Subject Clitics in Microcontact: A Case Study from Heritage Friulian in Argentina and Brazil

In this paper we present data from first generation immigrants (G1) and second and third generation heritage speakers of Friulian, a Rhaeto-Romance language spoken in North-Eastern Italy and also found in Argentina and Brazil. The target phenomenon is subject clitics (SCL s). We show that SCL s in heritage Friulian are in a process of being reanalyzed from being agreement markers to pronouns. While SCL s are obligatory in Friulian as spoken in Italy, they are often dropped in heritage Friulian in Argentina and Brazil; this phenomenon, we argue, needs to be interpreted as the drop of pronominal subjects, and not of agreement-like SCL s. We also demonstrate that the use of SCL s (reanalyzed as pronominal subjects) is conditioned both by grammatical factors (it happens more in some grammatical persons than in others) and by discourse factors (they are used more in the case of a continuation topic than in other topicalization contexts). This means that in heritage Friulian, discourse constraints on the expression of subjects are not being lost or weakened; in fact, against the general grammaticaliza-tion trend of pronominal forms, new discourse constraints are introduced.


Introduction
This article is concerned with language change in contact, in particular in situations of unbalanced bilingualism, such as the case of first-generation immigrants and heritage speakers (HS s).
The ever-growing field of heritage language studies has thus far revealed several generalizations regarding heritage grammars (i.e., Benmamoun, Montrul & Polinsky, 2013;Montrul, 2016;Polinsky, 2018). While change can be detected in all domains of grammar, certain domains may be particularly vulnerable. Already at the beginning of this century, various researchers observed that phenomena pertaining to the C-domain are vulnerable in bilingual acquisition (Hulk & Müller, 2000). This observation ultimately led to the formulation of the Interface Hypothesis (Sorace & Filiaci, 2006;Sorace, 2011), which claims that constructions in which syntax interacts with other language modules, such as discourse and pragmatics, are particularly vulnerable in bilingual populations.
Probably the most described phenomenon connected to the Interface Hypothesis has been the expression and the interpretation of subjects in prodrop languages in contact with non-pro-drop languages. Several studies on bilinguals whose weaker language is pro-drop have shown that, while these speakers have some command of the syntactic constraints related to the distribution of null subjects, they tend to overuse overt subjects in discourse contexts where null subjects are expected (e.g., Sorace & Filiaci, 2006;. Tsimpli et al. (2004) and , working on Greek-English and Italian-English bilinguals respectively, attribute this to bilingual speakers opting for the simpler system: while the English system offers only one option, i.e. overt subjects, Greek and Italian systems require the mastery of a more complex interaction between grammatical and discourse factors in order to select the correct, or more appropriate, form of the subject (overt or null). An increasing number of studies showing that heritage speakers do not have complete mastery of pro-drop has been released in the last few years (Carvalho & Bessett, 2015;Dubinina & Polinsky, 2013;Keating, VanPatten & Jegerski, 2011;Montrul, 2004Montrul, , 2008. On the other hand, some studies showed no effect of language contact on the realization and interpretation of null subjects (Flores & Rinke, 2020;Nagy et al., 2011;Rinke & Flores, 2018). In this article, we focus on the expression of subject clitics (SCL s) in heritage Friulian, a Rhaeto-Romance language originally spoken in North-Eastern Italy and also found in Argentina and Brazil. Following the methodology applied in Polinsky (2018), we present data from four first generation immigrants (G1, age of arrival of 10 years or older) and 12 second and third generation heritage speakers (HS s) of Friulian.
While SCL s are obligatory in Friulian as spoken in Italy, we will show that they are often dropped in heritage Friulian in Argentina and Brazil. On the basis of a number of considerations, we will argue that this phenomenon needs to be interpreted as the drop of pronominal subjects, and not of agreement-like SCL s. In other words, in the immigrant varieties, the syntactic status of these SCL s has changed. We also demonstrate that the use of SCL s, reanalyzed as pronominal subjects, is conditioned both by grammatical factors (they are used more in some persons than in others) and by discourse factors (they are used more in the case of topic continuation than in other topicalization contexts). This means that it is not the case that discourse constraints on the expression of subjects are being lost or weakened in these heritage varieties; in fact, new discourse constraints are introduced with respect to the described variety spoken in Italy. These constraints are also found in other pro-drop languages, and as such are not new. They are, however, new for this specific variety.
Before discussing the data and the issue, a premise is in order. The data of heritage Friulian were collected on fieldwork, through interviews with G1 and HS of Friulian. Finding these speakers was not an easy task, both because most G1 and HS have abandoned Friulian altogether, and because the Friulian community is rather scattered. Moreover, G1 speakers are rather elderly, which made it impossible to collect abundant data or ask them to perform a large number of tasks.
When G1 of Friulian left Italy they were mostly monolingual and therefore not fluent in Italian. In Italy, around the 1950s, Friulian entered into systematic contact with Italian. The version of Friulian that G1 emigrants speak stems, however, from a pre-contact situation. To ascertain the changes that emerged due to contact between Friulian and Spanish and Portuguese we would need to compare heritage Friulian with monolingual speakers of Friulian. These monolinguals would have to speak the language that was spoken in the 1950s before it entered in contact with Italian. These monolingual speakers do not exist: Friulian is spoken nowadays only in a bilingual situation; furthermore, it has changed since the time when these monolingual speakers left because of intensive contact with Italian.
Although we are aware that the question of how contact with Italian affected the grammar of modern Friulian is important in the analysis of change in contact, the issue will not be discussed in this article. In the present study, we relied on grammars from the 1950s and '60s, where the variety spoken by G1 Friulian was documented. The Friulian clitic system that we consider in the rest of the article is therefore that which is found in the grammars describing Friulian at the beginning/ half of previous century, which is the time when G1 speakers left Italy to emigrate to Argentina and Brazil. The SCL system described in these grammars (Della Porta, 1922) is identical to that described in more recent studies on the syntax of Friulian (Benincà, 2014(Benincà, , 2015Benincà & Vanelli, 2016). Some minor changes in the phonetics of SCL s do not affect the system that we are examining here.
Given the absence of a proper baseline variety, the reference to the system presented in these studies is the only way we have to show that change happened. Despite this seemingly restricted set of data, we believe that documenting a variety that is close to extinction is of great importance. Furthermore, our aim is to describe the syntax of SCL s in heritage Friulian; given that the speakers' grammars are rather consistent, we take this sample as being representative enough to draw some generalizations and attempt an analysis.
This article is organized as follows. In the next section we discuss the grammatical and discourse properties of pro-drop, followed by an overview of previous research concerning this phenomenon in bilingual populations. Sections 3 and 4 illustrate the pronominal systems in Friulian and in the contact languages, Argentinian Spanish and Brazilian Portuguese, respectively. In Section 5 our data collection is presented, followed by the results in Section 6; Section 7 offers an analysis of null subjects in heritage Friulian. Section 8 contains our conclusions.

Why Pro-Drop Is Called Pro-Drop
Null subjecthood is the phenomenon whereby some languages can leave the overt subject of a clause unexpressed. Null subject languages are usually referred to as pro-drop as it is assumed that the element left out is pronominal. The reason for this assumption is the observation that full DP s usually introduce new referents in a clause, while pronouns do not always (if we leave aside ostension, pronouns must refer to an element already mentioned in the discourse). In well-behaved pro-drop languages, the subject can be left out when referring to an element already present in the discourse, usually a topic (Frascarelli, 2007;Frascarelli & Hinterhölzl, 2007;Kwon & Sturt, 2012;Jimenez-Fernández, 2016;Frascarelli & Jimenez-Fernández, 2019;. In (1), the pro subject is interpreted as being co-referential with the subject of the preceding clause. In (2), an overt pronoun does not necessarily corefer with the subject of the preceding clause, but it can. The postverbal position of the pronoun is more felicitous because it bears a contrastive focus, which is more consistent with the use of an overt pronoun. In (3), the repetition of the full DP can only be licensed if the two DP s have different referents. In particular, the post-verbal, focused position is ruled out with a co-referential reading.
(1) Il professore i è entrato. pro i ha iniziato a parlare the professor is entered pro has started to talk "The professor entered. He started to talk" (2) Il professore i è entrato. Mentre (lui i,j ) iniziava a parlare (lui i,j ), the professor is entered while he started to talk he una ragazza si è alzata di scatto a girl is stood suddenly "The professor entered. When he started to talk, a girl suddenly stood up" (3) Il professore i è entrato. (Il professore ??i,,j ) the professor is entered the professor ha iniziato a parlare (il professore *i,j ) has started to talk the professor "The professor entered. The professor started to talk" These sorts of considerations caused linguists to postulate that the silent element, in sentences like (1), is a pronoun and not an overt DP. Notably, the observation that pro is a pronominal element is taken into account by Rizzi (1982) (Rizzi, 1982, p. 143) Not all pro-drop languages have the same distribution of pro. Some languages are radically pro-drop, others are subject pro drop for all persons, others can drop the pronominal subject only for some persons (see Holmberg, 2005, andD' Alessandro, 2015 for an overview of the distribution of pro).

Conditions on the Interpretation of Pro
Regarding the interpretation of pro, it has been shown that, in case of multiple possible antecedents, pro tends to refer to the sentential subject (Calabrese, 1986;Carminati, 2002), while overt pronouns tend to refer to lower arguments. As an example, consider (4), from Carminati (2002): (4) Marta scriveva frequentemente a Piera quando pro / lei era Marta i wrote frequently to Piera j when pro i,j / she i/j was negli Stati Uniti in the US "Marta frequently wrote to Piera when she was in the US" (83).
While Italian syntax allows pro to potentially corefer with both antecedents, Marta and Piera, Carminati (2002) shows through a number of experiments that native speakers of Italian prefer the interpretation according to which pro refers to the item in Spec, TP (i.e., in the canonical sentential subject position). The overt pronoun selects a lower antecedent instead (in this case, Piera). Carminati's work sets an important milestone in defining the conditions for pro licensing and highlights that null subjects involve more than just syntax.
That the use of pro is not only determined by syntax is shown by several studies, most notably those put forward by Sorace et al. regarding the processing and interpretation of pro,1 and those proposed by Frascarelli (2007), Jiménez-Fernández (2016), Miyagawa (2017) and many others. Sorace's work focuses on L2 acquisition and bilingualism, while Miyagawa, Frascarelli and Jiménez-Fernández focus on pro in monolingual varieties. The conclusion they all reach is, however, rather similar: topicality plays an important role in licensing pro. The pronoun is more likely to be overtly expressed in sentences where there is a shift in topic with respect to the previous discourse. As an example, compare (5) and (6) from Spanish (Montrul, 2004;p. 128): (5) Pepe no vino hoy a trabajar. *Pepe / ?Él / pro estará enfermo. Pepe not came today to work. Pepe / he / pro will be sick. "Pepe did not come to work today. He must be sick." (6) Hoy no fui a trabajar. Pepe / él / *pro pensó que estaba today no went.1sg to work Pepe / he / pro thought that I was enferma sick "Today I did not go to work. Pepe/he thought I was sick." (5) is an example of topic continuation: the topic of the second sentence is the same as in the first sentence, which makes a null subject the most felicitous option. In (6), there is a topic shift: the subject changes from the 1st person to a 3rd person, thus requiring an overt pronoun (or a full DP).

2.3
Pro-Drop in Contact The realization of null and overt subjects in pro-drop languages has been one of the main topics in studies on bilingualism, as mentioned above. Several studies have shown that bilingual populations whose weaker language is a null-subject one tend to overgeneralize overt subjects to pragmatically infelicitous contexts. This tendency is attested in bilingual children (e.g., , adult L2 learners (e.g., Sorace & Filiaci, 2006), L1 attriters (e.g., Tsimpli et al., 2004) and HS s (e.g., Montrul, 2004). This is especially true when one of the languages is not a pro-drop language, like English.
In contrast, some studies on the expression of the subject when two Romance languages are in contact have shown that null subjects may be easily handled by bilingual speakers (Carvalho & Child, 2011;De Souza, Chaves, & Simioni, 2018). In particular, the varieties of Uruguayan Spanish and Portuguese spoken on the border with Brazil display rates of null subjects that pattern with canonical pro-drop languages (like Spanish) rather than with non-pro-drop or partial pro-drop languages (like Brazilian Portuguese).
These data contrast with other studies that demonstrate an overextension of overt subject pronouns even when both languages are pro-drop (Bini, 1993 andSorace &Serratrice, 2009, for Italian-Spanish;Margaza &Bel, 2006, andLozano, 2006 for Greek-Spanish;Guido Mendes & Iribarren, 2007 for Brazilian Portuguese and Spanish). However, as noted by Filiaci, Sorace and Carreiras (2014), it may be the case that the division of labor between null and overt pronouns in the two null-subject varieties in contact is not identical.
The literature also reports a so-called person effect. First person singular seems to be the most frequent overt pronominal subject, even in cases where it does not serve a pragmatic function like, for instance, indicating contrastive focus or a shift in topic. This person effect has been attested both in monolingual acquisition (Serratrice, 2005 for Italian children) and bilingual Frasson, D ' Alessandro and van Osch Heritage Language Journal 18 (2021) 1-36 acquisition (Pinto, 2006 for Dutch-Italian children), as well as in HS s (Schmitz, Di Venanzio, and Scherger, 2016 for Italian and Spanish HS s in Germany).

Pronominal, Clitic, and Null Subjects in Friulian
Friulian is particularly interesting as far as pro-drop is concerned, as it has two series of subject pronouns, one tonic one the other clitic (see Benincà, 2014Benincà, , 2015Benincà & Vanelli, 2016;Poletto, 1993Poletto, , 2000. Friulian tonic pronouns are strong pronouns, according to the traditional tripartite model initially proposed by Cardinaletti and Starke (1999). On a par with their Italian or Spanish counterparts, Friulian tonic pronouns are stressbearing elements that can appear in isolation or can be coordinated. Friulian has generally been analyzed as a canonical null subject language, in that strong pronouns are not obligatory and are normally realized only when they serve some discourse function. However, Friulian differs from canonical null subject languages like Italian in that it has an additional set of obligatory SCL s. These are not real pronominal elements: they are inflectional heads, on a par with verbal agreement endings. These SCL s are also found in other varieties spoken in northern Italy, such as Venetan (Benincà, 1994) and Trentino (Brandi & Cordin, 1989; see also : Poletto, 1993: Poletto, , 2000Rizzi, 1986). The situation is illustrated in (7): Jo o feveli furlan. I I.scl speak Friulian 'I speak Friulian.' In example (7) we see that the overt subject and the SCL co-occur; this is possible because SCL s are inflectional elements and do not function as subjects. Unlike pronouns, they also need to be realized in both conjuncts in coordinated structures, as shown in (8), and do not allow for non-clitic material to be inserted between them and their verbal host, as shown in (9).
These conditions to ascertain the nature of SCL s were first identified by Rizzi (1986) for northern Italo-Romance varieties. More specific studies on Friulian (Benincà and Vanelli, 2015;Gaglia, 2012) show that the system of this language is particularly uniform: SCL s are present with finite verbs in all persons and are realized in all syntactic contexts. As a general rule, if a language has a SCL for a specific person, it must use it: SCL s are not optional.
The only exception is instantiated by clitic clusters: SCL s are not realized when they appear in a cluster (including negation, direct and indirect object, impersonal and reflexive2), as illustrated in (10).
(10) Un sarpint, îr, (*al) lu à copât. a snake yesterday he.scl it.ocl has killed 'Yesterday he killed a snake.' The conditions under which we expect a SCL are quite easily identifiable, which makes Friulian a good candidate for the analysis of change in contact. The complete paradigm of subject pronouns and clitics in Friulian is outlined in Table 1.  Note furthermore that Friulian varieties display considerable microvariation in the paradigm of SCL s. In the case of 1st person singular and plural, 2nd person plural and 3rd person singular and plural, allomorphy exists in different dialects of Friulian, which is observable in Table 1.3 As far as pro-drop is concerned, Friulian is a fully-fledged null subject language: full subjects can be dropped in the same structural and pragmatic conditions that determine subject drop in Italian or Spanish. Recall that Friulian SCL s are inflectional elements, and therefore their occurrence in a clause does not amount to saying that this is a non-null subject clause.
Previous research (Frasson, in press) on HS s of Venetan, a northern Italo-Romance variety, has shown that SCL s tend to be reanalyzed as weak pronouns, in the terms of Cardinaletti and Starke (1999). In this system, pronouns are divided into three classes: a strong one, a weak one and a clitic one. The trigger of the asymmetries that emerge among classes is structural deficiency: the more deficient a pronoun, the fewer phonological, morphological, syntactic and semantic features it realizes. While we cannot provide conclusive evidence that the same reanalysis from clitics to weak pronouns is also taking place in heritage Friulian, in this article we show that Friulian SCL s do display pronominal behavior, since they are not licensed in the context of doubling of a lexical or pronominal subject except when these are clearly topicalized (Section 6.1, Table 4); besides, SCL s can be dropped in the second conjunct in coordinated structures (Section 6,1, Table 6) and the restriction on their appearance in clitic clusters is violable4 (Section 6.1, Table 5). This syntactic behavior, in Rizzi's (1986) terms, is typical of pronouns, not of clitics. We therefore take these findings to indicate that they are in a process of reanalysis, from inflectional to pronominal elements. In the next section we present some relevant data on null subjects in Argentinian Spanish and Brazilian Portuguese, the two contact varieties, that may affect the system of subject pronouns of heritage Friulian.

Null Subjects in Argentinian Spanish and Brazilian Portuguese
The varieties of Friulian under analysis are spoken in Argentina and Brazil, where they underwent extensive contact with Spanish and Portuguese, respectively. The local varieties of Portuguese and Spanish are relevant for our study, in that they represent the dominant contact languages for most of the speakers. Therefore, their status with respect to null subjecthood, the nature of pro and the conditions on its realization presented in section 2.1, are crucial to define the role of contact in shaping the change in heritage Friulian. Brazilian Portuguese has been analyzed as a partial-pro-drop language.5 Based on the analysis of a corpus of written texts, Duarte (2000) shows that 2nd person singular subject pronouns are much more frequent in Brazilian Portuguese (90% of the occurrences) than in European Portuguese (24%). Similarly, Barbosa et al. (2005) show that spoken Brazilian Portuguese displays a higher number of overt pronominal subjects than European Portuguese and argue that might be related to the fact that verbal morphology in Brazilian Portuguese is reduced. For this reason, an overtly realized subject must appear in Spec-TP in Brazilian Portuguese. In spoken varieties, such a pronoun is generally a weak, phonologically reduced counterpart of the strong pronoun, as illustrated in (11), the structure of which is exemplified in (12). As noted by Kato (1999), when the strong pronoun is found in a left dislocated position inside the complementizer field, a weak pronoun can double it. The doubling of DP subjects through (weak) pronouns seems to be particularly common in Riograndense Brazilian Portuguese, the colloquial variety spoken in Rio Grande do Sul, the area where the data for this study were collected. To be completely sure of the status of the contact language with respect to null subjects, we interviewed one monolingual Brazilian Portuguese speaker in Caxias do Sul and one in Bento Gonçalves. A preliminary analysis of spontaneous speech shows that lexical DP s subjects realized by these speakers are generally doubled with an agreeing pronominal element, as shown in (13): (13) Riograndense Brazilian Portuguese O gaucho ele fala "buenas". the gaucho he speaks buenas 'A gaucho would say "buenas." ' The doubling seems to be restricted to DP subjects: there are no cases of phonologically reduced pronouns doubling tonic pronouns.
As for Argentinian Spanish, we rely on the research that has been conducted on the expression of the subject in the Rioplatense variety. In the literature on Spanish dialects (see in particular Silva-Corvalán, 2001), the Rioplatense variety is described as a pro-drop language, in which null subjects are generally favored. The only notable difference between Rioplatense and European Spanish regards the realization of the 2nd person singular subject pronoun with a non-specific referent or in its impersonal use; in this case Silva-Corvalán (2001) shows that the overt realization of the pronoun is much more frequent in Rioplatense (55% of the total occurrences of 2nd person singular) than in European Spanish (19%). However, when the referent is specific, Rioplatense Spanish does not exhibit a preference for the overt pronoun.
Some of the points we highlighted here will become relevant in the discussion of the heritage Friulian data: the two contact languages have a different distribution of pro. As a matter of fact, the conditions under which a null subject is accepted in the contact language seems to play a role in determining the realization of a pronominal subject as an overt or a null element in heritage Friulian. Before moving to the presentation of our results, we introduce some methodological notes.

Methodology
The data used in this study were collected in March-April 2019 in Argentina (Buenos Aires and Colonia Caroya) and Brazil (Bento Gonçalves and Ivorà).
The interviews were carried out by two interviewers in Argentina and Brazil. The tasks, as well as the instructions, were pre-recorded in Friulian by a native speaker of the language in order to avoid prompting the speaker for the contact language. These stimuli in Friulian were accepted as perfectly comprehensible, and accepted by the speakers without any objection. If the speaker needed extra information, it was provided in Spanish or Portuguese, i.e., in the contact language. All interviews were recorded and subsequently transcribed.

5.1
Participants Our study involves 9 Friulian speakers in Argentina and 7 in Brazil, for a total of 16 speakers (4 females, 12 males); the speakers' ages range from 57 to 93 years. While we are aware that this number of speakers is low for an experiment, we still believe this is a good representative sample. First, the number of speakers who are alive at the moment and who still speak the language is extremely low: Friulian in Southern America is an endangered language. It is not possible to provide a total number of Friulian speakers in Argentina and Brazil at the moment, as there may be recent emigrants as well as many heritage Friulians who do not speak the language.
As stated in the introduction, while this situation is far from optimal for drawing generalizations, this study is mainly qualitative in nature. We believe that these varieties, though underrepresented, need to be documented before their total extinction, and that syntactic phenomena are worth investigating whether they are present in a population of 2 million or 20 speakers.
All the speakers of Friulian we interviewed are native speakers but are dominant in the contact language (Argentinian Spanish or Brazilian Portuguese). All Brazilian informants are third generation HS s born in remote areas of the state of Rio Grande do Sul; five Argentinian speakers are second generation HS s born and raised in Buenos Aires, while the remaining four speakers were born in Italy and moved to different cities in Argentina when they were respectively 18, 14, 14 and 10 years old. The general tendency in both countries is to use the contact language in the vast majority of cases; speakers use Friulian only at home or with other Friulian speakers. This is particularly evident in the Friulian communities of Brazil, which are still quite isolated from the bigger Portuguese-majority speaking cities. In Argentina, Friulian always alternates with Spanish, even among members of the community. Other potentially relevant differences between the Friulian communities in Argentina and Brazil regard their level of education and knowledge of other languages. All the participants in Argentina also speak Italian (G1 immigrants are bilingual Friulian-Italian speakers and have learned Spanish when they moved to Argentina; HS s are bilingual Friulian-Spanish speakers and have learned Italian as a second language in Argentina); besides, four participants have a university degree, four participants have a high school degree, and only one participant only completed primary school. On the other hand, all participants in Brazil are Friulian-Portuguese bilingual speakers and were never exposed to Italian; five participants finished primary school, one participant attended the first two years of primary school and only one participant has a high school degree.6 The participants' details are summarized in Table 2. While we are well aware of the speaker pool's non-homogeneity, we remind the reader that finding HS s of a non-standard variety in Brazil and Argentina is a difficult enterprise, as stated above. Furthermore, we take the convergence of data that were collected in different areas and from speakers with a variety of profiles as a sign that the change at issue is not exceptional, but rather the result of contact. Moreover, as we will show, the contact language seems to play a role in determining the output of change.

Tasks and Materials
At the beginning, a sociolinguistic questionnaire was proposed to all informants to obtain information on their background, including the degree and quality of their exposure to Friulian and to the contact languages, their language dominance, their education and family situation. Two tasks followed the sociolinguistic questionnaire: a forced-choice task and a spontaneous production task. Not all participants carried out both tasks: it was not possible to use audio stimuli with some of our elderly informants. Ten participants carried out both tasks; for 2 participants in Brazil only spontaneous production data were collected, while for 3 participants in Argentina only data for the forcedchoice task were collected. That not all participants carried out both tasks, as well as the differences between informants' background as we described in Section 5.1, do not provide ideal conditions for a study and we are aware of its limits, as already pointed out. However, the data we collected still allows us to make some remarks on the nature and the distribution of SCL s in heritage Friulian.

5.2.1
The Forced-Choice Task The forced-choice task consisted of 24 items. Eight items targeted SCL s, but only four will be reported here, because the others are not relevant to the questions asked in this study.7 Sixteen items targeted other constructions such as differential object marking. Participants had to choose between two proposed sentences, one with and one without a SCL. The order of the two sentences was counterbalanced across the task. The aim of the task was to test whether clitics are still agreement markers in heritage Friulian or whether they are analyzed as pronouns. The tested contexts are: Participants were presented with auditory stimuli recorded by a Friulian native speaker. Each recorded item contained a pair of sentences testing one of the phenomena in i-iv; after listening to the sentences, participants were asked to choose the one they preferred. It is well known that a forced-choice task does not bring about conclusive evidence regarding the speakers' grammar, all the more for HL speakers who generally express less straightforward acceptability judgments (see Polinsky, 2018Polinsky, , 2020. For this reason, we cross-checked these data with spontaneous production.

5.2.2
Spontaneous Production In the production task, informants were asked to tell a short story about their past. In the case of first-generation immigrants, the interviewer asked the informants to say something about the experience of arriving at the destination country; in the case of HS s, the interviewer asked the informants to say something about their childhood. In this way, we managed to obtain a uniform corpus of spontaneous speech containing several sentences both in the present and the past tense.
Spontaneous production was used to support elicited data in the questionnaire and to test whether the use of clitics depended also on other linguistic or non-linguistic features.

6.1
Forced-Choice Task The forced-choice task confirmed our expectation that clitics are undergoing a process of reanalysis as pronouns8 as shown in Tables 3-6: SCL s in items marked with an asterisk (*) display a pronominal behavior and are not grammatical in Friulian as described in previous studies on this variety. This is particularly evident in the context of doubling, which is lost in most cases. Table 3 shows that both G1 and HS s disfavor the doubling of a non-topicalized subject (through the use of the clitic e in sentence 18). As stated above, Friulian as described in grammars would show clitic doubling in these contexts. The only grammatical way to utter these sentences in Friulian should be the one in Example (18). Table 3 shows that most of our informants chose the version without a subject clitic, which is not grammatical in Friulian.9 8 A statistical analysis of these data is not possible due to the low number of sentences. 9 A reviewer points out that the possibility of dropping the SCL in doubling contexts is expected for pronouns, as we showed in Section 4 for Brazilian Portuguese. Recall, however, that Friulian SCL s should not be dropped in this context: as agreement markers, they are always obligatory and do not distinguish between topicalized and non-topicalized DP subjects.

Frasson, D ' Alessandro and van Osch
Heritage Language Journal 18 (2021) 1-36 In topicalization contexts, such as in Table 4, the picture is different: topicalized subjects are doubled in most cases, as expected in Friulian. Most participants accepted the doubling of topicalized subjects (20), with only one HS in Argentina preferring the sentence without the SCL (21). This suggests that, in sentences with a lexical subject, topicalization is the context in which SCL s are most likely to appear. This is reminiscent of the situation in Standard French and colloquial Brazilian Portuguese, in which a subject pronoun can double a topicalized subject.
The restriction imposed on the realization of SCL s in clusters, which has been described for Friulian as spoken in Italy, is generally lost in HS s in Argentina and Brazil (22), while it seems stable in G1 speakers in Argentina (23).
In the context of coordinated structures, we see that first-generation speakers normally accept the repetition of the clitic in both conjuncts (24), as shown in previous studies on Friulian. HS s, on the other hand, seem to allow for an optional realization of the clitic in the second conjunct (24).  Table 6 shows that most informants chose the expected structure with a subject clitic in each coordinated conjunct. However, two HS s in Argentina and one in Brazil chose the structure with only one subject clitic. Recall that this should not be possible in Friulian: subject clitics are obligatorily realized every time a finite verb appears. The data presented in Tables 3-6 confirm that Friulian SCL s in Argentina and Brazil are changing from inflectional to pronominal elements.

6.2
Spontaneous Production Data For the spontaneous production data, all sentences containing finite verbs were coded for the following linguistic variables: person, subject type, cluster, verb class, and clause type. For the variable 'person,' 2nd person was excluded from the analysis. This was done for two reasons. First, in Friulian, 2nd person singular SCL s behave differently from 1st and 3rd person in that they are unaffected by the constraints on clitic clusters described in Section 2: they are realized in table 5 Realization of SCls in clitic clusters  all contexts and in the literature are generally treated differently from other SCL s.10 Second, there were too few instances of sentences with 2nd person singular and plural (only 6% of the data) to be able to include them in a statistical model. First person singular and plural, as well as 3rd person singular and plural, were merged into one category, as we did not observe significant differences within the same person and between singular and plural. For 'subject type,' a distinction was made between DP s (which included both demonstrative and full DP s), null subjects, pronouns and quantified pronouns,11 but the latter category was excluded due to the low number of cases (only nine). The variable 'cluster' refers to whether there was some other intervening clitic (such as a negator, a direct or indirect object, a reflexive or an impersonal clitic), in which case an SCL is not expected according to previous studies on Friulian. The variable 'verb class' consisted in three categories, namely transitive/unergative, unaccusative/passive and copular constructions, and the variable 'clause type' distinguished between matrix clauses and subordinate clauses.12

G1 (AR) HS (AR) HS (BR) Total
We also coded each sentence for whether there was a continuation or a shift in topic. If the referent of the SCL was the subject of the previous sentence, the sentence was coded as Topic Continuation; otherwise, if the referent was not the subject of the previous sentence (including the cases of reintroduction of an old referent after some time and introduction of a totally new referent), the sentence was coded as Topic Shift (see also De Prada Pérez, 2009). Topic continuation is an important factor that determines pronoun use in canonical null subject languages (Frascarelli, 2006;Frascarelli & Hinterhölzl, 2007;Jiménez-Fernández, 2016;). If SCL s in Argentinian and Brazilian 10 According to Renzi and Vanelli (1983) and Vanelli (1998), second person singular needs a different treatment from other subject clitics. As a matter of fact, it is the only person that is realized in all cases and in all varieties. In Poletto's (2000) analysis of various subject clitics, second person singular occupies a special HearerP position, the lowest one in the clitic field that she proposes. We leave the investigation on 2nd person for future research. 11 Besides Friulian, our study targeted other heritage northern Italo-Romance varieties, such as Venetan, Trentino and Piedmontese, which were not included in this article. Some of these homeland varieties, in particular the Venetan ones, exhibit a different distribution of subject clitics with lexical, pronominal and quantified subjects (see Benincà, 1994). Friulian subject clitics behave more regularly in this respect, as they double all types of subjects. All the varieties were uniformly tested for the same varieties, even though we would not expect a difference in the doubling of various subjects in Friulian. 12 As in the case of subject type, the literature on Friulian subject clitics does not show evidence of a different distribution with different verb or clause types. These variables were nevertheless included in the analysis, as they have been proved relevant in the realization of subject clitics in other varieties in our study. See in particular Benincà (1994), Manzini and Savoia (2005) and Poletto (2000) for a discussion on the realization of subject clitics in relative clauses and unaccusative constructions with postverbal subjects.
Friulian indeed behave like pronouns, we may expect them to obey a similar discourse constraint. In total, 580 sentences were coded, of which 375 contained an SCL and 205 did not. Two generalized mixed effects models were run using the lme4 package (Bates et al., 2012) from statistical tool R (R Development Core Team, 2017). The first model included all the data and the second model included only sentences with null subjects, as this context is crucial to determine whether the SCL is analyzed as a pronoun or an agreement marker.
In both models, all variables mentioned above (person, subject type, cluster, verb class, clause type and topic) were included, as well as the variable 'group' , for which orthogonal sum-to-zero contrasts were set; one contrast compared the G1 speakers to all HS s, and the other compared the two HS s groups in the different countries with each other. The two-way interactions between group and all other factors were also entered into the models.
The first model, which included all subject types, rendered a significant effect of group, which was modulated by interaction effects with some of the other variables. For instance, a significant interaction between group and cluster (β = 1.91, SE = 0.68, t = 2.80, p = .005) and subsequent Tukey posthoc tests showed that only in sentences with no other intervening clitics, G1 speakers were significantly more likely to produce SCL s (26) than the Argentinian HS s (27), (p < .001) as well as more than the Brazilian HS s (p = .01). In sentences containing clitic clusters there was no difference between the three groups: all participants tended to drop the SCL. This is illustrated in Figure 1. Similar effects were found in the model which included only null subjects. A significant interaction between group and cluster (β = −2.84, SE = 0.90, t = −3.15, p = .002), followed up by Tukey post hoc tests, indicated that in null subject sentences without clusters, the G1 immigrants in Argentina were most likely to use an SCL, followed (with a significant difference; p = .016) by the Brazilian HS, and the Argentinian HS s, who produced least SCL s of all, and significantly Frasson, D ' Alessandro and van Osch Heritage Language Journal 18 (2021) 1-36 less than the G1 (p < .001), and marginally significantly less than the Brazilians HS s (p = .07). Post hoc tests moreover indicated that only the immigrants and the Brazilian HS s were more likely to produce more SCL s in sentences without clusters (G1: p < .001; HS_Bra: p < .001), but the Argentinian HS s did not differ significantly between sentences with and without clusters (p = .19) These effects are shown in Figure 2.
The full model revealed another interaction effect, namely between group and person (β = −1.42, SE = 0.69, t = 02.06, p = .039), indicating that both HS groups were significantly more likely to produce clitics in the 3rd person (28) than in the 1st person (29) (p = 0.001 for the Argentinian HS s; 0.026 for the Brazilian HS s), while the first-generation did not make this distinction. This effect can be seen in Figure 3. As for the model on null subjects only, a significant main effect of person (β = 1.50, SE = 0.37, t = 4.04, p < .001) was found there as well. While the interaction with group was not significant in this model, post hoc comparisons were carried out to check whether similar patterns could be found as in the model on the entire dataset. These showed that, in sentences with null subjects only, Brazilian HS s did not differentiate significantly between the 1st and 3rd person clitics (p = .12). Neither did the G1 (p = .80), similar to the first model. The Argentinian HS s on the other hand preferred SCL s with 3rd person in null subject sentences, just as they did in the model on all sentences. These effects are shown in Figure 4.
In the model on the entire dataset, subject type also had a significant effect on clitic use (β = 1.21, SE = 0.41, t = 2.90, p = .004). Pairwise comparisons indicated that all speakers were significantly more likely to drop the SCL when it doubled a pronoun (30) than when it doubled a null subject (31) (β = 3.35, SE = 1.40, t = 2.90, p = .01), as shown in Figure 5. The rate of dropped SCL s with lexical subject was between that of pronouns and null subjects, but the difference with each of the two extremes was not significant. This seems to indicate that the most appropriate context for these speakers to produce a SCL is when Finally, the full model rendered a significant effect of topicality (β = 1.05, SE = 0.29, t = 3.64, p < .001), indicating a preference to produce a clitic when there was a shift in topic with respect to the previous sentence. The interaction with group was marginally significant as well (β = −1.06, SE = 0.59, t = −1.80, p = .07), which led us to explore the comparisons through Tukey post hoc tests. These showed that, in fact, only the Argentinian HS s were significantly more likely to produce clitics in topic shift contexts than in topic continuation contexts (example 32) (p = .02). For the Brazilian HS s, the same tendency was present, but the difference was not significant, (p = .10), and neither was the effect in the first-generation group (p = 0.93). These effects are illustrated in Figure 6. The above-described effect of topicality on SCL use was found in all sentences. The second model confirmed the significant effect of topicality for sentences with null subjects only (β = 1.17, SE = 0.34, t = 3.48, p < .001), though it was not modulated by group. This means that in sentences with null subjects, all speakers, regardless of group, were more likely to produce a SCL in topic shift contexts than in topic continuation contexts with null subjects. The effect is illustrated in Figure 7.

Discussion
The results of the forced-choice task presented in Section 6.1 indicate that SCL s in the varieties of Friulian spoken in Argentina and Brazil show a tendency towards reanalysis as pronouns. The tests (in particular the realization of the SCL s with topicalized and non-topicalized subjects and the repetition of the SCL in both conjuncts in coordinated structures) showed that SCL s allow also for a pronominal interpretation, which is not expected in Friulian as spoken in Italy. This situation resembles the results shown in Frasson (in press) for the heritage varieties of Venetan spoken in Brazil. However, SCL s are not dropped at random, as revealed by the spontaneous production data. One important syntactic factor that constrains the use of SCL s is person: 1st person SCL s are more likely to be dropped by HS s than 3rd person subjects. Our data suggest that the peculiar contact situation in which heritage Friulian is spoken may have favored an acceleration of the process and the use of 1st person SCL s has therefore drastically declined in the span of one generation.13 The type of subject also influenced the likelihood of a subject to be omitted: most clitics were produced in sentences with null subjects, and the fewest in sentences with pronominal subjects. These findings further supports the notion that SCL s are indeed interpreted as pronominal elements: if the SCL is analyzed as a pronoun, it makes sense that it would be less felicitous in combination with another pronoun in the same sentence.
The spontaneous production data furthermore revealed an effect of the presence or absence of another clitic. In Friulian spoken in Italy, the presence of another clitic is the only context in which clitics can, and in fact should, be dropped. The data presented here show that both G1 immigrants and HS s seem to adhere to this constraint, perhaps unsurprisingly, given their general tendency to drop clitics in other contexts. However, the judgment data show a different picture when it comes to clitic clusters: while G1 immigrants correctly chose the sentence without the SCL, most HS s chose the sentence containing the SCL. It should be noted that the judgment task contained only one sentence with a clitic cluster, and therefore it would be appropriate to question whether this is truly representative of all sentences of this type. The critical sentence in the judgment task contained a null subject, and as we know from the production data, the type of subject influences the likelihood of a SCL to be produced. In fact, sentences with null subjects invoke the most frequent use of SCL s among HS s, compared to sentences with pronominal lexical subjects. However, a separate model on the spontaneous production data, which only included sentences with null subjects, showed a similar result: all three groups preferred to drop the SCL in the presence of another clitic. Discrepancies between judgment and production data are not uncommon in linguistic research, and especially in HS s (Bowles, 2011;Montrul, Foote, & Perpiñán, 2008). The reason for this is probably that HS s do not have much metalinguistic / explicit knowledge about their heritage language, given that 13 An anonymous reviewer pointed out that 1st and 2nd persons are 'stage topics' or 'permanently available topics,' i.e., pronouns whose referents are always available in the discourse (Erteschik-Shir, 2007). Therefore, if SCL s are turning into pronouns, they are expected to be subject to discourse properties, hence be dropped more for the 1st and 2nd persons than for 3rd. We agree with the reviewer that it may be easier to retrieve 1st and 2nd persons in the discourse and thus the corresponding pronouns might be dropped more easily. they were never exposed to formal instruction in the language. Metalinguistic knowledge is particularly useful in tasks that target this type of explicit knowledge such as forced choice judgment tasks. Finally, an important result that arose from the spontaneous speech data concerns the effect of topicality on SCL use. This finding regards the distribution of SCL s in the context of topic shift or continuity: SCL s are most likely to be dropped when the topic is the same as in the previous sentence. This means that SCL s, when used as pronouns, conform to conditions for the licensing of pro in line with what was shown by Frascarelli (2007), Miyagawa (2017) and Jimenez-Fernández (2016): they are most likely to be used in sentences in which there is a shift in the topic.
This finding is important in many respects. Firstly, it strengthens the idea that SCL s in heritage Friulian can be reanalyzed as pronominal elements: recall that Friulian SCL s have been described as obligatory agreement markers and topic distinctions should not be relevant to their realization. Secondly, it contradicts previous research showing that the syntax-discourse constraints are harder to master in bilinguals, as shown by the Interface Hypothesis. The second-generation HS s in our study introduced a new discourse-related condition on the realization of the reanalyzed SCL s, demonstrating a particular sensitivity to discourse factors, rather than a weakening or loss thereof. We would like to argue that this departure from other studies may be because most of them have considered bilinguals who speak one pro-drop language and one non-pro-drop language, the latter being English in most cases. In the present study, though, we are dealing with two pro-drop languages, which are typologically similar (see Rothman, 2010 ff. on the relevance of perceived typological similarity for transfer between contact languages). This distinction would explain why other studies on bilinguals speaking two Romance languages have also failed to identify problems with null subjects (Carvalho & Child, 2011;De Souza, Chaves, & Simioni, 2018;Rinke & Flores, 2018): null subjects remain unaffected when two null-subject, typologically similar languages are in contact.
Further studies are necessary to establish whether the difference is due to the setting of the pro-drop parameter alone, or also the typological proximity between the two languages involved. The present study makes clear that that the predictions made by the IH do not work in the case of the language contact we analyzed.
Evidence that the contact languages play an important role is further provided by the differences observed between the HS s in Argentina and Brazil. One difference regarded the use of SCL s in clusters in sentences with null subjects. Although HS s in Brazil and first-generation immigrants produced more SCL s in sentences without clitic clusters, in line with previous descriptions of Friulian, the Argentinian HS s did not make this distinction. Moreover, the HS s in the two countries differ with respect to the effect of topicality. HS s of Friulian in Argentina clearly used SCL s in a way that is compatible with the conditions on the interpretation of pro we presented in section 1.2: SCL s are most likely to be dropped in sentences with no shift in the topic with respect to the previous topic. The same tendency was present in HS s of Friulian in Brazil but it did not reach statistical significance. Both these differences seem to suggest that HS s of Friulian in Argentina are in a further stage of a reanalysis process.
The most obvious explanation for this difference is the contact language, in particular the specific configuration of the contact language with respect to pro-drop. As briefly described in Section 2.2., Argentinian Spanish is considered a full-fledged pro-drop language, in which overt pronouns are used only in specific discourse contexts, such as when there is a shift in topic. Brazilian Portuguese, on the other hand, is typically described as a partial pro-drop language, which exhibits more overt subjects and in which the topicality effect is not as strong as in a pro-drop language like Argentinian Spanish.
However, other differences between the groups should be taken into account. Another possible factor behind the difference between HS s in Argentina and Brazil is the different generations to which the speakers belong. Our informants in Brazil are all third-generation HS s (they are the grandchildren of the original immigrants), as the migration wave to Brazil happened earlier than the one to Argentina, where all our informants are second generation HS s (the children of the original immigrants). This difference may account for the increased evidence of the topic continuity effects in second generation speakers in Argentina than in third generation speakers in Brazil. Second-generation speakers may introduce new interface conditions on the realization of the (reanalyzed) SCL s, but such conditions may be gradually lost in third generation speakers: the integration of information from syntax and discourse in the realization of SCL s requires a high processing load and is therefore recessive.
The two HS groups also differed from each other in the amount of contact they had with the majority language. HS s in Brazil live in isolated communities, farther away from big cities, which are still mainly composed of descendants of Friulian immigrants; as a result, they use their two languages in a diglossic fashion: Friulian is used with other members of the heritage community, while Brazilian Portuguese is normally used in all other contexts. HS s in Argentina (at least the ones in Buenos Aires), on the other hand, mix their two languages much more, using Spanish even with other Friulian speakers. This may result in a higher degree of cross-linguistic influence as a result of incomplete acquisition and/or attrition of the HL (e.g., Montrul, 2008).
Finally, the heritage Friulian-Argentinian speakers' knowledge of Italian as a second language may have influenced the results. These speakers are trilingual, and their Friulian could have been influenced by two pro-drop languages, Spanish and Italian. This group's trilingualism may have enhanced the reanalysis of clitics as pronouns and their discourse-related behavior.
While the present study makes it impossible to differentiate conclusively between the possible effects of contact language, generation or other external factors (amount of contact with the majority language, knowledge of other languages), it contributes to the discussion on the importance of the specific language combination in bilingualism research.

Conclusion
The study described in this paper aimed to explore the use of SCL s in G1 immigrants and second and third generation HS s of Friulian in Argentina and Brazil. The data from a forced-choice task and a spontaneous production task revealed that SCL s in heritage Friulian are in the process of being reanalyzed as pronominal subjects, as they are no longer obligatory. Their dropping, however, is not random: it is constrained by various linguistic and extra-linguistic factors. For instance, 1st person clitics are dropped more often than 3rd person clitics. Moreover, clitics are dropped more often when the topic of the sentence is the same as in the previous sentence, which is another reason to assume that these clitics are analyzed as pronominal elements. Discourse-constraints are not only not lost, but rather added to a domain that does not feature the same constraints in previously described varieties of Friulian, which indicates that the syntax-discourse interface is not always vulnerable in bilingual populations, as predicted by the Interface Hypothesis and as attested in previous work. We argue that a possible explanation for this departure from other studies might be that, unlike the majority of previous research, we have focused on closely related languages. This finding underlines the importance of expanding our research to include more language pairings, particularly typologically related languages.