Construções impessoais com si  em italiano têm sido o foco de inúmeros estudos. Muitas análises, como as de Cinque (1988), Chierchia (1995) e Dobrovie-Sorin (1996, 1998, 1999), foram propostas para definir os intrigantes padrões de concordância das construções com si . Neste trabalho, mostro que todos os diversos padrões de concordância derivam da dupla natureza do si  como um núcleo e como um DP (Chomsky 1995:249). Minha análise não postula propriedades especiais para o si  que o tornariam peculiar com relação a outros clíticos. Pode-se considerar que o si  age apenas ao nível sintático e não, como em outras propostas, no léxico.


Impersonal si constructions belong to the wider group of impersonal constructions, which are used to introduce a generic, unspecified subject in an utterance and to make general statements about groups of people. There are several strategies for obtaining these results in Italian, impersonal si being one of the most commonly employed. An example of a si construction is given in (1):

Figure 1.

The sentence in (1) is a statement about a property of a generic group of people; the absence of si would result in a sentence with a specific subject, as shown in (2). Italian is a pro-drop language. When the subject is a pro, as in (2), it needs to refer to somebody deducible from the context or already introduced in the discourse:

Figure 2.

Impersonal si constructions show a number of puzzling agreement patterns. In the next section, I examine impersonal si with transitive verbs.

1. Impersonal SI with Transitive Verbs in the Present Tense

In the present tense, si constructions with transitive verbs show two main agreement patterns, exemplified in (3) and (4).

Figure 3.

(3) and (4) have the same meaning, are made up of the same lexical items but display two different agreement patterns.

In this paper, I address the following question: what causes the difference in agreement patterns between sentences (3) and (4)? The difference in agreement patterns, I argue, is caused by the presence of si.

In the next section, I provide an overview of the theoretical background that I will use for the analysis of si constructions. In section 3, I first introduce Anagnostpoulou’s (2000)[1] model for the analysis of double object constructions. Then I discuss similarities between the agreement patterns in double object constructions and those in impersonal si constructions with transitive verbs. Next, I propose to extend Anagnostopoulou’s model to the analysis of impersonal constructions. More explicitly, I propose to analyze si constructions with verb-object agreement as quirky subject constructions. In section 4 I present an analysis for impersonals with unergative and unaccusative verbs. Finally, section 5 contains my conclusions.

2. Theoretical background

2.1. Uninterpretable features

According to the model outlined in Chomsky (1995, 1999)[2,3], syntactic expressions must be legible at the interface between the syntactic system and the other systems. This means that all the features which would not be interpretable by the other systems, e.g. by the phonological or by the logical system, need to be eliminated before the interface levels are reached. More explicitly, Chomsky (1999)[3] proposes a mechanism to eliminate uninterpretable features which can be briefly summarized as follows: Some features on lexical items have no value and need to be valued (and consequently eliminated) before the interface with other systems is reached, or the derivation will crash. The valuation takes place when a Match relation between phi-features on lexical items is established. Such a relation is established as soon as lexical items enter the derivation (i.e., are merged). The Match relation triggers an Agree relation; under Agree unvalued features can be valued and deleted from narrow syntax. Following Chomsky (1999)[3], I assume that the Agree relation doesn’t necessarily take place in a specifier-head configuration, but can be a long-distance relation, yet subject to locality conditions.

The choice of a derivational approach implies a ‘step-by-step’ definition of the relations between lexical items. Therefore, the mere presence of si in the Numeration cannot justify the disappearance of a Theta-role or of Case. Si has to enter into Match and Agree relations with other items and interact with them syntactically. Such an interaction takes place locally.

2.2. Phases and derivations

Chomsky (1995, 1999)[2,3] outlines a strictly derivational model for syntactic structures. The relations between lexical items are not representationally defined (as in the Government and Binding framework) and are established during the derivation. Following the ‘Derivation by Phase’ approach, I assume that the relations between lexical items are established as soon as they are taken from the Numeration1 and merged. However, the deletion of the features that are valued via Agree only takes place at the end of a phase.2 According to Chomsky (1999)[3], vP and CP are phases. Although the concept of phase doesn’t play a crucial role for my analysis, I will follow Chomsky’s 1999[3] model unless otherwise indicated.

2.3. The features of si

In this paper, I argue that si is not a special lexical item that absorbs a Theta-role or Case (contra Cinque 1988[4], Reinhart and Siloni 1999[5], Reinhart 2000[6]; pro Dobrovie-Sorin 1998, 1999[7,8]). I show that si does not have any special property, except the one deriving from its clitic nature. As a clitic, si exhibits a double status as a DP and a head (see Chomsky 1995:249)[2].

I assume that Italian has two different si’s: an impersonal one and an anaphoric one. These si’s have different features, and different behavior with respect to Agree relations.

I assume that impersonal si is referential, and consequently can both value phi-sets and have its Case features valued. As shown by Chierchia (1995)[9], impersonal si refers to an unspecified group of humans, which usually includes the speaker. Among the phi-features of impersonal si, the most relevant for this analysis is number. Si has a plural feature, as shown in (5):

Figure 4.

In (5), si controls PRO and triggers plural agreement on the adjective. Therefore, si can be considered as a plural.

The status of anaphoric si, exemplified in (6), is different from that of impersonal si.

Figure 5.

Anaphoric si needs to be bound by an antecedent in order to have its phi-features valued (see Reuland 2001[10], Law 2002). In other words, it is not referential. For this reason, I propose that anaphoric si cannot value phi-features, in contrast to impersonal si, which can. This also means that anaphoric si cannot act as an intervener in checking operations.

2.4. The argumental status of si

There is large disagreement about the argumental status of si. Many proposals have been made, among which the most relevant for the agreement problems I am examining are Burzio’s (1986)[11], Cinque’s (1988)[4] and Dobrovie-Sorin’s (1996, 1998, 1999)[7,8,12].

Burzio (1986)[11] simply considers si as an argument. Cinque (1988)[4], on the other hand, postulates the existence of two different si’s. Si is usually restricted to finite clauses. However, si is allowed in certain untensed clauses, namely in Aux-to-Comp (see Rizzi 1981, 1982)[13,14] and Raising structures with transitive and unergative verbs. (7) is an example of an Aux-to-Comp construction with a transitive verb, and (8) is an example of a Raising construction with a transitive verb [from Cinque 1988, 524-525].[4]

Figure 6.

Cinque proposes to consider these instances of si as argumental ones (+arg), which can be present in general only with verbs that project an external Theta-role. The other si, which can be present with any class of verbs, i.e. also with verbs that do not assign an external Theta-role, is a non-argumental one (-arg) [from Cinque 1988[4], 522, si with an unaccusative verb (arrive) and with a passive]:

Figure 7.

According to Dobrovie-Sorin (1996, 1998, 1999)[7,8,12], though, it is not necessary to postulate the ± argumental nature of si. What Cinque calls a +arg si is actually a passive si, which cannot be marked with Nominative. The only Nominative si is the one that Cinque defines as –arg. Si is not licensed in non-finite clauses because it is a Nominative clitic and in Italian Nominative clitics are not allowed in non-finite clauses. Transitive and unergative Aux-to-Comp and Raising structures allow si just because si in this case is not Nominative but Accusative. In other words, the si that is licensed in some non-finite structures (such as 7 and 8) is a middle-passive si, and not a Nominative one. Dobrovie-Sorin’s analysis has several advantages, since it can be extended to other Romance languages, such as Romanian, which doesn’t have Nominative clitics but has si constructions. In minimalist terms, however, one wonders why if there is a theta-position available for a DP and if there is exactly one DP present in the numeration, namely si, one should merge si in a non-theta-position and merge an expletive in subject position and let the chain formed by the two items absorb the external Theta-role. The considerations that led Cinque and Dobrovie-Sorin to discuss the argumental status of si were mainly related to the Projection Principle and to the division between D-structure and S-structure. In a model that doesn’t make use of these levels of representation, most of the arguments necessarily disappear. For further discussion see Manzini & Savoia (2000)[15], Embick (2000)[16], McGinnis (1997, 1999)[17,18], and Raposo & Uriagereka (1990)[19], among others.

In my analysis, I postulate no restrictions on the merging sites of si, which will be merged in an argument position whenever such a position is available.

2.5. Theta-roles

Following Hale & Keyser (1993)[20] and Chomsky (1995)[2], among others, I assume that thematic roles are determined configurationally.

According to Burzio’s Generalization, if a verb does not assign an external Theta-role it does not assign accusative Case. However, as has been pointed out by many linguists (cf. Marantz 1991[21], Burzio 2000[22], Reuland 2000[23]), Burzio’s Generalization has to be revised and decomposed, as it links very different properties of a predicate such as structural case and thematic roles. For the analysis of the examples that follow I assume a structure in which external Theta-role and Accusative are not necessarily assigned by the same head, in the same projection.

Given these basic assumptions, all the anomalous agreement patterns in si constructions surface as the result of syntactic derivations that involve si.

3. The analysis

In this section, I present an analysis of impersonal si constructions that is based on Anagnostopoulou’s (2000)[1] analysis of double object constructions. After a brief overview of Anagnostopoulou’s model, I show that this model accounts for a kind of double object construction in Italian, the so-called self-Benefactive, as well. I then propose to adapt Anagnostopoulou’s model to impersonal si constructions, in order to account for the apparent mismatch in agreement patterns between transitive impersonals with verb-object agreement and transitive impersonals that don’t show such an agreement. I also show how the past participle agreement patterns can be accounted for by using a strictly derivational approach.

3.1.1. Anagnostopoulou’s proposal

Anagnostopoulou (2000)[1] proposes a double structure for the analysis of double object constructions, which finds independent motivation in the work of Marantz (1991, 1993)[21,24]. English double object constructions are well known for their alternation of a PP with a dative Benefactive:

(10) a. I give a book to John/him

b. I give John /him a book

Italian doesn’t have double object alternation with two DPs. Yet, there is a construction with personal pronouns that closely resembles the English double object alternation in (10). In (11) the Benefactive a lui is a PP. In (12), the Benefactive gli is dative:

Figure 8.

In both (11) and (12) un libro is Accusative. This can be shown by substituting a pronoun for the DP object un libro, as in (13):

Figure 9.

In spoken Italian there is a very interesting kind of double object constructions, exemplified in (14):

Figure 10.

In this construction, which I call a self-Benefactive construction, si is anaphoric, and thus it is inflected according to the DP that binds it, as shown in (15):

Figure 11.

In (14), though, si cannot alternate with a PP, as shown in (16):

(16) *Giannii legge un libro a Giannii

Gianni reads a book to Gianni

‘Gianni reads a book to Gianni’

The alternation between (14) and (16) is not blocked because of the impossibility of double object alternation in Italian, which is in fact possible, as (11)-(12) show. (16) is out because of anaphora constraints. The two instances of Gianni must corefer, and in traditional terms there is a Principle C violation.

In (14) si is clearly dative, as can be shown by substituting for it a third person pronoun, which in Italian shows morphological case (see 17).

Figure 12.

Thus, (14) is a special double object construction in which si is a Benefactive dative clitic, coreferential with the subject Gianni.

Anagnostopoulou (2000)[1] proposes two different structures to account for double object constructions: one with a double v, which includes a causative head (vCAUS, v1) and an applicative head (vAPPL, v2) (see 18), and one with only one v (see 19). In the structure with a double v, the applicative head introduces the Benefactive (see Marantz 1993[24], McGinnis 1998[17], Anagnostopoulou 1999)[25], while the causative head, which is higher, introduces the external argument (see 18). In the second structure, the Benefactive is merged in the specifier of VP and the external argument in the specifier of v (see 19).

Figure 13.

In some languages, e.g. Spanish, Albanian and Icelandic the applicative head v2 assigns morphological dative (see 20 for Icelandic):

Figure 14.

Anagnostopoulou (2000)[1] proposes that there is morphological dative on the Benefactive in this group of languages if and only if there is an applicative head that can assign dative. As I have shown in (17), si in (14) is dative. Si also indicates a Benefactive in (14), and therefore Italian is one of the languages that mark the Benefactive with dative case.

Adopting Anagnostopoulou’s proposal, thus, we can say that in (14) si is merged in the specifier of the dative assigning head v2. Notice that for the analysis of self-Benefactive constructions with a double v I assume that v2 in Italian doesn’t assign accusative Case but only dative case. The derivation for the self-Benefactive construction in (14) is as follows:

  • The DP object un libro is merged with the root verb.
  • v2 is merged with the VP.
  • Si is merged in the specifier of v2.
  • Si gets dative case from v2. In this construction si is an anaphor, and therefore its phi-features are not valued. This means that it doesn’t take part in valuing operations until its referent is established. Once this happens, si gets the Benefactive role in spec, v2.
  • v1 is merged. It Matches the object DP un libro and assigns it Accusative.
  • T is merged.
  • The subject DP is merged in the specifier of T. There it has its Case features valued (Nominative).
  • The subject also enters a Match relation with si, whose phi-features it values according to the anaphoric mechanism proposed in Reuland (2001).[10]
  • Figure 15.

    Independent motivation for the claim that Benefactives are marked with dative in Italian (and thus that si is marked with dative in self-Benefactive constructions) is provided by a class of verbs that are inherently self-Benefactive, such as riservarsi (to keep for oneself), accaparrarsi (to hoard), assicurarsi (to secure), or procurarsi (to get oneself). Such verbs are most commonly used in the reflexive form, and have a self-Benefactive meaning. The si which appears on these verbs is an anaphor. In (22a) si refers to the subject Maria. In (22b) the self-Benefactive verb with si is located in an embedded sentence. Si still refers to Maria, but it has to surface as a pronoun because of anaphora constraints. It surfaces as a dative pronoun. I take this to show that Benefactive si is dative in Italian.

    Figure 16.

    Observe that since impersonal si doesn’t show any inflection, the morphological marker of dative does not surface on si. Despite this, I take (17) and (22) to show that dative is there.

    3.2. Impersonal si constructions with transitive verbs

    In the previous section we have seen how a self-Benefactive construction can be derived according to Anagnostopoulou’s proposal. In this section, I propose to extend this analysis to impersonal si constructions with transitive verbs. Specifically, I propose to adopt the structure with two v’s for the derivation of impersonal si constructions with verb-object agreement of the kind exemplified in (3) and the structure with only one v for the structure with no verb-object agreement in (4).

    As briefly shown in 1.1., impersonal si constructions with transitive verbs in the present tense display two main agreement patterns, exemplified in examples (3) and (4), here repeated as (23) and (24):

    Figure 17.

    In (23) the verb agrees with the Nominative object. The object is a real object. This can be shown by substituting for it the partitive clitic ne, which can only be merged as an internal argument, as shown by Belletti and Rizzi (1981)[26] and Burzio (1986)[11] among others:

    (25) In Italia se3 ne mangiano

    in Italia si of them eat

    ‘In Italy people eat them’

    In section 3.2.1., the construction in (23) will be shown to be a quirky subject one. Quirky subject constructions are a very well known phenomenon of Icelandic (Sigurðsson 1996[27], Tarldsen 1994, 1995)[28,29]. In such constructions, the Nominative DP has been proved to be an object (Alexiadou to appear, Zaenen, Maling & Thrainsson 1985[30], Sigur∂sson 2000[31]). The parallelism between (23) and quirky subject sentences in Icelandic provides further evidence for the objecthood of gli spaghetti.

    A third consideration can be added about the object status of gli spaghetti in (23). Stardardly, objects are associated with the Theme Theta-role, or in general with the lower Theta-role of the thematic hierarchy (Grimshaw 1990)[32]. In (23) the verb mangiare (to eat) assigns two Theta-roles: an Agent and a Theme. It is clear-cut that the Theta-role that is assigned to gli spaghetti is the Theme. Therefore, gli spaghetti is an object.

    The object status of spaghetti in (24) is deducible along the same lines of reasoning followed for (23).

    Observe that the Case of the object gli spaghetti in (23) is Nominative, while the Case of the object spaghetti in (24) is Accusative. This is shown in (26) and (27) respectively, where the DP object is replaced by a pronoun, which in Italian is marked for Case:

    Figure 18.

    (23) and (24) are thus two parallel constructions that mean exactly the same and vary only for their agreement patterns.

    3.2.1. Verb-object agreement

    In the previous section we have seen that there is a construction, namely (23), in which the verb agrees with the Nominative object. In this section I first show that a parallelism exists between (23) and the self-Benefactive constructions of the kind exemplified in (14), and then I extend the analysis of self-Benefactive constructions to (23).

    Consider the following sentences:

    Figure 19.

    Figure 20.

    (28) is a self-Benefactive construction, (29) is an impersonal one. These two constructions will be shown to have the same underlying structure. The structural case of the object dei buoni libri is not the same for (28) and (29). In (28) dei buoni libri is Accusative, as can be shown by substituting a pronoun for the DP (see 30); in (29), the DP object dei buoni libri is Nominative, as shown in (31)4 :

    Figure 21.

    I argue that the different Case on the DP object is due to an intervention effect performed by the impersonal si in (29), which is not performed by the anaphoric si in (28). The impersonal si is, in fact, referential, while the anaphoric one isn’t (see section 2.3). In (28) si is dative, as shown in section 3.1.1. Also in (29) si is dative. Therefore, a dative assigning v must be present. As we saw in 3.1.1., Anagnostopoulou (2000)[1] proposes that dative is assigned in double object constructions (and self-Benefactive ones) by an applicative head (v2). I propose to introduce a second v, which I will call v2 to keep the parallel with self-Benefactive constructions, also for impersonal si constructions. v2 assigns quirky dative. Observe that this v2 is not an applicative head, but just a quirky dative assigning head. (29) is thus an example of a quirky dative construction.

    The derivation of (29) is as follows:

  • The DP dei buoni libri is merged with the verb, in the complement position, where it receives its internal Theta-role. Its structural case features need to be valued.
  • v2 is merged with the VP.
  • Si is merged in the specifier of v2, and gets quirky dative. I assume that although si is marked with quirky dative, it can still intervene in checking operations (see Zanen, Maling and Thrainsson 1985[30])5. In this position, si also receives the external Theta-role.
  • v1 is merged; si is a DP, has phi-features (see 2.3) and is still visible for Case assignment. Therefore, v1 enters a Match+Agree relation with si, which is able to value the phi-features on v1 because of its intrinsic referentiality. After this valuation, si becomes inactive for further syntactic derivations, and only moves because of its clitic nature.
  • The direct object dei buoni libri stays without its Case feature valued.
  • T is merged, and the verb raises to T.
  • Si cliticizes on T.
  • A Match+Agree relation takes place between T and the direct object, which gets nominative Case and values the phi-features on T, which agrees with it.
  • The EPP on T is checked by an expletive pro, if present in the Numeration.
  • (32) [TP pro [T sij –leggonoi [v1P ti [v2P tj [VP ti dei buoni libri]]]]]

    Basically, si performs an intervention effect, preventing the transfer of Accusative by v1 to the direct object. Thus, the direct object has to wait until T is merged in order to receive its Case, which will be Nominative. Observe that this analysis suggests that v1P does not constitute a phase, contra Chomsky (1999)[3]. If it did, it would not be possible for the object to go to Spell-Out without being assigned Case, and the derivation would crash.

    Figure 22.

    If no pro is present in the Numeration, the EPP on T is checked by the DP object dei buoni libri, which is the only DP available for raising, since si has cliticized and cannot check the EPP any longer. If the object raises, the sentence sounds as follows:

    (34) Dei buoni libri si leggono in Italia

    some good books si read-3RD PL in Italy

    ‘In Italy people read good books’

    Observe that when the object raises si has cliticized (i.e. has become a head) and hence doesn’t block the object raising. Its trace doesn’t block raising either.

    (35) [TP [DP dei buoni libri]k [T sij- leggonoi [v1P tk ti [v2P tj [VP ti tk ]]]]]

    The derivation in (32) shows that it is not necessary to postulate properties for si that would tell it apart from other clitics. The mere presence of si in the Numeration doesn’t imply an absence (or an absorption) of the external Theta-role and of Accusative case. The external Theta-role is in fact assigned in impersonal si constructions. As a matter of fact, a by-phrase, which introduces an Agent, is licensed in passive contexts, which lack an Agent role, (see 36) but not in impersonal si contexts (see 37), which have it assigned.

    Figure 23.

    Evidence that si checks Accusative in verb-object agreement constructions is provided crosslinguistically by Romanian (Dobrovie-Sorin 1996, 1998, 1999)[7,8,12]. In Romanian there is an impersonal construction that mirrors the Italian one, namely an impersonal se construction with verb-object agreement. In such a construction, se shows Accusative case. I take this as a piece of evidence that se-si actually gets Accusative, and does not block its assignment (Cinque 1988)[4]. The Romanian counterpart of (29) is (38):

    Figure 24.

    3.2.2. Quirky subjects in italian and icelandic

    In 3.2.1. I have proposed that Italian impersonal si constructions are quirky subject constructions. This claim isn’t ungrounded, as I will show in this section.

    Icelandic quirky subject constructions, as the one exemplified in (39), show the following properties:

  • A Nominative object agreeing with the verb
  • A dative subject
  • A restriction on the person feature on the object
  • Figure 25.

    In (39), the verb agrees with the Nominative object. In addition to that, there is a person restriction on the object (Sigurðsson 1996)[27], which can only be 3rd person, as shown in (40):

    Figure 26.

    Italian impersonal si constructions show exactly the same agreement patterns as Icelandic quirky subject constructions. In particular, they exhibit Nominative object-verb agreement. In addition to that, si constructions present a person restriction on the Nominative object, which can only be 3rd person (see 41- 42). That impersonal si is dative will be shown in details in section 3.2.3.

    Figure 27.

    These striking similarities constitute crosslinguistic evidence for considering impersonal si as a quirky subject construction.

    3.2.3. No verb-object agreement

    The alternative agreement pattern for si constructions with transitive verbs was exemplified in (4) and (23), and is repeated here in (43):

    (43) In Italia si mangia spaghetti

    in Italy si eats-3RD SG spaghetti-PL ACC

    ‘In Italy people eat spaghetti’

    In (43) there is no agreement of the verb with the object. The ending of the verb is the default third singular one. Like in (23), also in (43) the object is an internal argument, as shown in (44):

    (44) In Italia se ne mangia

    in Italy si of them eats-3RD SG

    ‘In Italy people eat them’

    (44) shows that in (43) the object is a real object, i.e. an internal argument. The tests applied for (23) in section 3.2. are also valid for (43). In addition to that, in (43) the object is Accusative, as shown in (21), here repeated as (45):

    Figure 28.

    In (43) there is no V-O agreement: the verb exhibits the 3rd person singular default ending and the object bears Accusative. According to my proposal, if Accusative is assigned to the direct object no intervention effect of si can possibly have occurred. For this kind of sentences I assume in fact the second structure proposed by Anagnostopoulou (2000)[1], namely the one with one single v. The existence of only one v means, in Anagnostopoulou’s terms, that there is no dative assigning head (i.e. there is no v2). Impersonal si doesn’t show any inflectional morphology, and thus it is hard to detect the case that si bears. Yet, I argue that si in (43) is not dative, and that the construction in (43) lacks a dative assigning head. This statement is not unsubstantiated. Anagnostopoulou (2000)[1] shows that there is a strong correspondence between lack of the dative assigning head and presence of a bare noun object. Specifically, she shows that when there is no dative Benefactive (and thus when there is only one v) it is the Benefactive that checks the only Case available (i.e. Accusative) and the real object is licensed by (abstract) incorporation (see Baker 1996)[33]. In order to have incorporation, a bare object is required (see Baker 1988[34] and Van Geenhoven 2001)[35]. Anagnostopoulou’s (1999)[25] proposal is summarized in table 1.

    Table 1.

    a. 2 v ’s DAT morph. on the Benefactive & DP object
    b. 1 v no DAT morph. on the Benefactive & bare noun object

    In other words, in some languages whenever the dative assigning head is missing the object of the construction must be a bare noun. This implication is bi-directional. (46) and (47) show that Italian is one of the languages for which this implication holds.

    Figure 29.

    Example (46) shows that the presence of a bare noun excludes the possibility of a dative Benefactive. Example (47) shows that if a Benefactive is present a bare noun is not licensed. Anagnostopoulou’s equation: ‘bare noun = no dative assigning head’ holds for Italian. No v2 is present when a bare noun object is available.

    The pattern proposed by Anagnostopoulou for double object constructions seems to hold for si impersonal constructions as well. In particular, (48) and (49) show that, in sentences with no agreement between the verb and the object, the object must always be a bare noun. According to Anagnostopoulou’s equation, when the object is a bare noun (as in the sentences with no verb-object agreement) there is no dative. In (48), a no verb-object agreement construction, a bare noun object is required. Thus, there cannot be a dative in the sentence, and therefore si is not a dative. In (49), on the other hand, which is a verb-object agreement sentence, a bare noun object is very odd. There is a DP object that can license a dative. Si is dative in this case:

    Figure 30.

    I conclude that Anagnostopoulou’s equation is also true for impersonals, and thus that there is no dative assigning v when the object is a bare noun. Moreover, following Baker (1988, 1996)[33,34], the object can be taken to incorporate into the verb. The derivation of (43) is thus as follows:

  • The object spaghetti is merged with the verb.
  • v1 is merged, and the object incorporates into the verb.
  • Si is merged in the specifier of v1 and gets the external Theta-role.
  • T is merged; si immediately cliticizes on it; si’s Case features are valued by incorporation/cliticization on T. Note that this kind of incorporation/cliticization doesn’t trigger agreement.
  • An expletive pro is merged in the specifier of T to check the EPP.
  • The verb in T shows the default 3rd person singular inflection because of lack of agreement with a DP (Benveniste 1966)[36].
  • Figure 31.

    3.3 Past participle agreement with transitive verbs

    In the past tense (passato prossimo) of impersonal si constructions with verb-object agreement, the past participle shows agreement with the Nominative object:

    Figure 32.

    In particular, the auxiliary is plural if the object is plural, and the past participle agrees in number and gender with the object. Italian speakers do not perceive the non-agreeing form as grammatical (see 53).

    Figure 33.

    The phi-set on Italian past participle is incomplete, because the participle lacks person. Following Chomsky (1999)[3], I will consider the past participle as having unvalued Case features. The direct object, which also has unvalued features, is phi-complete and can enter a Match relation with the participle. The derivation of (52) runs as follows:

  • The direct object molte macchine is merged with the verb.
  • The past participle head is merged, and the verb moves to it. From there, pp enters a Match relation with the direct object, which values the unvalued phi-features on pp, according to the mechanism proposed in Chomsky (1999)[3]. The direct object remains with its Case features unvalued, as pp is not phi-complete, non-referential and cannot value the object’s Case. Both pp and the object still need to have their Case features valued.
  • v2 is merged; si is merged in the specifier of v2. There it gets dative case and the external Theta-role. Despite this case, si can still intervene in checking operations.
  • v1 is merged. It enters an Agree relation with si, which gets its Case features valued and values the phi-features on v1. The past participle and the direct object are still with their Case features unvalued.
  • The auxiliary is merged in T. Si cliticizes on T and is no longer visible for any Agree relation. The auxiliary has unvalued phi-features.
  • T establishes an Agree relation with pp, whose features are still visible. The Case feature on pp is valued. However, the phi-set of pp is incomplete, and thus it cannot value the phi-features on T.
  • T looks deeper down and Matches with the direct object, which is phi-complete and can value the phi-features on T. As a result, the direct object gets nominative Case.
  • The EPP on T is checked by an expletive pro.
  • If the object is not a DP but a clitic, as in (54), the auxiliary shows the default third singular ending and pp agrees with the object clitic:

    Figure 34.

    The sentence in (54) exemplifies a property of Italian, namely the fact that the past participle agrees with the object clitics. If this agreement is missing, the sentence sounds ungrammatical, as in (55):

    Figure 35.

    Furthermore, in (54) the auxiliary must show the 3rd singular default ending. The plural ending, which would show agreement of the auxiliary with the object, is ruled out:

    Figure 36.

    We are thus dealing with a construction with no verb-object agreement. Therefore, only one v is present. The derivation for (54) is the following:

  • The direct object is merged with the verb.
  • pp is merged and the verb raises to it and establishes and Agree relation with the direct object. The Case features of both pp and direct object are unvalued.
  • v1 is merged, and Agrees with pp.
  • Si is merged in spec, v1.
  • v1 values pp’s Case features, but pp cannot value v’s phi-features because it is phi-incomplete. Thus, v1 establishes an Agree relation with the direct object and values its Case. The object gets Accusative.
  • Si cliticizes, and so it cannot establish any relation with T. The direct object raises via clitic movement and cliticizes on the auxiliary.
  • The EPP feature on T is checked by pro.
  • The verb in T gets the third person singular ending as a result of lack of agreement with a DP.
  • 3.4. An extra dative

    In the previous sections I have proposed a model for the analysis of transitive si constructions. Before turning to other verb classes a further observation needs to be made. The core idea of my proposal is that si in V-O agreement constructions is dative. The following sentence seems to constitute counterevidence for my statement:

    Figure 37.

    In (57), mi is clearly dative. How is it possible then that si is also dative? I can give a tentative answer by saying that mi in (57) is not a real dative, but a so-called ethical dative, which is not a real case. It is a well-known fact that a dative can always be converted into a PP of the form a +DP in Italian. This is possible for sentences like (58), where the dative is a Benefactive:

    Figure 38.

    In (57), however, substituting a PP for the dative mi doesn’t give us the same acceptability as in (58):

    Figure 39.

    This shows that the only dative pronouns can be sticked in an impersonal si construction. A characteristic of ethical datives in Italian is that they can only be realized by pronouns. Therefore, if (57) is indeed an instance of ethical dative, it constitutes no counterevidence to my proposal.

    4. Impersonal si with unergative and unaccusative verbs

    Unergative verbs show an interesting difference when compared to unaccusatives. The agreement patterns of the present tense in impersonal si constructions resemble those of unaccusatives. The past tense is instead different, for pp is also singular, and not plural as in the case of unaccusatives.

    The present tense of an unergative impersonal si construction is shown in (60):

    Figure 40.

    In (60) the verb shows the default 3rd singular ending. The past tense of (60) is (61):

    Figure 41.


    In (61), the auxiliary shows the default 3rd singular ending and the participle shows the default singular masculine ending7.

    The present tense of an unaccusative impersonal si construction is shown in (62):

    Figure 42.

    In (62), just like in (60), the verb is at the present tense and shows the default 3rd singular ending. However, the past tense of an unaccusative impersonal is different from the past tense of an unergative construction, as shown in (63):

    Figure 43.

    In (63), the auxiliary shows the default 3rd singular ending while the participle is plural masculine.

    In the previous section I have proposed an analysis for impersonal si constructions with transitive verbs. This section will be devoted to the analysis of unergative and unaccusative impersonals.

    4.1. Impersonal si with unergatives

    For the analysis of impersonals with unergative verbs I follow the lines proposed by Hale and Keyser (1993)[20], according to which unergatives are actually transitives with the direct object (theme) incorporating into the root by conflation. I argue that the object is syntactically projected, but it has no phonological realization.

    As shown in the previous section, the agreement patterns of unergatives resemble those of unaccusatives only partially. In the present tense, the agreement patterns are the same as those of unaccusatives:

    Figure 44.

    In (64) si is merged in the specifier of the only v available. The phi-features on v are valued by the ‘null’ direct object. Si doesn’t trigger any Agree relation because it cliticizes on T as soon as T is merged.

    In the past tense, the past participle shows a masculine singular ending, and the auxiliary is singular:

    Figure 45.

    I argue that the inflection of pp is due to its agreement with the cognate object, which, being phonetically non-realized, triggers the default third singular agreement on the pp. The derivation of (65) runs as follows:

  • The direct object is merged with the verb. The past participle head is merged, and it Agrees with the direct object, having its phi-features valued (i.e. getting the masculine singular ending).
  • v is merged, and si is merged in its specifier.
  • The auxiliary is merged in T, and si immediately cliticizes on it. Its Case features are valued by incorporation/cliticization on the auxiliary.
  • The EPP on T is checked by pro.
  • The verb on T shows the default ending because of lack of agreement with a DP.
  • (66) [TP pro [T sii –è [vP ti [ppP telefonatoj [VP tj (DO)]]]]]

    The derivation of (65) is also illustrated in (67):

    Figure 46.

    4.2. Impersonal si with unaccusative verbs

    In the present tense, the finite unaccusative verb shows the default third singular ending, as in (62), here repeated as (68):

    Figure 47.

    Following Kratzer (1994)[37], I assume that unaccusative verbs have no v projection. Si is merged in the internal argument position. Evidence that si is generated in complement position is offered by the agreement patterns of unaccusative impersonals in the past tense, where the participle exhibits a plural masculine ending. This is not explainable in other ways than with a pp-si agreement. The derivation of (68) runs as follows:

  • Si is merged with the verb.
  • T is merged, and the verb raises there. Si cliticizes on the verb in T, and thus it cannot value the phi-features on T.
  • Si is merged with the verb.
  • T is merged, and the verb raises there. Si cliticizes on the verb in T, and thus it cannot value the phi-features on T.
  • (69) [TP pro [T sii –arrivaj [VP tj ti ]]]

    The derivation in (68) is also illustrated in (70):

    Figure 48.

    In the past tense, as I have already pointed out, there is a mismatch in number between the auxiliary and the past participle:

    Figure 49.

    The plural ending on the past participle is given by its agreement with si. Si is merged in complement position, and (71) is derived as follows:

  • The past participle is merged with si. Si Agrees with pp and values its phi-features, which get the plural masculine inflection. pp is phi-incomplete, though, and it cannot value Case on si.
  • The auxiliary is merged on T.
  • Si cliticizes on T and can no longer enter any Agree relation.
  • The auxiliary shows the default ending as a result of lack of Agree.
  • pro checks the EPP on T.
  • The Case feature on si is valued by the incorporation of si into T.
  • (72) [TP pro [T sii –è [ppP ti arrivatij [VP tj ti ]]]]

    Figure 50.

    7. Conclusions

    In this paper I have examined some peculiar agreement patterns for impersonal si constructions in Italian. I have suggested a strictly derivational analysis (Chomsky 1999)[3], proposing that the syntactic structure of impersonal si constructions with transitive verbs can be analyzed according to the patterns outlined by Anagnostopoulou (2000)[1] for double object constructions. What differentiates double object constructions with a Benefactive si from impersonal si constructions is the nature of si, which is non-referential in the case of anaphoric si and referential in the case of impersonal si. All the various agreement patterns derive from si’s double nature as head and DP (Chomsky 1995)[2] and from locality conditions. There is no need to postulate special properties of si, such as absorption (or withdrawal) of Case or Theta-role, which are not shared by other clitics. Finally, I have examined the unergative-unaccusative puzzle, and I have proposed an analysis that accounts for the singular-plural alternation on the past participle in terms of lack vs. presence of agreement of pp with si.


    I am indebted to the following people for helpful comments and fruitful discussion: Peter Ackema, Artemis Alexiadou, Adriana Belletti, Fabian Heck, Tanya Reinhart, Luigi Rizzi, Ian Roberts and Manuela Schönenberger. I would like to thank an anonymous reviewer for his/her careful comments and generous suggestions.


    1. "On Nominative Case Features and Split Agreement" Alexiadou A.. In: Brandner E., Zinsmeister H., eds. New Perspectives on Case Theory.. .
    2. On Double Object Alternations and Clitics Anagnostopoulou E.. Massachusetts: University of Crete; 1999.
    3. "Two Classes of Double Object Verbs: the Role of Zero Morphology" Anagnostopoulou E.. Zero Morphology.2000.
    4. Incorporation: a Theory of Grammatical Function Changing. Baker M.. Chicago: University of Chicago Press; 1988.
    5. The Polysynthesis Parameter Baker M.. Oxford: Oxford University Press; 1996.
    6. "The syntax of ne: Some Theoretical Implications" Belletti A., Rizzi L.. The Linguistic Review .1981;1:117-154.
    7. "Morphological Passive and pro-drop: the Impersonal Construction in Italian" Belletti A.. Journal of Linguistic Research.1982;2:1-34.
    8. Problèmes de linguistique générale Benveniste E.. Gallimard; 1999.
    9. Italian Syntax Burzio L.. Dordrecht: Reidel; 1986.
    10. "Anatomy of a Generalization" Burzio L.. In: Reuland E., ed. Arguments and Case. Amsterdam: John Benjamins; 2000 .
    11. "The Variability of Impersonal Subjects" Chierchia G.. In: Bach E., Jalinek E., Kratzer A., Partee B., eds. Quantification in Natural Languages. Ordrecht: Kluwer; 1995 .
    12. "A Minimalist Program for Linguistic Theory" Chomsky N.. In: Hale K., Keyser S. J., eds. The View from Building 20. Massachusetts: MIT Press; 1993 .
    13. The Minimalist Program Chomsky N.. Massachusetts: MIT Press; 1995.
    14. “Derivation by Phase” Chomsky N.. Massachusetts: MIT; 1999.
    15. “On Si Constructions and the Theory of arb” Cinque G.. Linguistic Inquiry .1988;19:521-581.
    16. Gender Corbett G.. Cambridge: Cambridge University Press; 1991.
    17. “Syntactic Configurations and Reference: se/si in Romance” Dobrovie-Sorin C.. In: Zagona K., ed. Grammatical Theory and Romance Languages. Amsterdam: John Banjamins; 1996 .
    18. “Impersonal se Constructions in Romance and the Passivization of Unergatives” Dobrovie-Sorin C.. Linguistic Inquiry.1998;29:399-437.
    19. “Se-si Type Anaphors” Dobrovie-Sorin C.. 1999.
    20. “Unaccusative Syntax and Verbal Alternations” Embick D.. In: Alexiadou A., Anagnostopoulou E., Everaert M., eds. The Unaccusativity Puzzle. Oxford: Oxford University Press; 2000 .
    21. Argument Structure Grimshaw J.. Cambridge: MIT Press; 1990.
    22. “On Argument Structures and the Lexical Expression of Syntactic Relations” Hale K., Keyser S. J.. In: Hale K., Keyser S. J., eds. The View from Building 20. Cambridge: MIT Press; 1993 .
    23. "Expletives and Agreement in Scandinavian Passives" Holmberg A.. Working Papers in Scandinavian Syntax .2000;65:35-64.
    24. “Facets of Romance Past Participle Agreement” Kayne R.. In: Benincá P., ed. Dialect Variation and the Theory of Grammar. Proceedings of the GLOW Workshop in Venice. Dordrecht: Foris; 1989 .
    25. Parameters and Universals Kayne R.. Oxford: Oxford University Press; 2000.
    26. “The Event Argument and the Semantics of Voice” Kratzer A.. Amherst: University of Massachusetts; 1994.
    27. “Past Participle Agreement with Pronominal Clitics and Auxiliary Selection in Italian and French” Law P.. In: Proceedings of Going Romance 2001. 2002 .
    28. “The Syntax of Object Clitics: si in Italian Dialects” Manzini M. R., Savoia L.. In: Salvi G., ed. Festschrift Renzi. 2000 .
    29. “Case and Licensing” Marantz A.. In: Westphal G., Ao B., Chae H. R., eds. Proceedings of the Eigth Eastern States Conference on Linguistics. Ohio State University: Department of Linguistics; 1991 .
    30. “Implications of Asymmetries in Double Object Constructions” Marantz A.. In: Mchombo S. A. , ed. Theoretical Aspects of Bantu Grammar 1. California: CSLI Publications; 1993 .
    31. “Reflexive External Argument and Lethal Ambiguity” McGinnis M.. In: Curtis R., Lyle J., Webster G., eds. Proceedings of WCCFL 16. California: CSLI Publications; 1997 .
    32. Reflexive Clitics and the Specifiers of vP McGinnis M. Massachusetts: MIT Press; 1999.
    33. “What Applicative Heads Apply to” Pylkkänen L.. In: Minnick M., Williams A., Kaiser E., eds. Proceedings of the 24th Annual Penn Linguistics Colloquium. Upenn Working Papers in Linguistics, 7.1; 2000 .
    34. “Object Agreeement in the Impersonal -se Passive Construction in European Portuguese” Raposo E., Uriagereka J.. In: Dziwirek K, Farrell P., Mejias -Bikandi E, eds. Grammatical Relations: A Cross Theoretical View. Satnford: SLA; 1990 .
    35. “Against the Unaccusative Analysis of Reflexives” Reinhart T., Siloni T.. Massachusetts: UIL-OTS Utrecht-Tel Aviv University; 1999.
    36. The Theta system : Syntactic Realization of Verbal Concepts Reinhart T.. OTS Working Papers in Linguistics.2000.
    37. Arguments and Case: Explaining Burzio’s Generalization Reuland E.. Amsterdam: John Benjamins; 2000.
    38. “Primitives of Binding” Reuland L.. Linguistic Inquiry.2001;32:439-492.
    39. “La montée du sujet, le si impersonell et une règle de restructuration dans la syntaxe italienne” Rizzi L.. Recherches Linguistiques.1976;4:158-185.
    40. “Nominative Marking in Italian Infinitives and the Nominative Island Constraint” Rizzi L.. In: Henry F., ed. Binding and Filtering. London: Coom Helm; 1981 .
    41. Issues in Italian Syntax Rizzi L.. Dordrecht: Foris; 1982.
    42. “Icelandic Finite Verb Agreement” Sigur∂sson H. A.. Working Papers in Scadinavian Syntax.1996;57:1-46.
    43. “The Locus of case and Agreement” Sigur∂sson H. A.. Working Papers in Scadinavian Syntax.2000;65:65-108.
    44. “Reflexives, Pronouns and Subject/Verb Agreement in Icelandic and Faroese” Taraldsen T.. Working Papers in Scandinavian Syntax.1994;54:43-58.
    45. “On Agreement and Nominative Objects in Icelandic” Taraldsen T.. In: Haider H, Olsen S., Vikner S, eds. Studies in Comparative Germanic Syntax. Dordrecht: Kluwer; 1995 .
    46. “Impersonal Passives: A Phase-Based Analysis” Svenonius P.. Massachusetts: University of Tromsø; 2000.
    47. “Noun Incorporation” Van Geenhoven V.. Glot International .2001;5:261-271.
    48. “Case and Grammatical Functions: the Icelandic Passive” Zaenen A., Maling J., Thrainsson H.. Natural Language and Linguistic Theory.1985;3:441-483.