr pers 's [3pers] feature. Thus, a first/second person isubject would be prohibited from occurring with a verb that has third person singular morphology." This analysis has shortcomings, the first, most important of which was already mentioned within it. As noted above, it relied on not deleting the NOM feature when checked by the expletive, an assumption with "no convincing principled reason" (Tor tora 1999:401). Moreover, although not noted, the checking of the πfeature of the pvS should not be any more necessary than that of its #feature, as is noted on that same page: "given Chomsky's assumption that the phifeatures (i.e. the person and number features) on the argument are are [+Interpretable], they do not have to be checked" and "the Plu ralFF(DP) does not get checked against anything, but since it is [+Interpretable], it does not need to be checked." One might argue that, in that account, clashing πfeatures are lo cated within a single head, while clashing #features never are. But this clash is not about checking, as quoted. Some further mechanism would be required to establish and rule out the clash.
This analysis shows resemblances to those in the footnotes of Brandi and Cordin (1989) and Belletti (2005) in that its treatment of the πrestrictions is based on a clash between features of an expl and the pvS in such data. There are clear similarities also in the role that the expl plays in these accounts in determining the agreement that appears on the tensed verb. This account, however, is inspirational in light of more recent Minimalist mechanisms that are otherwise unexplored in this domain, in that it separately relates the single CaseLicensing head to the two important elements in such data, the expl and the pvS. Relationships between a single probe and multiple goals have been proposed in the literature, and a new account using these ideas will be developed in the next section that connects the data discussed so far to other important data in Romance grammars.

Cyclic/Multiple Agree
In the previous sections, we identified four important empirical and theoretical issues in naprpvS as requiring explanation. First, Bolognese uses a special clitic AI that is not one of its SCLs, while in other grammars there may be a SCL or no clitic at all. Bolognese thus give overt evidence that is new about what is occurring in data of this nature. Second, tensed verbs in the relevant data show 3S agreement. Third, the data shows πrestrictions, permitting only pvS.3. Fourth, the standardly assumed connection enacted by Agree be tween determination of agreement on T and CaseLicensing by it does not seem to hold in the same way in naprpvS as it does in cRpvS. An expl, perhaps with 'default/neutral' fea tures, determines agreement on T while the pvS appears to have no connection to T under standard assumptions about Agree. Optimally, the analysis of all these factors should be unified in a primitive distinction between naprpvS and cRpvS, with that distinction capable of capturing the variation observed within naprpvS.
We propose that these issues are all related to the expl that occurs in naprpvS, which is a (nondefault, nonlastresort) expl with specified features (discussed below) that is available in the grammars that have such data but not in those that don't. This expl is distinct from the default one in data like (1012) that occurs when there is no nominal argument available for Agree, as required by interface conditions to value and delete the features of the probe in this domain. It is also distinct from the one that occurs in cRpvS data, when it and the pvS act as a single nominal available for EPP and the interface requirements of the probe. As we connect and explain the four issues in naprpvS, we show the role that the special nature of this expl, its true independence from the pvS (unlike in cRpvS), plays in Minimalist mechanisms developed for independent reasons.
For the first two issues, the presence of AI in naprpvS and tensed verbs with 3S, we start by adopting the common notion that it is an effect of the expl that the tensed verb is 3S, meaning that (some form of) an Agree relation holds between it and T (we address the CaseLicensing issue below). This is appropriate both for naprpvS data like (12) which have a 3S tensed verb and the clitic AI, and for the other distinct Bolognese expletive constructions in (1012) which have a 3S tensed verb and the SCL.3SM (a)l. The verbal agreement is identical in the two, though the clitics are different. As noted above, our proposal rests on the difference between the two expls in these two sets of data: in (1012), the expl has the 3SM features considered default/neutral and natural for expletives in many grammars (Brandi and Cordin 1989, etc), while in naprpvS, we propose that Bolognese specifies an expl that is simply 3S. The clitic AI, which always and only appears together with this specified expl.3S, results from whatever cliticization process generally holds of argument clitics, but in this case applying to the specified expl.3S. (As noted in the previous section, the clitic attaches to a head beneath T, where SCLs attach.) This specified expl.3S is sufficient to value 3S agreement on the tensed verb, and this proposal thus correlates the first two properties of naprpvS (AI and 3S agreement) by means of standard effects of features involved in Agree. The lack of M on the expl.3S furthermore explains the lack of a SCL in naprpvS even if it moves to a position where subject cliticization could occur (since there is no SCL.3S without gender in Bolognese), and thus also of inversion in interrogative clauses in naprpvS.
For the third and fourth issues regarding πrestrictions and CaseLicensing, the addition of this specified expl.3S in naprpvS effectively adds a second nominal to the domain of a single probe, introducing the conditions necessary to be subject to the same mechanisms underlying the identical πrestrictions seen commonly in Romance data like Bolognese (1516). We first discuss these mechanisms, then return to explaining how the specified expl.3S underlies these last two issues. gives 'Pevel is giving me/us/you/you to him,her,them.' The explanation of Person Case Constraint effects like these are based, in recent accounts, on the elaboration of standard Agree known as Cyclic or Multiple Agree (Béjar and Rezac 2003, 2009, Nevins 2007, 2011. In the wellformed (15), a DCL related to an indirect ob ject occurs with an ACL.3 related to a direct object. In the illformed (16), a DCL related to an indirect object occurs with an ACL.1 or an ACL.2 related to a direct object. In each, ccommand holds between the two. Nevins (2007: 293) says that "it is assumed that these clitics double an underlying argument structure where the indirect object ccommands the direct object." Béjar and Rezac (2009: 46) refers to the dative DP as "closer," in this same, ccommandbased sense. For the illformed (16), Nevin's Multiple Agree approach holds that the third person dative interferes between the probe and the first or second person ac cusative that the dative ccommands (an intervention effect). In Béjar and Rezac's Cyclic Agree approach, a first or second person accusative, probed in a first cycle, leaves the probe unable to probe the ccommanding third person dative in a second cycle (a probe exhaustion effect). In the wellformed (15), these effects do not arise: a first or second person dative doesn't interfere between a probe and the lower third person accusative, or that accusative doesn't deplete the probe in the first cycle.
It is often noted that these PCC effects seem to apply only to clitics. Nevertheless, both major approaches refer to the underlying position of the cliticrelated elements, pre sumably because the ccommand relations between the clitics themselves would depend crucially on the analysis chosen, and on the application of the definition of ccommand to head internal elements, which are unnecessary complications of the discussions. C command holding between the related phrases captures the necessary generalizations and provides a basis for their explanations. Additionally, both approaches extend their mech anisms to phenomena not involving clitics, such as agreement displacement and omnivo rous number. Moreover, these approaches are based in elaborations of Agree, which also underlies standard treatments of agreement, of course, and CaseLicensing as well (and note that Rezac 2009:47 andNevins 2011:955 explicitly tie these approaches to CaseLicensing).
PCCeffects do not arise in every grammar, but they do so in Bolognese (and many other Romance grammars), so Bolognese certainly has whatever factors make PCCeffects possible, and according to Multiple/Cyclic Agree, they are a set of mechanisms that can, in some grammars, have an effect on more than data involving only clitics. Our analysis builds on this, and in particular claims that the specified expl and the pvS in naprpvS are subject to Multiple/Cyclic Agree and that the two remaining issues can thus be given a simple analysis unified with the first two by the effect of the specified expl. Because not every grammar with PCCeffects induced by Multiple/Cyclic Agree also exhibits data like naprpvS (e.g. Italian, which has PCC but not naprpvS), grammars like Bolognese must have some additional mechanism or element that those others lack. This additional mechanism, as noted above, is precisely the specified expl added to the derivation in those grammars that, like Bolognese, have it available.
Since there are thus two independent nominals in the domain of the single probe head T in naprpvS, the expl.3S and the pvS, both requiring a connection to it, the elab oration of Agree as Multiple/Cyclic Agree can apply there, and it explains the link be tween the third and fourth noted issues (πrestrictions and CaseLicensing of pvS). This contrasts with (1012), with the SCL.3SM (a)l, where there is no DP argument requiring CaseLicensing, and thus the (possibly default) expl.3SM is the only nominal available to satisfy EPP and to value and delete the uninterpretable features of T, as required, and pre sumably getting CaseLicenced in the process. (Expletives of this nature are illformed in Caseless, or nullCase, environments: English *It to rain bothers me vs. For it to rain bothers me; Bolognese *An um pièṡ brîṡa ed piôver 'It doesn't please me to rain' vs. An um pièṡ brîṡa ch'al piôva 'It doesn't please me that it rains'.) It contrasts also with cRpvS, where the pro in SpecT, whether called expletive or not (as in Belletti 2005, where it is contrasted with the 'expletive' pro), serves only to satisfy EPP, and it is only the features of the pvS that interact with those on T, simultaneously determining agreement on T and CaseLicensing pvS, either by longdistance Agree or by sharing them with pro. The pro and the pvS in cRpvS share the properties of a related preverbal subject, and they are, in effect, one nominal (which is made literal in Belletti 2005). On the other hand, in naprpvS data like (12) and (56), we have two nominals needing to relate to T. We have agreed with the literature that it is an effect of the expl that the tensed verb is 3S, indicating that Agree(T,expl) holds. Agree(T,pvS) should also need to hold for CaseLicensing of pvS, but this would result in pvS determining agreement on the tensed verb, contrary to fact. As noted, however, this is exactly the effect of Multiple/Cyclic Agree, an elaboration of Agree established for independent reasons. In data like (1012) and cRpvS (where pro and pvS split the duties of a single DP) there is only one DP available for CaseLicensing and determination of agreement, the SCL.3SM and the pvS respectively, and simple Agree will find the one DP without problem; Multiple/Cyclic Agree would be redundant, un necessary, it would effectively reduce to simple Agree since there is only one goal in the domain of the probe T. In naprpvS, however, there are two independent nominals in the domain of T, the only available CaseLicenser. Both need to be CaseLicensed, and by standard logic, simple Agree can only find the higher one, which is the expl, since it is the one that raises to SpecT for EPP. The pvS in naprpvS would thus never be CaseLicensed, if simple Agree applied, because of the intervention effect of expl.
Instead, since Multiple/Cyclic Agree is available in such in a grammar, it can and does apply in this sort of data. As standard in these approaches, the individual φfeatures probe separately. A full discussion of the differences between the Multiple Agree and the Cyclic Agree approaches is beyond the scope of this paper, but either one should be com patible with our proposal, which requires only the shared essential notions of the ability of a probe to relate to more than one goal, the relevance of ccommand between them, and the role of π in CaseLicensing. Several differences between them will result in different specific assumptions about the structures involved. Most significant is the location of the two nominals relative to the probe: Multiple Agree has the probe above both, Cyclic Agree has the probe between the two (and permits upward probing). In effect, this means that in the former, the expl must be below T, then raising to SpecT for EPP, while in Cyclic Agree, the expl must start in SpecT. Both of these ideas are common in the literature, and this paper will make no arguments one way or the other. The specific mechanisms for capturing πrestrictions in each approach differ in a correlated way, the intervention vs. probedeletion effects described above.
Let us now turn to the implementation of these proposals in the Bolognese data investigated. Given space limitations, the following is presented only in terms of Multiple Agree. Consider the structure in (17), in which (i) the heads and projections associated with auxiliary verbs and participles are ignored, since they are irrelevant to the discussion, and (ii) the structure indicates only an unergative verb: the structure would be identical (above the Tdomain) with an unaccusative verb: In naprpvS data like (12) and (56), the pvS merges in its Tposition, and does not undergo Amovement, or any other movement relevant here. The specified expl merges between between T and v/VP, because (i) it must ccommand the pvS, including the external argu ment of an unergative verb, and (ii) its associated clitic AI cliticizes below T, as discussed. When T merges, its πfeature probes in accordance with Multiple Agree. If the pvS is 3 (12), π on T simultaneously probes and can find both expl and pvS, since no intervention effect arises from an expl.3 between T and a pvS.3. This successful multiple probe by π CaseLicenses them both: T is valued as 3, the cases of the two DPs are valued as nomi native, and these all can be deleted. This mechanism resembles Tortora's (1999) idea, but with no unmotivated assumption about the nondeletion of the NOM feature of Agr/T. Si multaneously, the # feature of T probes and finds the expl, valuing T as S, 4 another aspect of the notion that the agreement on T is due to the presence of the expl. If the pvS is 1 or 2 (56), and π attempts to probe both expl.3S and pvS simultaneously, it finds the expl that ccommands the pvS, but the feature 3 on expl creates the discussed intervention effect between T and the pvS with 1 or 2, blocking this part of Multiple Agree. Since the pvS's casefeature is unvalued, it can't be deleted (it is not CaseLicensed), and data like (56) is thus illformed.
Contrast the structure for naprpvS in (17) with the general characterization of cR pvS in (18), which highlights the important points, but abstracts away from particular analyses, which are not crucial here. Relevant data include (7) and (8).
This is simple Agree. While personcase effects occur in Bolognese and other Romance grammars, they show no omnivorous number nor numbercase effects, and no reason to suppose that #, unlike π, ever involves Multiple/Cyclic Agree. This is directly related to the role that π, as opposed to #, plays in CaseLicensing. See Nevins (2011) for discussion.
The pro/expl in SpecT is associated with the pvS, and they have the same features (dashed line). It is surely expletive in the sense that it does not refer (separately from the pvS), though it may or may not be labelled as such. It may or may not have moved from a lower position, perhaps in close association with, or within, the pvS. This pro/expl, which either has no features or has features identical to and dependent on those of the pvS, merely satisfies EPP, while Agree(T,pvS) simultaneously determines agreement (and any SCL) on the tensed verb and enacts CaseLicencing of the pvS. Together, the pro/expl and the pvS act like a single element, in particular like a preverbal subject in many grammars, including those in Bolognese (34).
Finally, the structure in (19) represents data like (1012), where there is no argu ment nominal in the domain of T with which it can Agree. A (default) expl.3SM occupies SpecT to satisfy EPP as well as to determine agreement (and the appropriate SCL) on T, and it is presumably also CaseLicensed there by T, as discussed above. This expl might start in a lower position and satisfy Agree(T,expl) there, and then move to SpecT for EPP (dashed arrow).

Conclusions and Extensions
In Bolognese, an expl that is specified as 3S and independent of the pvS is introduced into naprpvS. This expl.3S underlies the four issues of this data type, including the clitic AI that is related to it, the obligatory 3S agreement of the tensed verb, the πrestrictions that rule out a pvS.1 or pvS.2 that it ccommands, and the CaseLicensing of the nonassociate pvS apparently without agreement between it and the sole CaseLicensing head in the domain, T. With only one probe head available for CaseLicensing the two nominals, the standard application of Agree is insufficient. As a result, the application of the elaboration of Agree as Multiple/Cyclic Agree, independently motivated in the grammar, is induced by the introduction of the independent expl.3S, which in turns provides a novel explanation for these πrestrictions that is thus linked to completely independent explanations of the πrestrictions in previously unconnected Romance data.
The naprpvS data differs from two other constructions standardly assumed to in volve expletives. The expls in those differ both from each other, and from the one in naprpvS. In cRpvS, the expl either is without features other than whatever is required to satisfy EPP or has the same features as the pvS in the data. The two elements thus act as a single nominal related to the probe T. In constructions without a nominal available to Agree with T, a default expl.3SM behaves like a preverbal subject, satisfying EPP, deter mining agreement on T (including the SCL.3SM), and being CaseLicensed by it.
This analysis for naprpvS can be extended to other grammars with simple, typically morphological adjustments. For example, Trentino lacks a clitic in data like naprpvS, and Fiorentino displays an impersonal clitic identical to its SCL.3SM. The same analysis as proposed here for Bolognese can apply to these, with only the features of the expl and the morphological realizations of the clitic involved differing. In Trentino, there is the above quoted "morphological gap," which we take to mean that there is no clitic in Trentino associated with its specified expl.3S that is comparable to Bolognese AI. In Fiorentino, there is this same morphological lack as in Trentino of a clitic comparable to AI, but, unlike in Bolognese and Trentino, the introduced expl is 3SM and not 3S. This specification is rich enough to make possible a SCL.3SM when probed by T. Borgomanerese is quite similar to Bolognese, and we could hold that its clitic ngh, glossed as LOC, is actually an equivalent to AI (and perhaps homophonous with a locative). Other possibilities could arise, given the idea that expls with particular featuresets can be specified in particular grammars. This idea seems reasonable, with the variation in the formal properties of expletives that has already been explored. Moreover, Poletto (pers. comm.) notes that, in Rodoretto di Prali, weather verbs seem to have expl.3SF, which suggests that there, at least, the default doesn't occur, or, possibly, that a deeper look at the notion of the default is required. A possibility under the present account is that specific grammars could potentially specify the features for expletives not only in naprpvS, but also in domains without argument nominals, even if many of them do not do so, or they could specify a different set as default. In this way, Rodoretto di Prali may be held to specify 3SF, which would be easy enough to acquire through positive evidence, every time it rained. All this suggests that the notion of specification of features for expletives merits deeper exploration.