The production of preverbal and postverbal subjects by Italian heritage children: Timing of acquisition matters

Heritage language (HL) speakers seem to diverge from monolingual speakers in the acquisition of syntax–discourse interface phenomena. However, most of the studies reporting this finding do not make any distinction between different types of syntax–discourse interface structures. Therefore, it is an open question whether these structures are difficult for HL speakers across the board or whether different types of structures are associated with different acquisition outcomes. In this study, we investigate whether the timing of the acquisition of syntax–discourse interface structures among monolingual children affects their acquisition by HL children. We consider the acquisition of the alternation between preverbal and postverbal subjects with unaccusative, unergative and transitive verbs in Italian. This alternation involves the syntax–discourse interface across all verb types. However, mastery of this alternation has a different timing in monolingual acquisition depending on the verb type: it emerges earlier in association with unaccusative verbs than unergative and transitive ones. We elicit narratives in Italian from 42 Italian heritage children living in Germany and ranging in age from 7 to 14 years. We focus on the analysis of preverbal and postverbal subjects, considering the effect of verb type and discourse status of the subject. The children show a good mastery of the alternation between preverbal and postverbal subjects with unaccusative verbs, whereas they exhibit difficulties with the same alternation in association with transitive and unergative verbs. Furthermore, postverbal subjects with transitive verbs are particularly vulnerable to cross-linguistic effects from German. The results of the study suggest a gradient interpretation of heritage speakers’ difficulty with syntax–discourse interface structures, with timing in monolingual acquisition being a relevant factor that modulates the degree of this difficulty. The results are not affected by children’s dominance of exposure to Italian or age.


Introduction
Heritage speakers (HSs henceforth) acquire the heritage language (HL) under particular conditions of language exposure. They are exposed to the HL at home from birth, but they usually experience a shift in language exposure from the HL to the societal language (SL) when entering kindergarten. They usually receive formal instruction in the SL. As a result, it is common for them to become more proficient or dominant in the SL than the HL in late childhood and adolescence (Benmamoun et al., 2013;Rothman, 2007;Torregrossa et al., 2023). Due to these conditions of language exposure, HSs may exhibit different acquisition processes and outcomes than monolingual speakers.
Across the literature, several studies investigate which linguistic domains are particularly vulnerable in HL acquisition (e.g. Benmamoun et al., 2013;Montrul, 2018;Montrul & Polinsky, 2011). Most of these studies reveal that HSs diverge from monolinguals in linguistic phenomena at the interface between syntax and discourse (see, among others, Kaltsa et al., 2015;Montrul, 2004;Polinsky, 1997;Silva-Corvalán, 1994;Tsimpli et al., 2004). By contrast, HSs and monolinguals are shown to have similar patterns with respect to the mastery of 'core' syntactic phenomena (see Kupisch et al., 2014, for adjective placement in heritage French with German as SL; Montrul et al., 2008 for subject wh-questions in HL Spanish/SL English; but see De Prada Pérez & Pascual y Cabo, 2012, for the observation of a different pattern, with HSs being more vulnerable to core syntactic phenomena than syntax-discourse interface ones). Studies show different ways in which HSs differ from monolinguals. For example, HSs' language production and comprehension seem to be vulnerable to cross-linguistic effects from the SL. Kaltsa et al. (2015) observe that Greek HSs (dominant in Swedish, their SL) produce overt pronouns referring to subject antecedents in Greek to a greater extent than monolinguals (see Montrul, 2004, for similar results in HL Spanish). In pro-drop languages such as Greek, the alternation between null and overt subject pronouns is regulated by syntax-discourse interface properties, because null pronouns appear in topic-continuity contexts (referring, for example, to a subject antecedent), whereas overt subject pronouns tend to mark topic-shift (referring, for example, to an object antecedent, see Torregrossa et al., 2021). The production of overt pronouns in topic-continuity contexts can be analysed in terms of cross-linguistic influence from Swedish, a non-pro-drop language requiring the use of overt subject pronouns in both topic-continuity and topic-shift contexts. However, the same result can be interpreted as a simplification of the complex form-function mapping associated with the use of overt pronouns in Greek.
Similar simplifications are observed in the production of other linguistic structures involving the syntax-discourse interface, such as the omission of differential object marking in Spanish as a HL (Montrul, 2004;Montrul & Bowles, 2009). Other studies even argue in favour of a restructuring of linguistic phenomena in the HL, which may lead to the emergence of new grammars. For example, Polinsky (2008) shows that Russian HSs express aspectual distinctions using analytical forms (e.g. light verbs like 'be', 'become' or 'do') instead of the synthetic morphological forms found in monolingual Russian.
To account for the divergent behaviour of HSs, several accounts have been proposed, ranging from representational to processing-related ones. For example, the observation of cross-linguistic effects or the emergence of new grammars supports a representational account (see, however, Serratrice, 2016, andSharwood Smith &Truscott, 2014, for a processing account of cross-linguistic effects). By contrast, in her formulation of the Interface Hypothesis (IH, henceforth), Sorace (2005Sorace ( , 2011 proposes that bilingual speakers -including HSs -have difficulties with syntax-discourse interface phenomena because the processing efforts of coordinating linguistic (e.g. the use of certain morphosyntactic forms) and non-linguistic information (e.g. discourse) add up to the processing efforts related to the inhibition of one language in favour of the other (see Montrul & Polinsky, 2011, for the extension of the IH to studies on HL). This may lead to the abovementioned simplification of the grammatical patterns of monolingual grammars.
It should be noted that the studies mentioned so far do not make any distinction between different types of syntax-discourse interface phenomena. Therefore, it is still unclear whether HSs differ from monolingual speakers in the mastery of all interface phenomena, or whether phenomena are associated with different acquisitional paths and outcomes. In the present study, we differentiate syntax-discourse interface phenomena on the basis of their timing in monolingual acquisition. In particular, we test to what extent the degree of lateness of acquisition of different interface phenomena among monolingual speakers modulates their acquisition in bilingual contexts, with phenomena acquired earlier being easier to master than phenomena acquired later (see Schulz & Grimm, 2019;Tsimpli, 2014).
The acquisition of postverbal subjects in Italian offers us a privileged viewpoint for investigating the extent to which the timing of monolingual language acquisition of different linguistic structures affects the success of their acquisition among HSs. As we show in section 'The distribution of preverbal and postverbal subjects in Italian', the distribution of subject-verb (SV, henceforth) and verb-subject (VS, henceforth) word orders in Italian is regulated by syntax-discourse interface principles that differ depending on the type of verb class (unaccusatives vs unergatives/transitives). Previous studies show that Italian monolingual children master the syntax-discourse principles regulating the alternation between SV and VS with unaccusative verbs earlier than unergative or transitive verbs (see section 'Acquisition of the alternation between SV and VS in monolingual speakers'). Our study investigates whether this difference is reflected in a different acquisition path and outcome of these structures by HSs. By doing so, we intend to verify if different types of interface phenomena pose different acquisition problems for HSs. Montrul (2018) underlines the importance of investigating HSs' language development during the school-age period because it represents the missing link between the early stages and the ultimate attainment of HL acquisition. As mentioned above, HSs usually receive formal instruction in the SL and not in the HL. This may affect the stabilization of complex structures, such as the ones involving the syntax-discourse interface, which tends to be related to literacy exposure Bowles & Torres, 2021;Caloi & Torregrossa, 2021;Rinke et al., 2019;Van Osch et al., 2019). However, only a few studies consider school-aged HSs. The present study intends to contribute to this line of research, by testing school-aged HL children (between 7 and 14 years) living in Germany and attending German-Italian bilingual schools.
Before introducing our study, we show which factors account for the distribution of SV and VS in Italian (section 'The distribution of preverbal and postverbal subjects in Italian'). Then, we report the results of previous studies that have dealt with the acquisition of these structures among Italian monolinguals (section 'Acquisition of the alternation between SV and VS in monolingual speakers') and bilinguals (section 'Bilingual acquisition of preverbal and postverbal subjects').

The distribution of preverbal and postverbal subjects in Italian
Italian is classified as a Subject-Verb-Object (SVO) language but allows for postverbal subjects (VS) under certain conditions. Rizzi (1982) argues that the availability of VS is related to the pro-drop nature of Italian: postverbal subjects are licensed by a phonetically null element in the specifier of the inflectional phrase. The distribution of VS depends on the interaction between lexical factors -as related to the class of the verb -and the information structure of the subject (Adger, 1995;Belletti, 1998Belletti, , 2001Belletti, , 2004. With unaccusative verbs, VS represents the unmarked word order, since it is used in broad-focus contexts, that is, when the sentence is 'all-new'. In these contexts, the postverbal subject is base-generated as an internal argument of the verb and exhibits object-like behaviour (Belletti, 1988;Burzio, 1986). However, VS can also be used in narrow-focus contexts, for example, when the subject is marked as new information and the verb is given in previous discourse (Belletti, 1988;Zubizarreta, 1998). 1 For example, the sentence in (1) is appropriate both as an answer to a question like 'What happens?' -introducing a broad-focus, all-new context -and a question like 'What comes?', which serves to elicit an answer with the subject constituent corresponding to the wh-operator in the preceding question and expressing new information (i.e. narrow focus on the subject).
The use of the SV-order with unaccusative verbs serves to mark the topicality of the subject of unaccusative verbs by means of the movement of the subject constituent from the internal argument position to the sentence-initial position. In (2), the subject il treno ('the train') appears in the preverbal position because it has been already mentioned in the previous question: it functions as a given topic. The narrow focus of the answer corresponds to the constituent al secondo binario ('at platform 2') in this case.
( With unergative and transitive verbs, the mapping between the position of the subject and its information structure shows an opposite pattern as compared with unaccusatives: SV represents the unmarked option, which is used in broad-focus contexts, as shown in (3), featuring the unergative verb fischiare 'to whistle'.
By contrast, VS is used to mark the subject as narrow focus (Belletti, 2001(Belletti, , 2004. The sentence in (4) is a felicitous answer to a wh-question like 'What has whistled?': the subject il treno ('the train') corresponds to the wh-operator of the question and appears in the postverbal position.
The same holds for transitive verbs. In (5), the constituent il treno ('the train') carries a narrow-focus interpretation since it answers the preceding subject wh-question. The direct object la macchina ('the car') -which has been already mentioned in the question -can be referred to by using a clitic pronoun or a clitic left dislocation in the preverbal position: 2 According to Belletti (2004), postverbal subjects of transitive or unergative verbs are derived via the movement of the subject constituent to the specifier of a low-focus projection, which dominates the verb phrase. This movement of the subject constituent is followed by the remnant movement of the other constituents (i.e. the verb and the object of transitive verbs) to the left periphery of the sentence (see Belletti, 2004, for further details).
To summarize, both SV and VS can be considered as marked word orders in Italian, depending on verb class (SV with unaccusatives and VS with unergatives or transitives). Therefore, the use of SV or VS in Italian requires mastery of the syntax-discourse interface conditions which affect their alternation. In the next section, we will show that the alternation between VS and SV is acquired earlier with unaccusatives than with unergatives (or transitives) by Italian monolingual children. This will lead us to investigate whether this difference in acquisition timing is reflected in the extent to which HSs master this alternation with the two types of verbs. Tsimpli (2014) refers to the timing of the acquisition of linguistic structures (with reference to monolingual language acquisition) as a relevant factor to consider when accounting for bilingual children's language acquisition process and outcome. Distinct focus should be put on the interaction of the timing of acquisition with the age of onset of bilingualism and the quantity of input a bilingual receives in the two languages. In particular, she proposes a three-way distinction of linguistic structures based on their timing of acquisition as early, late and very late (see also Schulz & Grimm, 2019, for empirical evidence in favour of this proposal). The different timing depends on where the structures are situated in the grammar. 'Core' syntactic structures (i.e. structures that involve only narrow syntax, see section 'The distribution of preverbal and postverbal subjects in Italian') tend to be acquired early, that is, children master them by the age of 3 years (or even earlier). 'Peripheral' syntactic structures (i.e. structures that are situated at the interface between syntax and discourse or cognition) tend to be acquired late, from 4 years onwards and may differ from each other in their degree of lateness. For example, some structures may emerge when the child is between 4 and 5 years (the late ones), whereas others may emerge after 5 years (the very late ones). This may be because, for instance, the acquisition of very late structures requires more cognitive resources than the acquisition of late ones (see Tsimpli, 2014, for discussion).

Acquisition of the alternation between SV and VS in monolingual speakers
In the present article, we examine two different types of interface structures (see section 'The distribution of preverbal and postverbal subjects in Italian'). We show that these structures differ from each other in their timing of monolingual acquisition. This conclusion is not based on the assumption that one structure involves more cognitive resources than the other, but rather on previous studies on the acquisition of these structures by monolingual children. Therefore, our classification of these structures as late or very late is based on solid empirical evidence as reported in previous studies.
First, it should be noted that Italian monolingual children acquire the distinction between verb classes from an early age. Lorusso et al. (2005) find that Italian monolingual children between 18 and 36 months produce a greater amount of VS in association with unaccusative verbs as compared with unergative and transitive verbs. As shown in section 'The distribution of preverbal and postverbal subjects in Italian', VS is the unmarked word order with unaccusatives. Therefore, its use is not affected by discourse factors. The observation that postverbal subjects with unaccusatives are acquired by the age of 3 years is consistent with Tsimpli's idea that 'core' syntactic structures are acquired early. By contrast, Italian children seem to master the syntax-discourse interface conditions involved in the alternation between the use of SV and VS at later stages of acquisition. Vernice and Guasti (2015) conduct two experiments based on sentence repetition tasks featuring appropriate and inappropriate SV and VS structures. In the first experiment, they manipulate the position of the subject (preverbal vs postverbal) and the type of verb (unaccusative vs unergative), always using indefinite subjects. They show that Italian monolingual children ranging in age between 4 and 5 years are more likely to repeat a SV after they hear a VS with an unergative verb rather than with an unaccusative verb. The second experiment is identical to the first one except for the fact that the subject is always definite. In this case, the children are more likely to produce SV rather than VS when they hear a VS. This holds for both unaccusatives and unergatives, suggesting that children prefer to move definite subjects to the preverbal position independently of verb class. Crucially, the comparison between the two experiments shows that the children are more prone to produce a SV after a VS with an unaccusative verb and a definite subject than a VS with an unaccusative verb and an indefinite subject. Overall, the study suggests that by the age of 5 years, children have knowledge of the distribution of SV and VS as motivated by both type of verb (i.e. unaccusative) and the discourse status of the subject (its definiteness in authors' terms; see also Cairncross & Dal Pozzo, 2022).
As for the use of VS with transitive verbs, Abbot-Smith and Serratrice (2015) use a picture-matching task to analyse the comprehension of different word orders by children of three age groups (2;6, 3;6 and 4;6). The children hear transitive sentences (with novel verbs) featuring a volitional agent and an affected patient and point at the corresponding picture. Target sentences vary in their word order: SVO, SO clit V and O clit VS. All groups perform at chance with the O clit VS order, that is, with a discourse-given object in the preverbal position and a narrow-focus subject. Even the eldest children fail to comprehend this kind of sentence (e.g. Lo baffa il gatto 'It clit "baffes" the cat'), even if they are accurate with the other two word orders. The authors argue that children have difficulties with the comprehension of O clit VS because it is very rare in child-directed speech and, as a result, they interpret the postverbal constituent as a patient and not as an agent (see Belletti & Contemori, 2012, for similar evidence on the acquisition of postverbal subjects in object relative clauses). This study suggests that 5-year-old children have not acquired the use of VS with transitive verbs yet.
Therefore, the abovementioned studies suggest that in Tsimpli's (2014) terms, the alternation between SV and VS is a late-acquired phenomenon with unaccusatives, but a very late-acquired phenomenon with transitives. The timing of acquisition of these structures matches the complexity of their derivation, with SV with unaccusatives involving movement of the subject in preverbal position and (O)VS with transitives involving movement of the subject to a low-focus projection as well as the remnant movement of the other constituents to the left periphery of the sentence (see section 'The distribution of preverbal and postverbal subjects in Italian'). However, if one does not want to assume the syntactic account presented in section 'The distribution of preverbal and postverbal subjects in Italian', the greater syntactic complexity of VSs featuring transitive verbs compared with VSs featuring unaccusative verbs may also be related to the observation that VSs with unaccusatives involve either no movement in broad-focus contexts or movement of the subject to a low focus position in narrow focus contexts. In contrast, VSs with transitives involve movement of two constituents (i.e. the object in the left periphery of the sentence and the subject to a low-focus projection). Furthermore, it is not excluded that the difficulty associated with the production of VS with transitive verbs is related to the occurrence of clitics in these structures, given that clitic production may be a vulnerable domain in HL acquisition (see Torregrossa & Bongartz, 2018 on Italian heritage children living in Germany).
Finally, it should be noted that based on previous studies, it is impossible to establish when Italian monolingual children acquire the use of VS with unergative verbs to mark the subject as new or contrastive focus. However, this is not problematic for the present study, given the limited number of unergative verbs occurring in the corpus of collected narratives (see section 'Methodology' for further details).

Bilingual acquisition of preverbal and postverbal subjects
The acquisition of VS structures by bilingual speakers has been investigated across several studies within the framework of the IH (Sorace, 2005(Sorace, , 2011. In this section, we use the term 'bilingualism' broadly, including L2 speakers who have learnt the target language in adulthood as well as speakers who have been exposed to it from birth. The IH capitalizes on the distinction between 'internal' and 'external' interfaces (Sorace & Filiaci, 2006). Internal interfaces involve the integration of information across different linguistic domains (e.g. lexicon-syntax and syntax-semantics), while external interfaces concern the interaction between linguistic domains and language-external resources (e.g. discourse constraints and cognitive abilities). In section 'The distribution of preverbal and postverbal subjects in Italian', we show that VS is the unmarked word order with unaccusatives. In other words, the use of VS with this class of verbs does not depend on discourse factors, but instead on the mapping between a lexical item (i.e. an unaccusative verb) and the associated syntactic structure. By contrast, VS with unergative and transitive verbs corresponds to a marked word order, which is motivated by discourse factors (i.e. focus marking on the subject constituent). According to the IH, bilinguals and near-native L2ers should exhibit difficulties only with the latter type of structures (VS with unergatives and transitives). This is confirmed by the study by Belletti et al. (2007) on near-native Italian L2ers speaking English as L1. Based on an elicited production and a story-telling task, the authors show that L2ers and monolingual controls pattern alike only in the use of VS with unaccusative verbs, while L2ers use a lower amount of VS with unergative and transitive verbs (see Lozano, 2006, for similar results in L2 Spanish). Similarly, Listanti and Torregrossa (submitted) investigate the acquisition of VS in L2 Italian across different proficiency levels. They find that the use of VS with unaccusatives emerges at the intermediate level and remains stable along the developmental path. By contrast, the use of VS with unergative and transitive verbs emerges only at advanced levels of proficiency.
Similar results have been found in Italian as a HL. Using an elicited production task, Caloi et al. (2018) investigate the use of VS by adult HSs whose SL is German, as compared with Italian monolingual controls. The participants of this study have to answer wh-questions eliciting an answer featuring a focus on the subject with different verb types. Contrary to the monolingual participants, the HSs do not use VS as a strategy to mark narrow focus on the subject with unergative and transitive verbs (see also Belletti & Leonini, 2004). Rather, they produce SV structures in which the subject constituent is marked as new information by prosodic means, which corresponds to the answering strategy typically employed in German (see also Note 1). By contrast, HSs converge with monolinguals in the production of VS with unaccusative verbs.
To our knowledge, no previous study has been conducted on HL Italian children in this domain. School-aged HSs speaking a language combination similar to the one considered in the present study have been tested by Van Osch et al. (2019). The authors analyse the production of VS in Spanish by three groups of HSs with Dutch as SL differing in age (9 years old, 13 years old and adult, respectively), as compared with Spanish monolingual controls. The results show that age plays a significant role in the acquisition of the conditions of use of VS: school-aged children produce VS only in association with unaccusative verbs, exhibiting sensitivity to verb type, but not to discourse factors. By contrast, adult HSs behave like monolingual controls. The results suggest that the discourse-related constraints related to the use of VS in Spanish are in principle ultimately acquirable by HSs, but with some delay compared with monolingual speakers.
It should be noted that all aforementioned studies analyse the use of VS rather than the discourse-driven alternation between SV and VS. Overall, the results are compatible with the IH, given that the use of VS involves different interfaces with unaccusatives, on one hand, and unergatives and transitives, on the other hand. However, the alternation between SV and VS is motivated by discourse factors with all verb types (unaccusative, unergative and transitives). The IH does not distinguish between different types of syntax-discourse interface structures. In this sense, Italian HSs should show difficulties not only in the use of VS with unergative and transitive verbs but also in the use of SV to mark the subject of unaccusative verbs as topics. However, as observed in section 'Acquisition of the alternation between SV and VS in monolingual speakers', the use of SV with unaccusative verbs seems to be acquired earlier than the use of VS with transitive verbs by Italian monolingual children. Therefore, if the timing in monolingual acquisition modulates the acquisition of interface phenomena, we should expect different acquisitional outcomes for the alternation between SV and VS with unaccusative and transitive verbs, respectively.
A study by Schulz and Grimm (2019) provides clear empirical evidence in favour of the concept that timing in monolingual acquisition affects the acquisition path and outcomes of bilingual children. The authors compare the acquisition of different phenomena by monolingual German children, simultaneous bilingual children (with German as one of their two languages) and early second-language learners of German. The children are tested in two rounds, at the ages of 4 years and 4 months and 5 years and 8 months, respectively. For the purposes of the present study, it is interesting that in the first testing round, the monolingual children and simultaneous bilingual ones behave alike in the mastery of early-acquired phenomena (SV agreement, telicity), but not phenomena that are acquired late or very late (complex sentences, wh-questions, negation and case marking). In the second testing round, the simultaneous bilinguals catch up with their monolingual peers in the acquisition of late phenomena (complex sentences and wh-questions) but not very late ones (negation and case marking). No effect of dominance of language exposure is found. The results suggest that the path of acquisition of structures with a different acquisition timing does not differ between monolinguals and simultaneous bilinguals, except for some delay exhibited by the latter group. In Schulz and Grimm's terms, simultaneous bilingual children just need some time to acquire the target structures successfully.
For the present study, we do not exclude that dominance of language exposure may be relevant, since we are more concerned with older HL children than the ones considered in Schulz and Grimm (2019). For example, Torregrossa et al. (2021) show that dominance of language exposure in the HL has a positive impact on the accuracy with which Greek HL children produce referring expressions in discourse, which is also a very late-acquired phenomenon, as the one considered in the present study (see also Paradis & Navarro, 2003, for similar evidence). Therefore, timing in monolingual acquisition and dominance of language exposure are both taken into account in this study. Furthermore, cross-linguistic influence from German (the SL) may affect the production of VS in Italian by Italian HSs. In the next section, we compare German and Italian in terms of word order, to identify possible sources of cross-linguistic influence from the SL to the HL.

Preverbal and postverbal subjects in German
German is a verb-second language, meaning that finite verbs always occupy the second position in main declarative clauses. The verb can be preceded by any constituent, such as the subject (as in (6)), an adverb (as in (7)) and the object (as in (8)). The three sentences are adapted from the Italian sentences (1), (2) and (5) above, respectively. In these three sentences, we use the past tense (Präteritum). If the verbs were in past perfect (Perfekt), the auxiliary would have occupied the second position in the clause and the associated past participle would have occupied the sentence's final position (Grewendorf, 2005).
The sentence in (6) overlaps with the corresponding SV(XP) Italian sentence at the superficial level. The sentence in (7)  'come') nor by information structure (i.e. narrow focus on the subject), but rather by the sentence-initial adverb and the verb-second constraint. Without the adverb, the sentence would be ungrammatical (i.e. *Kam ein Zug 'came the train') and SV would be used instead (i.e. Ein Zug kam). The same analysis applies to (8). The sentence overlaps superficially with the corresponding OVS Italian sentence, except for the absence of clitic resumption. Speakers may use this sentence to mark the subject constituent der Zug (the train) as the narrow focus (e.g. to answer a question like 'What hit the car?'). However, the OVS word order would be triggered anyway by the object in the sentence-initial position.
To summarize, German allows for SV and (XP)VS as in the case of Italian. However, the use of (XP)VS in German is unspecified with respect to the distinction between verb class (unaccusative vs unergative/transitive) and information structure categories. Here, we follow Müller and Hulk (2001) in claiming that cross-linguistic influence may occur in situations in which there is structural overlap between a bilingual's two languages: Language A only allows for one morphosyntactic option and Language B for two. In these cases, cross-linguistic influence from A to B is expected to occur, with the overlapping option being produced in Language B in situations in which the production of the non-overlapping option would be more appropriate. Based on the above considerations, we should distinguish two situations of overlap when considering the use of SV and VS in German and Italian, respectively. If a constituent in sentenceinitial position does not occur, German allows for only one option (i.e. SV), whereas Italian allows for two (SV and VS, not considering information structure). Müller and Hulk's (2001) hypothesis predicts the use of SV in Italian in contexts in which the use of VS would be more appropriate. For example, Italian HSs with German as their SL may produce SV with unaccusative verbs in broad-focus contexts or unergative verbs selecting a narrow-focus subject (as shown in the abovementioned study by Caloi et al., 2018).
However, if a constituent in sentence-initial position occurs, there is a full overlap between the two languages. Nevertheless, Italian postverbal subjects have a more complex feature specification than German ones, since the distribution of the former is motivated by information structure (e.g. [− topic] as in the case of unaccusatives and [+ focus] in the case of unergatives and transitives), whereas this is not necessarily the case of German postverbal subjects. Following Müller and Hulk (2001), we expect crosslinguistic influence from the 'simpler' system (German in this case) to the more complex one (Italian) -see Serratrice et al. (2004) for a similar revision of Müller and Hulk's (2001) hypothesis in terms of complexity of feature specification. This may result in the production of (XP)VS structures in Italian in which the subject is not specified for the appropriate information structure feature.

The study
In this article, we present the results of a story-retelling task performed by school-aged Italian HSs living in Germany and attending Italian/German bilingual schools. We conduct an analysis of SV and (O)VS sentences based on verb type and discourse status of the subject. We investigate to what extent HL children's acquisition of different types of VS structures is modulated by their timing in monolingual acquisition and children's dominance of language exposure in Italian or German.
We focus on the acquisition of the discourse conditions licensing the use of SV versus VS in Italian. These discourse conditions vary depending on verb type (unaccusatives vs unergatives/transitives), as has been shown in section 'The distribution of preverbal and postverbal subjects in Italian'. Furthermore, mastery of the alternation between SV and VS emerges earlier with unaccusative verbs than with transitive verbs in Italian monolingual acquisition. We expect to observe that HL children master the alternation between SV and VS better with unaccusatives than transitives based on the concept that timing in monolingual acquisition modulates HL children's acquisition of linguistic structures (in line with Tsimpli, 2014 andSchulz &Grimm, 2019). By contrast, the observation of no difference in the mastering of SV and VS between unaccusatives and transitives would suggest that HL children have difficulties with syntax-discourse interface structures across the board (in line with Sorace, 2011).
We also analyse whether the dominance of language exposure in one or the other language interacts with HL children's acquisition of the alternation between SV and VS. We expect Italian dominant children to perform better than German dominant ones, especially in the use of very late-acquired structures (i.e. VS with transitive verbs), which would be in line with the results presented in Torregrossa et al. (2021).
Finally, we consider how far the production of (appropriate) SV and VS structures with unaccusative and unergative/transitive verbs is affected by children's chronological age because of the large age range considered in our sample (see section 'Participants'). Previous literature reports inconsistent findings on the relationship between HL acquisition and age. On one hand, some studies show an increase of HL children's competence with increasing age (Rodina et al., 2020;Torregrossa et al., 2023). On the other hand, other studies report that older HL children perform similarly, or even worse, compared with younger ones, which would indicate a differential acquisition path of HSs compared with monolinguals (e.g. Armon-Lotem et al., 2021).

Participants
Forty-two (23 females) Italian HL children ranging in age between 7 years and 5 months and 14 years and 1 month (M = 10 years and 11 months, SD = 27 months) took part in the study. Participants were recruited in two Italian-German bilingual schools, a primary school in Hamburg (N = 24) and a secondary school in Cologne (N = 18), respectively. German was the main language of instruction in both schools, with the number of hours dedicated to the teaching of Italian ranging between 7 (in Cologne) and 12 (in Hamburg). Before the study, the parents provided written informed consent. Both the parents and the teachers reported that none of the participants had previously identified speech, hearing or visual impairments. All children had at least one Italian-speaking parent. Thirty-three out of 42 were born in Germany. The remaining nine had been exposed to German no later than the age of 4 years.

Research instruments
Background questionnaires. The parents of the participants were administered a background questionnaire with the aim to identify patterns of language exposure across different contexts (e.g. home, outside-school activities, etc.) over time (currently and in the past -from 0 to 6 years). This was done to assess children's dominance. In particular, we used a subset of the questions in the questionnaire used by Bongartz and Torregrossa (2020) and Torregrossa et al. (2022) -see also Mattheoudakis et al. (2016) for information related to the design of the questionnaire. The questions used in this study tapped into children's home language history and current language use. We refer to Appendix 1 in Supplementary Materials for a description of the structure and the analysis of the questionnaire.
Vocabulary test. We used two vocabulary tasks (i.e. one for each language) as a measure of participants' verbal abilities. For Italian, we used the Renfrew Expressive Vocabulary Test (Renfrew, 1995; see Andreou et al., 2021, for the use of this vocabulary measure in Italian). The participants were asked to name 50 pictures representing different commonplace objects. For German, we used a productive vocabulary task normed for German monolingual children (Petermann et al., 2010), which consisted of 40 pictures meant to elicit 30 nouns and 10 verbs. In both cases, if no correct answer was given, we provided the participants with a semantic cue, to disambiguate the picture or avoid its misperception. If the participants were still not able to give any answer, we provided them with a phonemic cue consisting of the first syllable of the target word.
Narrative elicitation. The story-retelling task which was administered to the children was based on the Edmonton Narrative Norms Instrument (ENNI; Schneider et al., 2005). This assessment tool includes two structurally equivalent stories. Each story consists of 13 pictures representing a series of events involving two major characters (either a dog girl and a rabbit boy or an elephant girl and a giraffe boy) and two minor ones (either an elephant lifeguard and an elephant lady or a rabbit balloon seller and a rabbit lady). We chose to rely on this instrument because the actions represented in the pictures are useful to elicit different verb types. For example, the characters perform actions over objects, which should correspond to the use of transitive verbs, or they appear, disappear or move in the scene, which should correspond to the use of unaccusative verbs. Furthermore, the story has been designed to elicit topic shift (when the attention focuses from one character to the other) and topic maintenance (when a character remains in the scene for an entire episode). This allowed us to observe the effects of information structure on the use of different word orders. Each child was asked to retell one or the other story (counterbalancing the order across participants). They had to look at the pictures of the story, as appearing as a sequence of PowerPoint slides on a computer screen, and listen to a prerecorded female voice telling the story on the headphones. Then, they had to retell the story to the experimenter. The experimenter did not have visual access to the pictures, so that the children were encouraged to provide as many details of the story as possible to the experimenter. During the retelling phase, children could look at the pictures (see Torregrossa & Bongartz, 2018, for methodology).
We decided to rely on the retelling mode since previous studies have shown that bilingual children tend to produce longer and more complex narratives compared with when they have to tell the story from scratch (Otwinowska et al., 2018). Furthermore, Otwinowska et al. (2018) show that the retelling mode has a positive effect on macrostructure measures (i.e. the complexity of story structure), but not on microstructure ones, which includes lexical richness or the use of specific linguistic structures (i.e. pronouns vs full nouns in Andreou et al., 2020). In other words, the retelling mode should not affect children's use of VS across verb types in the case at stake in the present study. However, in order not to bias the children in the use of VS, we included no instance of VS in the stories, except for one case, in which the use of SV in association with an unaccusative verb would have been inappropriate. We refer to Appendix 2 in Supplementary Materials, in which we report the model stories and analyse the clauses containing a finite verb based on the position of the subject (preverbal vs postverbal), the type of verb (copular, transitive, unaccusative and unergative) and the discourse status of the subject (given vs new) -see section 'Analysis of the narratives' for the relevance of these parameters.

Data analysis
Analysis of background questionnaires and vocabulary tests. The aim of the present analysis is to calculate a dominance score for each child that takes into account two measures of language exposure (home language history and current language use, as extracted from the background questionnaires) and one measure of language proficiency (vocabulary). All measures are expressed as differentials between the score obtained in Italian and the one obtained in German, whereby a positive score indicates dominance in Italian and a negative score dominance in German. Since we used two different vocabulary tests for German and Italian, respectively, and these tests are based on a different number of items, the differential score related to dominance in vocabulary is calculated after converting each language-specific score into a percentage. Appendix 3 in Supplementary Materials reports the descriptive statistics for each measure (see Table S3.1).
To calculate the dominance scores, we perform an exploratory factor analysis (EFA). The aim of this analysis is to investigate how far the three dominance-related scores (home language history, current language use and vocabulary score) load on one factor. This factor corresponds to a general construct of language dominance. After verifying the assumptions for conducting an EFA based on the three variables at issue (see Table  S3.2 in Appendix 3 in Supplementary Materials for the results of the Kaiser-Meyer-Olkin Measure of Sampling Adequacy and Bartlett's Test of Sphericity), we verify the number of factors onto which the three variables load, by visualizing a scree plot (see Figure S3.1 in Appendix 3 in Supplementary Materials). This analysis confirms that there is a single factor underlying the three variables at stake. We use the factanal function in R (R Core Team, 2021) to conduct the EFA and extract the loading of each variable, to identify the contribution of the variable to the factor (Hartmann et al., 2018). Based on these loadings, we calculate a dominance score for each child. This score is the weighted sum of the values corresponding to each of the three variables (home language history, current language use and vocabulary score), with the weights corresponding to its loading in the above EFA. The final score has excellent reliability compared with the other three variables, with an α of .87. It is used as the child's general dominance score in the next analysis.
Analysis of the narratives. Children's stories were audio-recorded, transcribed and divided into units based on the occurrence of at least one finite verb. Since our analysis focuses on the position of overt subjects, units lacking an overtly expressed subject are discarded. The final dataset contains 421 units. We annotate the units for the following features: • • position of the subject; • • verb class; • • discourse status of the subject.
In the following sections, we introduce the labels used in the annotation system. All the examples are taken from the transcriptions of the children's narratives.
Position of the subject. The subject constituent can either precede (SV) or follow (VS) the verb. 98 units out of 421 display the VS order. The position of the subject is the outcome variable in the statistical model described in section 'Data analysis and results'.
Verb class. We initially consider three verb classes: unaccusatives, unergatives and transitives (for details on each verb class, see Levin et al., 1995;Sorace, 2000). However, we decide to group unergatives with another verb class, since there are very few units containing unergative verbs (34 in total). We include them in the group of transitive verbs, based on the evidence reported in sections 'The distribution of preverbal and postverbal subjects in Italian' and 'Acquisition of the alternation between SV and VS in monolingual speakers' that unergatives and transitive verbs both involve an external argument position, contrary to unaccusatives. Verb class is used as a predictor in our model. It has two levels, that is, unaccusatives versus transitives/unergatives.
Subject discourse status. We code each subject constituent for its discourse status, differentiating between subjects marked as 'given' and as 'focus' (see Riester & Baumann, 2013). We code the subject as 'given' whenever the corresponding referent has been mentioned in discourse. Given subjects tend to be marked for definiteness, as shown in (14) and (15) We code both new and contrastive constituents as 'focus' since both properties have been shown to trigger VS in Italian (Belletti, 2001; see section 'The distribution of preverbal and postverbal subjects in Italian'). We classify the subject as 'new' if the corresponding referent is introduced in discourse for the first time. Discourse-new subjects tend to be expressed by indefinite noun phrases, as shown in (16) and (17): Subjects are coded as contrastive based on two criteria, taken from Riester and Baumann (2013): i) if there is a parallelism with another referent in discourse; ii) if they occur in the scope of a focal operator, such as anche ('too') or solo ('only'). In (18), the subject constituent lei ('she') in the last sentence evokes the alternative la giraffa ('the giraffe') in the preceding one. Therefore, it is coded as contrastive based on the first criterion: The subject constituent occurring in (19) is classified as contrastive based on the second criterion:

Statistical model
Data are analysed using a generalized linear mixed-effects model in R version 4.1.2 (R Core Team, 2021). We run two models, both including word order (SV vs VS) as the outcome variable. In the first model, the predictor variable is the interaction between verb class (unaccusative vs unergative or transitive), the discourse status of the subject (given vs focus) and children's dominance score. In the second model, the predictor variable is the interaction between verb class, discourse status of the subject and children's age. We sum-coded the two-level predictors of verb class and subject discourse status as −0.5 and 0.5 and centred the continuous variables related to children's dominance and age. We include random intercepts for participants, with random slopes initially maximally specified.
Both models are fitted using the buildmer function in the 'buildmer' package (Voeten, 2021). This function uses lmer from the 'lme4' package (Bates, Maechler, et al., 2015). It starts by attempting to fit the most maximal model possible. If the model fails to converge, the function then simplifies the random effects structure via backwards stepwise elimination. Once the maximally converging model has been fitted, the function calculates p-values for all fixed effects using the 'lmerTest' package (Kuznetsova et al., 2017 The resulting models are the maximally converging models supported by the data (Bates, Kliegl, et al., 2015).

Results
Dominance scores. Figure 1 shows the distribution of the dominance scores across participants. We refer to Appendix 3 in Supplementary Materials for a description of how these dominance scores are calculated. The majority of the children is dominant in German (negative values in Figure 1). Furthermore, the Italian-dominant children are not as dominant as the German-dominant children (MIN =−2.23; MAX: 1.37). Overall, the group was German dominant (M = −0.52, SD = 1.00).
Interaction between verb class, subject discourse status and dominance. The maximally converging model includes the interactions between verb class, discourse status of the subject and children's dominance score as factors. Random intercepts of participants, with random slopes for the discourse status of the subject, are supported. The results of the analysis are related to the probability of observing a 'VS' response. 3 Table 1 reports the estimates, standard errors (SE), z-values and p-values for each predictor. We find a significant main effect of verb class, indicating that the probability of observing a VS significantly increases when the verb is unaccusative. This pattern is visualized in Figure 2 (left), which plots the predicted probabilities for the children to produce a VS according to the type of verb employed. The probability of observing a VS with V = unaccusative is much higher than the probability of observing a VS with V = unergative or transitive.   We also find a significant effect of the discourse status of the subject, indicating that the probability of observing a VS significantly decreases when the subject is given. This pattern is shown in Figure 2 (right), which plots the predicted probabilities for children to produce a VS sentence according to the discourse status of the subject. In this case, the probability of observing a VS with S = Focus is higher than the probability of observing a VS with S = given.
The interaction between the verb class and the discourse status of the subject is also significant. As Figure 3 suggests, the children seem to be sensitive to the discourse status of the subject when producing SV or VS with unaccusative verbs. The former is used to mark the subject as given, whereas the latter is used to mark the subject as focus. By contrast, the discourse status of the subject does not seem to affect the use of SV or VS with unergative/transitive verbs, given that SV tends to be produced both when the subject is given and when it is new. Pairwise comparisons confirm these observations (as run by using the emmeans function in R; Lenth, 2021). With unaccusative verbs, the probability of observing a VS is significantly higher if the subject is focus than if the subject is given (β = 3.7, SE = 0.82, z = 4.53, p < .001). With unergatives and transitives, there is no significant difference between focus and given subjects (β = 0.63, SE = 0.88, z = 0.72, p = .47). Finally, we do not find any effect of dominance. The effect of the interaction between verb type and information structure does not seem to be affected by children's dominance score.
Interaction between verb class, subject discourse status and age. The maximally converging model includes the interactions between verb class, discourse status of the subject and children's age as factors. Random intercepts of participants are supported. As in the previous model, the results of the analysis are related to the probability of observing a 'VS' response. 4 Table 2 reports the estimates, SE, z-values and p-values for each predictor.
We find the same significant effects as in the model reported in Table 1. In particular, we find a significant main effect of verb class and discourse status of the subject and a significant interaction between these two factors. There is no significant effect of age nor any interaction between verb type, discourse status of the subject and age, indicating that the effect of the interaction between verb type and discourse status of the subject does not vary depending on the age of the children.

Discussion
The present study investigates Italian HL children's acquisition of the syntax-discourse interface conditions that regulate the alternation between preverbal and postverbal subject positions with different types of verbs in Italian. In particular, we analyse to what extent the acquisition of these conditions reflects their timing in monolingual acquisition and is affected by children's dominance and age. We rely on the existence of two structures in Italian, allowing for the alternation between SV and VS order, namely unaccusative verbs (for which VS is the unmarked word order) and transitive verbs (for which VS marks the subject constituent as focus). As shown in section 'The distribution of preverbal and postverbal subjects in Italian', the alternation between SV and VS involves the syntax-discourse interface in both cases, with preverbal subjects being associated with givenness with unaccusatives and postverbal subjects with newness or contrast (i.e. focus) with unergatives or transitives. Crucially, however, the acquisition of the alternation between SV and VS has a different timing in monolingual acquisition across the two verb classes: acquisition occurs earlier with unaccusative verbs than with transitive verbs (and most probably unergatives). The first result emerging from our study is that the Italian HL children who took part in the experiment tended to produce more VS sentences if the verb was unaccusative rather than unergative or transitive (Figure 2, left). We also observe the production of a greater number of VS in association with a focus subject (rather than a given one), as shown in Figure 2, right. Overall, these results reflect some general tendencies observed in the use of VS in Italian, as discussed in section 'The distribution of preverbal and postverbal subjects in Italian'.
However, we also observe a significant interaction between the verb class and the discourse status of the subject. With unaccusative verbs, the HL children are significantly more prone to produce a VS if the subject is a new or contrastive focus than if it is given, reflecting the distribution observed in monolingual Italian. With transitive (and unergative) verbs, they tend to produce VS with both given and new-information or contrastive subjects. This suggests that the alternation of SV and VS is particularly vulnerable with this class of verbs.
These results are compatible with the idea that the timing of acquisition of a certain structure among monolingual children affects its acquisition among HL children. Italian HL children master the alternation between preverbal and postverbal subjects with unaccusatives, which is a late-acquired phenomenon in monolingual Italian, but do not master the same alternation with transitive and unergative verbs, which is a very-late-acquired phenomenon. Crucially, both phenomena involve the syntax-discourse interface, as shown in section 'Acquisition of the alternation between SV and VS in monolingual speakers'. Therefore, a categorical interpretation of the IH would predict that HL children have difficulties with both types of structure (Sorace, 2011). Our study suggests a more gradient interpretation of this hypothesis, according to which timing in monolingual acquisition modulates the degree of difficulty of syntax-discourse interface structures among bilingual speakers, with structures acquired earlier being easier than structures acquired later. In other words, the acquisition of certain structures by HL children may follow a specific path which mirrors the acquisitional path of these structures by monolingual children. If this is the case, we would expect HL children to be able to ultimately acquire even syntax-discourse interface structures that are acquired very late. This finding is compatible with evidence reported in previous studies that adult HSs perform like monolingual speakers even with syntax-discourse interface phenomena (see Kupisch & Rothman, 2018, for a discussion of this idea, as well as the aforementioned study by Van Osch et al., 2019).
It should be noted that in our analysis, we do not find age to modulate the interaction between verb type and discourse status of the subject (Table 2). In other words, we do not find any evidence of increasing competence in the use of SV and VS with increasing age, contrary to the results shown by Rodina et al. (2020) and Torregrossa et al. (2023) in relation to other linguistic phenomena. At first sight, this observation does not support our previous consideration that very-late-acquired syntax-discourse interface structures can be fully mastered with increasing age. Rather, it suggests differential acquisition between HSs and monolinguals (see section 'The study'). However, it is not excluded that the acquisition process is still ongoing among the HL children analysed in this contribution. In this sense, the children might need more time for a complete acquisition of the alternation between SV and VS. Furthermore, the assumption that HL children are, in principle, able to acquire very-late-acquired syntax-discourse interface phenomena does not mean that they will do so. Differential acquisition in comparison to monolinguals may still occur under certain exposure conditions. In the previous sections, we show that the participants of this study are relatively homogeneous in terms of language dominance (see section 'Dominance scores'), but heterogeneous in their literacy exposure, with the children in Cologne being exposed to fewer hours in Italian than the children in Hamburg. Crucially, the children in Cologne are also older than the ones in Hamburg (section 'Participants'). Therefore, the difference in literacy exposure in the HL between the younger and the older children may account for the lack of age effects in the analysis. This proposal is compatible with several studies showing that literacy exposure in the HL is more relevant for the acquisition of late-acquired phenomena than language exposure to the HL outside school (e.g. Torregrossa et al., 2022). However, this finding should be taken with caution given that the amount of hours dedicated to the HL at school is a rough indicator of literacy exposure to the HL, which should be integrated with the consideration of how the HL is taught and which literacy-related activities the children conduct outside of school.
Another interesting result emerging from our study is that dominance in language exposure (which has been shown to be crucial for the acquisition of marked word orders, see section 'Bilingual acquisition of preverbal and postverbal subjects') does not seem to modulate the interaction between verb type and discourse status of the subject in the production of SV versus VS. This means that children tend to master the alternation between SV and VS with unaccusatives but not with transitives independently of their dominance in German. This is unexpected based on other studies that have observed a crucial role of language exposure variables on the acquisition of interface phenomena (see section 'Introduction' and Torregrossa et al., 2021 on reference production). This finding might be related to our previous observation that the group of children considered in this study is more homogeneous concerning the dominance of language exposure than other groups of children considered in other studies. For example, the study by Torregrossa et al. (2021) also includes children who have a pronounced dominance in the HL, whereas most of the children in this study are dominant in the SL (see Figure 1).
As a concluding remark, we comment on the occurrence of transitive verbs featuring a discourse-given postverbal subject in the narratives produced by the HL children. In other words, HL children produce an unexpected marked structure which is infelicitous. A more careful analysis of these data reveals that in most of these sentences, the verb is preceded by another constituent -generally an adverb -as poi ('then') in (20) and (21): The structure of the sentences in (20) and (21) seems to reproduce the constituent order (AdvVSO) of the corresponding German declarative sentences, with the verb appearing in the second position (see section 'Preverbal and postverbal subjects in German' on the use of postverbal subjects in German). As argued in section 'Preverbal and postverbal subjects in German', the direction of cross-linguistic influence from German to Italian is expected in this case: Italian allows for AdvVS(O) under certain circumstances related to the interaction between verb type and discourse status of the subject (e.g. with unaccusative verbs selecting an unmarked subject or transitive verbs selecting a new-information one). By contrast, German exhibits AdvVS(O) independently of verb type or discourse status of the subject. Therefore, the occurrence of AdvVS(O) in the production of German-Italian bilingual children may be interpreted as the result of cross-linguistic influence from the system with a less complex featural specification (AdvVS(O) in German) to the system with a more complex one (AdvVS(O) in Italian).
This study is the first to interpret the acquisition of marked word orders in a HL under the lens of timing in monolingual language acquisition. However, we would like to mention two limitations of this study. First, the evidence related to the timing of the phenomena at issue in monolingual language acquisition is based on previous studies, since we did not include a monolingual group here. Our main aim is to understand how far HL acquisition is modulated by the 'complexity' of the phenomena (as reflected in their timing of acquisition), children's dominance and their age. To answer these research questions, no monolingual group is needed. Second, we should consider that the non-occurrence of a structure in semi-spontaneous speech does not necessarily reflect an 'incomplete' acquisition of this structure. In this sense, a full understanding of HSs' mastery of the use of SV and VS with different verb types can only be reached by triangulating evidence from several tasks, including, for instance, semi-spontaneous speech -which is usually 'freer' and more ecological -and elicited imitation -which tends to be more controlled.

Conclusion
Our study provided empirical evidence of a modulating role of timing of monolingual acquisition in the HL acquisition of syntax-discourse interface phenomena. We investigated the acquisition of the alternation between pre-and postverbal subject position in Italian by school-aged HSs whose dominant language was German. HSs' performance was shown to vary depending on the type of verb considered (unaccusative vs transitive). This difference reflected a different timing in the acquisition of these structures by monolingual (20)  Italian children, with subject position alternation acquired earlier with unaccusative verbs and later with transitive ones. The results of the study offered a nuanced view of the difficulty involved in the acquisition of syntax-discourse interface structures. The lack of significant effects of age and dominance did not necessarily contradict previous findings related to the acquisition of syntax-discourse interface structures. As for age, we did not exclude that the acquisitional process was still ongoing among the children considered in this study, whereby the most difficult structures would be eventually acquired with increasing age. As for dominance, we pointed to the importance of interpreting the results related to the acquisition of syntax-discourse interface structures among HL children in terms of their literacy exposure in and outside school, since literacy seems to play a crucial role for the acquisition of late or very late phenomena.

Aknowledgement
We would like to thank the children for their enthusiastic participation in our experiments and their parents and teachers for supporting the study. The paper benefitted from the constructive comments of our reviewers and our editor, to whom we are deeply grateful.
2. It should be mentioned that the questions in (4) and (5) could also be answered by using a preverbal subject (i.e. an SV order) which is marked by a rise-fall intonation. This is shown in (i) and (ii), where the block capitals indicate prosodic prominence on the narrow-focus constituent. The interface with prosody will not be taken into account in this contribution. We refer to Torregrossa (2012) for further details.