2.1 Venetan in Brazil as a heritage language

Most syntactic accounts of northern Italo-Romance varieties are based on varieties spoken in Italy. This is true also for Venetan, a north-eastern Italo-Romance variety which has been studied mainly in the region of Veneto, where the Central Venetan varieties are spoken. However, speakers of this language are also found in other countries, where Venetan is an autochthonous variety (coastal areas of Slovenia and Croatia) or a heritage language spoken by communities of emigrants and their descendants (Belgium, Germany, North and South America). This study focusses on BV, a variety of Venetan that has been spoken in southern Brazil since the mid-nineteenth century, as a consequence of a number of migration waves from north-eastern Italy that involved almost half a million native speakers of Venetan moving to Brazil. The language is still spoken in various communities of descendants of the original immigrants. As a consequence, Venetan is still spoken by second and third generation immigrants, more than half a million native speakers in total.

Benmamoun, Montrul & Polinsky (2013) use the term heritage speakers to refer to second generation immigrants, the children of original immigrants, living in a bilingual environment from an early age. This applies to our case, in that the majority of speakers of BV are sequential bilinguals, in Montrul’s (2008) terms: they are only (or mostly) exposed to BV until they go to school, where BP is introduced. More precisely, these speakers have BV as their first language (L1, the language they firstly acquired) and BP as their second language (L2, the language they acquired later); however, it’s rather exceptional for adult speakers to use predominantly BV. In other words, their L2, BP, becomes their primary language, i.e. their dominant language, while their L1, BV, becomes their secondary language.

The data used in this study are the result of different tests which involved 16 speakers from three different Brazilian states: Rio Grande do Sul (henceforth ‘RS’), Santa Catarina (‘SC’) and Paranà (‘PR’). All the speakers are second and third generation native speakers, grown up in small communities of BV speakers. These communities are not connected to each other; nevertheless, the patterns of change we find are similar in all of them. This suggests that we are in presence of a case of contact-induced change that affects structures of the L1 when contact with the L2 takes place.

2.2 Contact-induced change and cross-linguistic influence

We are describing a situation of intensive language contact which may lead to significant changes in the language of heritage speakers. As Montrul (2004) points out, we might expect their language to be affected by the condition of intense exposure and use of the majority language. Our hypothesis is that the changes we identify in a heritage language like BV are not endogenous, but rather the product of language contact. We will see that this is true in some cases, while in other cases we have to be careful in defining a change as contact-induced.

Following Hulk & Müller (2000), we expect a certain degree of cross-linguistic influence, intended here, in their terms, as the ability of bilingual children to process two different languages separately without excluding the possibility of one language influencing the other. Starting from Platzack (2001), particular grammatical domains have been identified that are most likely to be affected by mutual influence; these are usually at the interface between two modules of grammar. This idea has been developed in the works of Sorace and Serratrice (2009) and Sorace (2011) on the Interface Hypothesis, intended as the observation that language structures that involve an interface between syntax and another domain are more difficult to acquire than structures involving only syntax. The left periphery of the sentence, where the interface between syntax and information structure is encoded, is particularly affected. Another fundamental point which we have to consider in our analysis is the presence of minimally different syntactic structures in the L1 and in the L2: given their similarity, we propose that minimally differing structures are more prone to contact-induced change.

As pointed out in 2.1, most accounts of Venetan are based on Central Venetan varieties spoken in Italy, where the language is spoken in contact with Italian. Venetan is nevertheless spoken in contact with other languages as well.

In this study we focus on the system of subject pronouns of BV and on the availability of null subjects. In the following sections we will see that, among the detected changes, only one of them is to be intended as the direct consequence of contact between Venetan and the varieties of BP spoken in the South of the country (Catarinense, Paranaense, Sul-riograndense).

3.1 The twofold system of subject pronouns of Central Venetan

The system of subject pronouns of Central Venetan is made up of two different paradigms: a tonic one and a clitic one.

Central Venetan tonic pronouns are strong pronouns, according to the traditional tripartite model initially proposed by Cardinaletti & Starke (1999, henceforth ‘C&S’). On a par with their Italian or Spanish counterparts, Central Venetan tonic pronouns are stress-bearing elements that can appear in isolation (1a), can be coordinated (1b) or modified (1c). This is exemplified by Trevigiano, a Central Venetan¹ variety:

On the other hand, in the terms of C&S, Central Venetan clitic pronouns are deficient elements which cannot appear in isolation (2a), be coordinated (2b) or modified (2c).² Besides, SCLs double full subjects, carrying the same featural specification (2d).

We follow C&S in saying that strong and clitic subjects have different distributional properties that reflect a fundamental difference in their syntax: strong tonic subjects are full DPs, while SCLs are heads encoding only the φ-features of the subjects.

The two sets of Central Venetan subject pronouns, at least as far as declarative sentences³ are concerned, are exemplified in Table 1.

As we can see in Table 1, a relevant difference between strong and clitic pronouns is the presence of gaps⁴ in the paradigm of SCLs. The presence of gaps in the paradigm is fundamental for our study of the system of BV, as we see in Section 4.

Another crucial matter in our study concerns the role of SCLs in northern Italo-Romance varieties, which has been connected to pro-drop and to the availability of null subjects (‘NS’ henceforth) in a wide number of studies (Brandi & Cordin 1981, 1989; Rizzi 1986; Benincà 1994; Poletto 1993, 2000; Roberts 2010). These analyses provide evidence that SCLs do not behave like actual pronouns, but rather as agreement markers which share their φ-features with the subject and are realised inside the inflectional field, as shown in (3a–b).

This means that SCLs do not have the syntactic role of a subject; actually, the example in (3a–b) shows that, when SCLs do not co-occur with a DP subject or full pronoun subject, the “real subject” is null (see below), just like in Standard Italian.

Alexiadou and Anagnostopoulou (1998) propose that SCLs in Central Venetan contribute to satisfy the EPP, whereas in Standard Italian this is done exclusively by verbal morphology, through verb raising. They propose that Italian verbal morphology includes a nominal feature. For this reason, in Italian, verb raising is sufficient to check the nominal feature included in verbal morphology. On the other hand, verb raising in northern Italo-Romance may be not be directly triggered by EPP-checking. In these varieties the EPP is checked with the merge of a subject clitic on the inflection. In the remainder of the paper we follow Alexiadou and Anagnastopoulou in their analysis of EPP-checking. In Section 3.2 we extend this approach to BP.

3.2 Subject realisation in Central Venetan and in Brazilian Portuguese

In this section we compare the realisation of Central Venetan SCLs and BP subject pronouns.

BP, the contact variety relevant to this study, has been analysed as a partial NS language.⁵ Barbosa, Kato & Duarte (2005) show that spoken BP displays a higher number of overt pronominal subjects and that this might be related to the fact that the (reduced) verbal morphology in BP is not able to check the EPP anymore. In Alexiadou & Anagnostopoulou (1998)’s terms, BP is becoming a language where the EPP needs to be satisfied by an XP in the subject position.⁶

To put it differently, what Barbosa, Kato & Duarte (2005) suggest is that Spec-TP has to be filled by an overt pronominal element in BP matrix clauses. What seems to be most interesting is that in spoken BP this pronoun is generally a weak and phonologically reduced counterpart of the strong pronoun. When the strong pronoun is found in a left dislocated position inside the complementiser field, the weak pronoun can co-occur with it. In Table 2 we propose an adaptation of the model presented in Kato and Duarte (2014):

This development does not seem different from what happened in Old French, where weak subject pronouns started to double the subject, while at the same time the tendency to systematically fill subject positions with these items increased. Both in French and in BP we can observe a reduction of verbal morphology (see also Salvi 2011).

In the remainder of this paragraph we will see that the BP weak pronominal subjects seem to behave like those of Standard French⁷: unlike the Central Venetan SCLs described in 3.2, BP weak subjects are DP pronouns, behaving like phrases rather than heads. In the remainder of this paragraph we will present some relevant differences in the behaviour of Central Venetan SCLs and BP weak pronouns, on the basis of their distribution and position in the sentence.

The syntactic configuration of BP is therefore different from that of Central Venetan. As we have seen in this section, Central Venetan SCLs are bundles of features realised on T, while the subject position can be null. In BP, on the other hand, weak pronominal subjects have to occupy Spec-TP, which cannot be null.

Evidence that we are dealing with two different elements and two different subject positions in Central Venetan and BP can be found in negation and coordination, two tests originally proposed by Rizzi (1986) to ascertain the phrasal or head nature of clitics. The first test concerns the position of the clitic with respect to preverbal negation: a phrasal subject pronoun normally comes before preverbal clitic negation, while an agreement marker usually follows it.

(5) shows that in BP, the subject pronoun precedes the clitic negation não and does not occur in the same position as the dative clitic⁸: the subject is realised in a different syntactic position; in (6) we see that in Trevigiano (Central Venetan), on the other hand, the order with respect to the clitic negation no is inverted and the SCL cliticise onto the verb together with the dative clitic:

This difference is evidence for the fact that the element used in spoken BP is a weak pronoun, while the one used in Central Venetan is an agreement marker: if we consider that the negation clitic and the SCL form a clitic cluster on T, we may expect the SCL to follow it, as in (6); the inverted order (5), on the other hand, suggests that the position of the pronoun in the structure is different.⁹

The second test regards the obligatory repetition of SCLs in both conjuncts in coordination: an agreement marker must be repeated in both conjuncts in coordination (7a), while a weak pronoun is realised only in the first conjunct in unmarked contexts (8), as shown by Trevigiano and spoken BP examples, respectively:

The examples are consistent with Rizzi’s test on coordination, in that in (7a) we see that the SCL of Trevigiano is repeated in both conjuncts, while in (8) we see that the weak pronoun of BP is not repeated in the second conjunct. This is due to the fact that in (7a) the SCL is repeated every time a T head appears, as it is part of the verbal morphology, while this is not the case in (8). A structure like (7b), in which the SCL is not repeated in the second conjunct, is impossible in Trevigiano.

It has been pointed out by Vanelli (1998) that medieval northern Italo-Romance varieties did not have two series of subject pronouns, but just one set of strong subjects. These varieties have consequently developed a set of weak DP subject pronouns, parallel to those of BP and Standard French and thus went through a stage where they did not allow for NSs anymore; Central Venetan weak subjects have been consequently further reduced to agreement markers, and this represents the current stage of development of this variety.

In this section we presented a series of facts related to the realisation of the subject in the varieties included in this study. It has been pointed out that Central Venetan has a twofold system of subject pronouns: tonic pronouns on the one hand and SCLs behaving like agreement markers on the other hand; it has been also shown that BP requires an overt subject in most syntactic environments in matrix clauses and that it is developing weak subject pronouns to perform this function. On the other hand, null subjects in BP always require a syntactic antecedent, which is the closest c-commanding DP.¹⁰ In Section 4 we will explore two changes in the system of subject pronouns of BV, trying to identify possible causes behind the change. In particular we will try to demonstrate that contact between two minimally different syntactic systems can determine changes in the grammar that do not follow the expected path of development for a given structure.

4.1 The first person pronoun “mi” in Brazilian Venetan

Recall that in Central Venetan the two sets of pronominal subjects – tonic and clitic – are substantially different both in their structure and in their distribution: tonic subject pronouns are phrases, normally used when topicalised, while SCLs are agreement markers, obligatorily realised on T. Between the Central Venetan and the BV system of subject pronouns there are two striking differences regarding the distribution and the structure of the two paradigms of subject pronouns.

The first difference between Central Venetan and BV is the generalised realisation of the first person singular full pronoun even in non-contrastive environments (recall that Central Venetan has no SCL for the first person). In an analysis of spontaneous speech, the rate of overt 1sg pronominal subjects realised by our speakers of BV is 63,3%, of which 78,9% are used non contrastively; on the other hand, testing the same sentences on speakers of Trevigiano (Central Venetan), we noticed that the rate of overt 1sg pronominal subjects amounts to 13,3% and all of them are used contrastively. Recall, from Table 1, that mi in Central Venetan is a tonic pronoun; first person singular represents a gap in the set of SCLs, as there is no clitic element for first person singular.

Since mi is a tonic pronoun in Central Venetan, we would expect mi to be realised mainly as a topicalised element bearing contrastive stress, as Central Venetan is an NS language. However, in (9) and (10) we see that the use of mi in BV is much wider than in Central Venetan, where mi would be most probably interpreted as a topic. Example (11) shows that in Trevigiano the subject pronoun is unacceptable in non-contrastive environments.

An additional difference that emerges from the contrast between (9) and (11) lies in the imperfect verbal morphology. While modern Trevigiano has developed the ending -o for 1st person singular imperfective (iero in 11, possibly because of the contact with Italian), BV preserved a generalised -a ending for all persons (gera in 9) as in archaic Venetan varieties.

We know from Section 3 that the reduced verbal morphology has also been linked to the loss of pro-drop properties since the formulation of the parameter by Rizzi (1982). We propose therefore that the discourse conditions under which mi is realised in Central Venetan are no longer applied in BV. While in Central Venetan mi is a tonic pronoun, realised usually in cases of topic shift, in BV mi is also used as a weak pronoun, it is phonologically reduced, it does not bring any additional discourse-related meaning and it is realised quite regularly in main sentences. In Section 4.2 we show that the same interpretation that we gave for mi (a weak pronoun) can be given for BV third person pronouns as well.

In the previous section we saw that BP no longer allows for NSs in many contexts (Barbosa, Kato & Duarte 2005) in matrix clauses. One hypothesis we need to investigate is the possible influence of BP on the availability of NSs in BV. The change we identified in the realisation of mi as a weak pronoun, often realised in main sentences in BV, may be the consequence of the contact with BP, a system which requires an overt subject even in non-contrastive environments.

In Section 5 we will show that this is not the case. Before that, we discuss the change we detected in the syntax of SCLs in BV.

4.2 The system of subject clitics of Brazilian Venetan

In 3.2 we concluded that Central Venetan SCLs are not real pronominal elements, but rather agreement-like elements realised on a T head. This means that they do not occupy the canonical subject position, Spec-TP. Therefore Central Venetan, like Italian, is a canonical NS language. What differs is the way the EPP is checked in the two varieties: while in Italian this is checked by verbal morphology, in Central Venetan this has to be supplemented by the incorporation of an SCL encoding the φ-features of the subject.

Despite being obligatorily realised in most contexts, the use of Central Venetan SCLs in matrix clauses has some restrictions. In particular, also for SCLs we can identify some interface conditions in which third person SCLs are not realised in the presence of a non-dislocated full DP subject. According to Benincà (1994) the example in (12) is grammatical only if the DP Giorgio is interpreted as a topic, otherwise the SCL is normally absent.

The realisation of third person SCLs, depends, in other words, on the interpretation of the DP subject as a topicalised element. This brings us back to what we said in 4.1 with respect to the interface between syntax and discourse. If SCL realisation partially depends on discourse properties of the DP subject, we can say that the realisation of SCLs in (9) is related to interface conditions.

Turning now to BV, we show that in addition to the Central Venetan pattern, it displays also a pattern in which the SCLs resemble the BP weak pronouns behaviour in matrix clauses. In (13) we see that the condition on the realisation of third person SCLs in presence of a non-dislocated full DP subject that we showed in (12) can be applied to BV as well:

Although very similar, SCLs in BV and Central Venetan present some distributional differences suggesting that the status of SCLs in BV is different from that of Central Venetan. This difference might be due to the nature of the SCL (pronominal or inflectional). In order to identify this difference, we need to go back to the two tests by Rizzi we used to distinguish BP weak pronouns from Central Venetan SCLs: preverbal negation and repetition in coordination.

As far as the preverbal negation test is concerned, despite the presence of the regular Negation-SCL order, several cases of SCL-Negation order are attested (14) in BV; on the other hand, only the Negation-SCL order is accepted in Trevigiano (15a–b):

In (14) negation follows the subject clitic el. This shows that the SCL is higher than negation in the structure and suggests that BV SCLs may also be structurally different from Trevigiano SCLs (15a), which follow the clitic negation. This is consistent with the idea that SCLs, if non-pronominal, follow the negation (Rizzi 1986).

BV, on a par with Central Venetan and spoken BP, also has a set of tonic pronominal subjects. It is worth pointing out that some BV speakers would use the tonic subject (16a) to mark the topic shift, while the clitic in (16b) is used in non-contrastive contexts, since it cannot be dislocated in the left periphery:

In (16a) the full pronoun is considered as a topicalised subject, while in (16b) the clitic suggests that there is no topic shift; this becomes clear in (16c–d): we can say that only the form lu can be used in case of a topic shift (16c), while the dislocated use of the form el (16d) leads to ungrammaticality.¹² What is relevant to our study is the fact that the SCL, like the tonic pronoun, comes before preverbal negation, which could represent a change in the position of the clitic in the structure with respect to the original Central Venetan order. The test on the position of the SCL with respect to preverbal negation provides evidence that the change we detected is indeed affecting syntax.

As for the repetition of the SCL in coordination, the situation seems, again, different from that of Central Venetan.

In addition to the Central Venetan pattern, with the SCL repeated on both conjuncts, there are examples of coordinated structures that we can hardly reconcile with the description we gave of Central Venetan SCLs in terms of Rizzi (1986):

In (17) we have a DP subject in the first conjunct and no SCLs realised in either of the coordinated conjuncts. Unexpectedly the structure in (18) follows the same model presented in (17): the subject is realised only once, without being repeated in the second conjunct despite the fact that in (17) the subject is a full DP, while in (18) it is an SCL. According to the analysis we have adopted, SCLs should be realised in both conjuncts in coordination if they are not real pronominal elements, but inflectional elements.¹³ BV does not obey this constraint, differently from Central Venetan: (18) seems to resemble the BP distributional scheme for weak pronouns that we saw in (8), repeated below as (19):

This distribution suggests that the SCL in (18) is not occupying the expected position in the inflectional field, as we assumed so far for Central Venetan: BV SCLs are not used as agreement markers on the model of Central Venetan, but are weak pronouns, on the model of BP reduced pronominal subjects.

Following Alexiadou and Anagnostopoulou (1998) we propose that the EPP in BV can be checked by a weak pronominal subject located in Spec-TP as in (20a), while, according to the same authors, in Central Venetan this is done by an agreement marker realised on T as in (20b).

Central Venetan SCLs are normally realised proclitically to their verbal host as in (20b), and they cannot be separated from them by other phrases. In the remainder of this section we show that in BV we have cases in which an adverb can appear in between the SCL and the verbal host, suggesting once again that the correct interpretation of BV is (20a).

It is known that a clitic element can only be separated from its host by another clitic; this is in fact the case in Central Venetan. We do not expect the possibility of inserting a non-clitic element, like an adverb, between the clitic and its verbal host. This phenomenon, known as interpolation, has been attested for southern Italo-Romance by Ledgeway & Lombardi (2005) and for Triestino, a northern Italo-Romance variety, by Paoli (2008). In BV SCLs we find something similar to what Paoli describes for Triestino:

According to Ledgeway & Lombardi (2005), interpolation is a consequence of the different extent of clitic and verb movement. In our analysis, this means that the weak element can move independently from the verb to Spec-TP, higher than certain classes of low adverbs.¹⁴

There are two possible explanations for this interpolation: the first possibility is that the verb moves lower in BV than in Central Venetan; the second is that the adverb is topicalised. However the second hypothesis implies that el is in CP; as shown in examples (16c–d) el cannot be dislocated, so we propose that the correct analysis is a difference in the extent of verb movement in BV, which moves lower than in Central Venetan. Interpolation is not a possibility in the Central Venetan varieties which form the base from which BV emerged, so we should consider this as a consequence of contact as well. A number of studies has shown that the verb in BP moves lower than in European Portuguese, using temporal adverbs as evidence supporting this thesis. Particularly relevant to our analysis is the hypothesis presented in Ambar, Gonzaga & Negrão (2004) and Cyrino (2013). What these authors show is that temporal adverbs are always found preverbally in BP:

Cyrino (2013) argues that T consists of two positions, a higher T1 and a lower T2 position, and that in BP verbs can move only as high as T2, which is the reason why they appear after temporal adverbs.¹⁵

What we propose for BV is that there is more than one change involved: on the one hand, SCLs are not realised as agreement markers, but rather as phonologically weak pronouns, on a par with BP weak pronouns and Standard French SCLs; on the other hand, verb movement has somehow decreased in BV, under the effect of contact with BP, along the lines of Cyrino (2013).

Going back to the change in the structure of SCLs, from agreement markers to weak pronouns, this degrammaticalisation is somehow anomalous and unexpected: in the model of C&S this would mean that SCLs are reversing their process of weakening, regaining a part of the structure. The degrammaticalisation process is due to contact-induced change, whereby BV is following BP on its path becoming a non-NS language. We will analyse this process in Section 5.

5.1 Attrition, incomplete acquisition and interface conditions

When we talk about attrition in L1 we indicate the loss of linguistic skills in the first language due to exposure to another language; a feature that was initially acquired in the native language is consequently (partially) lost due to prolonged exposure to another language. Incomplete acquisition, on the other hand, indicates a situation where a feature was never acquired completely. According to Polinsky (2016) both phenomena may emerge in the heritage language, as well as in the dominant language. Recall that BV underwent two changes in the NS domain: the introduction of weak mi and the transformation of SCLs into weak pronouns. To link Polinsky’s hypothesis to our case study, the realisation of first person singular mi subject pronouns in BV in non-contrastive environments could be the consequence of attrition or incomplete acquisition.

Attrition is normally found in the first generation of immigration. In our case, our group of speakers is made up of second and third generation immigrants, all heritage speakers that were exposed to a language that was, in the first generation of immigrants, affected by attrition. It is therefore impossible, at this stage, to establish whether attrition or incomplete acquisition played a role in the change.

What we know is that the change in the conditions on the overt realisation of 1sg mi in BV involves a simplification of interface conditions, a process which is quite common in bilingual speakers, as the studies we present in the remainder of this paragraph show. These conditions can be applied to heritage speakers as well, as we have seen in the case of BV speakers.

Hulk & Müller (2000, 2001), for instance, claim that cross-linguistic influence is possible in bilingual children, at the interface between syntax and discourse, where a language A has a syntactic construction which seems to allow more than one syntactic analysis, while language B allows only one analysis for the same structure. We think that the interface conditions (in particular topic shift) under which overt mi is realised in Central Venetan are more difficult to acquire than the purely syntactic conditions under which weak pronominal subjects are realised in BP. It is thus tempting to define the change taking place in BV with respect to the overt realisation of 1sg pronominal subjects as the consequence of the contact with BP.

Sorace, Serratrice, Filiaci and Baldo (2009) studied the realisation of overt subjects in Italian-Spanish and Italian-English bilingual children. Their study evidenced that, regardless of whether the first language is an NS language (like Spanish) or not (like English), the rate of overt subjects in bilingual speakers of Italian is higher than the one observed in monolingual Italian speakers.¹⁶ Sorace & Serratrice (2009) have also claimed that there is a difference between internal interfaces (the ones involving formal features within syntax, which are acquirable in L2, e.g. the syntax-semantics interface) and external interfaces (the ones involving pragmatic conditions, which are problematic in L2 acquisition, e.g. the syntax-discourse interface). The Interface Hypothesis, as formally defined by Sorace & Serratrice (2009) and Sorace (2011), can be applied to our study: language structures that involve an interface between syntax and another domain are more difficult to acquire than structures involving only syntax. This proves that the change we detected in the realisation of mi as a weak pronoun, which is regularly realised in non-contrastive contexts, is not a direct consequence of the contact of an NS language (like Central Venetan) with a partial NS language (like the contexts of BP we analysed). Rather, it supports the hypothesis that heritage speakers, along with bilinguals, show a simplification of the interface conditions under which NSs are allowed.

Besides, Serratrice (2005) has observed that, despite being unequivocally identified by the context, 1sg is the most frequent pronominal subject, even in cases where it does not serve any contrastive function, in some children acquiring Italian. Pinto (2006) shows that the use of 1sg pronominal subject in Italian-Dutch bilingual children is not always pragmatically adequate, as 1^st person pronoun is realised also in cases where no topic shift has occurred; she also points out that the rate of overt 1sg subjects outnumbers that of other subject pronouns. Heritage speakers involved in our case study can be considered as a particular class of bilingual speakers, therefore we are able to apply these considerations also to BV.

In sum, the change we detected with respect to the overt realisation of 1sg pronominal subjects in BV is due to contact, but not to the syntactic configuration of BP, the contact language.

5.2 The syntax of subject clitics in contact

BV SCLs seem to allow for an interpretation that is different from the one that has been given for Central Venetan. The data we presented suggest that speakers tend to reanalyse agreement markers as weak pronouns. The case of BV is particularly interesting because we would not expect a process of degrammaticalisation where SCLs reverse reduction, adding some structure, contrary to what is normally predicted: from heads, hosted in a phrase, they tend to behave as proper phrases. What we introduced in 5.1, the role of the external interface between syntax and discourse in bilingual speakers (Sorace et al. 2009), seems to be fundamental in order to address this question properly as well. In other words, the change and, possibly, the simplification of the contexts where overt subjects have to be realised indirectly affected the syntax of SCLs. The change in the conditions on the realisation of 1sg pronominal subject in BV is not modelled on BP, but we propose that this was the starting point that led to the change we detected in SCLs.

The fact that speakers of BV realise mi as a non-contrastive subject pronoun under the conditions described by the interface hypothesis, may represent a precondition for the reanalysis of the structure of SCLs in heritage adult speakers: contact with BP induced a partial non-NS status in the language. There was no possible candidate to be used as a 1sg non-contrastive pronominal subject, therefore the strong pronominal form was selected. For most of the other persons there are agreement heads the use of which is obligatory and non-contrastive due to their clitic nature, which are reanalysed as weak pronouns. In sum, BV developed a new set of weak pronouns, partially derived from the tonic paradigm (mi) and partially from the SCL paradigm.

In particular, what emerges from the reanalysis of SCLs into weak pronouns is a structural change from Central Venetan to BV. The BV pronominal system is different from the original structure in Central Venetan as well as from the sentence structure of the language of young Venetan-Italian bilinguals in Italy (Klaschik & Kupisch 2016, Casalicchio & Frasson 2018). In Venetan-Italian bilinguals the change in the SCL system does not affect the syntax of SCLs, but rather the conditions on their realisation, which depend on the discourse interface. This situation supports the hypothesis according to which discourse conditions are involved. The case of BV SCLs, on the other hand, is different: the change affects the structure of SCLs, rather the discourse conditions on their realisation.

In conclusion, we can say that the detected change in the syntax of SCLs in BV is the result of contact with BP. In order to be able to say so, we verified that the change has been precisely modelled on the system of reduced pronominal subjects of BP and affected the way agreement is realised in BV. In BV, the EPP is checked by a pronoun in Spec-TP, and not by SCLs on T.

We also proposed that the change in the syntactic structure of SCLs has been probably favoured by another change, namely the reduction of the complexity of the interface contexts in which overt subjects are realised in NS languages.

5.3 The identification of contact-induced change in heritage speakers

The idea behind this study is an analysis of particular conditions that have to be met in order for us to be able to claim which changes are induced by the contact with a given syntactic system and which ones are not.

The case of BV is a relevant example; we identified different tendencies in native speakers of this variety and some of them are, in our view, indeed the consequence of the contact with a syntactic construct that minimally differs from the one we are investigating.

We identified a target group of speakers using the heritage variety, detecting a range of changes that was common among them. We noticed that the difference in the interface conditions under which subjects are realised has been frequently attested in studies on bilingual speakers, regardless of the combination of languages and of the availability of NSs in both of them. The fact that the 1sg pronominal subject tends to be overtly realised even when the topicalised interpretation is absent is therefore not sufficient in order to be intended as the reflex of the contact with BP, but rather as the tendency to simplify interface conditions that are computationally more difficult to learn. Nevertheless, this aspect is fundamental, since it constitutes the starting point in the reanalysis of SCLs as weak pronouns. This change, detected in adult heritage speakers of BV, is strongly dependent on the nature of language contact: an entire set of agreement heads, SCLs, are realised as phrasal subject pronouns in non-contrastive environments; this seems to have been modelled on the BP set of weak pronouns. This change in the syntax of SCLs is anomalous: we know that northern Italo-Romance SCLs underwent a process of reduction, starting out as nominative tonic pronouns, reducing then to obligatory non-contrastive subject pronouns, finally becoming agreement markers, heads realised in the inflection field and encoding φ-features of the subject. We would not expect these elements to reverse the grammaticalisation process to become pronominal again. The only factor we cannot rule out is the direct effect of the contact of weak pronouns of BP on Central Venetan SCLs, which gave rise to the situation we find in BV: a paradigm that uniformly behaves as a set of weak pronouns.

We can therefore conclude that SCLs are reanalysed in BV as phrasal pronouns under the direct contact with BP. This syntactic change was favoured by an initial change in the contexts where overt tonic pronouns are realised, due to the simplification of the external syntax-discourse interface, which is commonly attested in bilinguals.