Introduction

Once young children succeed at producing two-word utterances, there is a steady growth in the array of syntactic constructions that they can understand and produce. Their grammatical competence, however, does not develop homogeneously as various morpho-syntactic structures are mastered at a different pace and a rich literature on language development has focused on the factors of linguistic complexity that affect the emergence of different grammatical structures. Fine-grained comparisons between structure-specific growth trajectories are, however, relatively scarce in language acquisition research for many reasons, such as the practical difficulties of performing longitudinal studies or utilizing samples from large cross-sectional databases of children at different ages. Structure-specific developmental trends are often inferred only indirectly, and even the acquisition trends of grammatical structures that have received the most attention are hard to compare; this is true also for passives and Romance clitics that are the two constructions under investigation in this paper. There is, however, a remarkable source of information that might be exploited: the databases collected for the standardization of language assessment tests. These databases, primarily created for diagnostic reasons, usually cover an array of grammatical structures across different age groups, which are often much larger than the sample of participants tested in conventional experimental settings. Despite the great potential of these databases, their use in studies focusing on grammatical development remains limited. This is mostly due to the fact that participants' performance is considered only in aggregated form and scores collected for the various grammatical constructions are usually collapsed into a single composite score, making a comparison between different structures hard or impossible. This information, however, is available within the databases, and can be extracted through fine-grained analyses that go beyond total scores.

In this study, we looked at a large database of responses from Italian children between 4 and 10 years of age, part of the norming sample used to standardize the Batteria di Valutazione del Linguaggio in bambini dai 4 ai 12 anni / Battery for Language Assessment in children aged 4 to 12 years old (BVL 4–12; Marini et al. Reference Marini, Marotta, Bulgheroni and Fabbro2015). We targeted specific structures, so as to compare the developmental trend of sentences that differ in the morpho-syntactic realization of the internal argument of transitive verbs.

In Romance languages, including Italian, the internal argument of the verb, the theme/patient, appears in post-verbal position in canonical SVO sentences, while it appears preverbally in sentences such as passives and certain clitic constructions. It has been suggested that this phenomenon relates to movement operations, that are in turn associated with different degrees of grammatical complexity in language development. For example, Jakubowicz (Reference Jakubowicz2005, Reference Jakubowicz2011) proposed a complexity metric precisely defined in terms of the movement steps involved during the derivation. Her account capitalized on the initial preference for in situ wh-questions in French, both in children with typical development (TD) and those with developmental language disorders (DLD). In the same vein, Moscati and Rizzi (Reference Moscati and Rizzi2014, 2021) proposed that the developmental sequence of agreement configurations conforms to a movement-based metric of grammatical complexity. Indeed, children between 3 and 5 years of age mastered local agreement configurations in Italian that do not involve movement (e.g., Determiner-Noun agreement) earlier than those characterized by movement (e.g., Clitic-Past Participle agreement).

Following these previous approaches, other grammatical constructions, such as passives or clitics, can be investigated and compared once movement is taken as a primitive. In fact, both structures feature the dislocation of the internal argument, as illustrated by the Italian examples in (1–3):

In the SVO sentence in (1), the internal argument le ragazze “the girls” occupies its canonical post-verbal position. In (2) and in (3), instead, it surfaces to a position to the left of the finite verb/auxiliary, either as a clitic pronoun or as the subject.

Different approaches exist in the literature to characterize the fine-grained properties of (2) and (3). However, a movement step ^{Footnote 1} has been generally assumed in both clitic constructions (see Kayne Reference Kayne1975; Sportiche Reference Sportiche, Rooryck and Zaring1996; Belletti Reference Belletti and van Riemsdijk1999) and passives (Jaeggli Reference Jaeggli1986, Collins Reference Collins2005). This is schematically illustrated in (4), where the theme/patient is raised from its post-verbal position to a higher one to the left:

Clitic constructions and passives also share other grammatical properties. For example, they both trigger past-participle agreement in Italian, a property that has also been considered in relation with movement. Although the past-participle does not agree with the object in canonical SVO sentences as in (1), movement triggers past-participle agreement in gender and number in (2) and (3). The uninflected form salut-ato in (1) is replaced by salut-ate in (2) and (3), a form that carries the same gender and number [−ate = +fem, + plur] of the fronted argument.

Besides considerations based on their grammatical properties in adult language, developmental studies have also unveiled some similarities between the two structures in (2) and (3). Both are produced later than SVO declaratives and are typically mastered only after three years of age. Many factors may be at play here, and they may not be the same for clitics and passives. However, there is at least one family of hypotheses that shares a similar assumption, namely that the observed delay in the acquisition of these structures may be traced back to the movement step in (4). Borer and Wexler (Reference Borer, Wexler, Roeper and Willliams1987) proposed that the comprehension advantage of short passives (e.g., the window is broken) over long ones (e.g., the window is broken by the boy ) stems from the fact that the full movement-based computation is obligatorily required only in the latter case, while short passives may receive a simpler adjectival reading in which no movement is involved. The asymmetry between long—arguably requiring movement—and short passives has been replicated many times across different languages (Greek: Terzi and Wexler, Reference Terzi, Wexler and Hirotani2002; Russian: Babyonyshev and Brun, Reference Babyonyshev and Brun2003; Portuguese: Rubin, Reference Rubin2009; see also Pierce Reference Pierce1992 for Spanish and the role of post-verbal subjects). In Italian, Volpato et al. (Reference Volpato, Verin and Cardinaletti2016) investigated children's comprehension of passives. They manipulated several factors, and, when the comprehension of actional long “be” passives was considered, their results indicated that children were able to select the correct interpretation shortly before turning four years old.

This time window roughly overlaps with the age at which children begin to master the syntax of clitics. Italian-speaking children can retrieve the correct antecedent of a clitic computing the local domain by the age of three (McKee, Reference McKee1992), and this is apparently generalizable across Romance languages (e.g., in French: Hamann et al., Reference Hamann, Kowalski, Philip, Hughes, Hughes and Greenhill1997; in Spanish, Bauuw et al., Reference Baauw, Escobar, Philip, Sorace, Heycock and Shillcock1997). However, other comprehension problems persist longer. For example, Dispaldro et al. (Reference Dispaldro, Ruggiero and Scali2014) analyzed the comprehension of 3rd person object clitics in Italian, testing children’s ability to use gender and number—two features that may be checked through movement—to disambiguate between two possible referents in a picture-matching task. Their results showed that by the age of four, children’s performance was still not adult-like.

More evidence, showing that it is at around the age of four that children's difficulties with clitics disappear, comes from studies focusing on production. In Italian (Hamann et al., 1997; Tedeschi, Reference Tedeschi2009), Catalan (Gavarrò et al., Reference Gavarró, Torrens and Wexler2010), and French (Hamann Rizzi and Frauenfelder Reference Hamann, Rizzi, Frauenfelder and Clahsen1996; Prévost 2009), the omission of 3^rd person object clitics disappears at around this age (see also the large comparative study reported in Varlokosta et al., Reference Varlokosta, Belletti, Costa, Friedmann, Gavarró, Grohmann, Guasti, Tuller, Lobo, Anđelković, Argemí, Avram, Berends, Brunetto, Delage, Ezeizabarrena, Fattal, Haman, van Hout, de López, Katsos, Kologranic, Krstić, Kraljevic, Mi&ecedil;kisz, Nerantzini, Queraltó, Radic, Ruiz, Sauerland, Sevcenco, Smoczyńska, Theodorou, van der Lely, Veenstra, Weston, Yachini and Yatsushiro2016). The acquisitional complexity of clitics cannot be reduced to pronominalization in general, since the production of object pronouns in languages lacking a clitic series—as for example in English—does not seem to pose the same difficulty (Perez-Leroux et al., Reference Pérez-Leroux, Pirvulescu and Roberge2008; Serratrice et al., Reference Serratrice, Sorace and Paoli2004). Again, as for passives, movement-based accounts have been proposed for the late emergence of cliticization. This hypothesis dates back at least to Wexler (Reference Wexler1998), who proposed that the complete sequence of movement steps required by cliticization might be subject to maturation ^{Footnote 2} .

More recently, new movement-based accounts have been proposed and object clitic omissions have been linked to a broader class of movement constraints. As shown in Figure 4, the first movement step raises the internal VP argument across the agent. This may give rise to classic intervention effects (Friedmann et al., Reference Friedmann, Belletti and Rizzi2009) that have been documented in both adults and children (for an overview, see Biondo et al., Reference Biondo, Pagliarini, Moscati, Rizzi and Belletti2022). This line of reasoning was initially proposed by Zesiger et al. (Reference Zesiger, Zesiger, Arabatzi, Baranzini, Cronel-Ohayon, Franck, Frauenfelder, Hamann and Rizzi2010) for French and more recently by Arosio and Giustolisi (Reference Arosio and Giustolisi2019) for Italian. According to such accounts, intervention effects are modulated by the degree of similarity, in terms of shared grammatical features, between the moved constituent and the one it crosses over.

This brief review highlights several similarities between the two constructions, both on the theoretical and on the empirical side. Although a unifying account seems beyond reach at the moment, many hypotheses already share the core idea that object movement might be a common source of complexity in language development. To the best of our knowledge, however, the only explicit comparison between clitics and passives is the production study reported in Manetti (Reference Manetti, Baiz, Goldman and Hawkes2013). She observed that Italian-speaking children between 3;6 and 4;6 years systematically avoided passives, preferring the use of clitics. Manetti argued that passives could be harder than clitics at this stage, hence the preference for using clitic constructions instead of passives. To date, however, Manetti's work is an isolated case and no comparative study exists in comprehension. Moreover, it is still impossible to have a comparative estimate of the development of the two structures in child language. Our first goal is therefore to enrich the literature in this direction and extend the observation to children between 4 and 10 years old. This is well beyond the time window considered by previous investigations regarding clitics and passives, which have usually focused on the time window between 4 and 6. In most cases, however, children's performance was still not at ceiling, suggesting that some development might still occur in the grammar school years. For this reason, we decided to also include older children.

A second major goal of this study concerns the relationship between working memory and sentence comprehension. As already stated, grammatical development is characterized by great interindividual variability. This may depend not only on sociolinguistic factors (e.g., environmental variables related to parental education and professional profile; Riva et al., Reference Riva, Cantiani, Dionne, Marini, Mascheretti, Molteni and Marino2017; Romeo et al., Reference Romeo, Leonard, Robinson, West, Mackey, Rowe and Gabrieli2018) but also on a child’s cognitive profile. Indeed, morpho-syntactic abilities hinge on the interplay between accumulating linguistic knowledge and growing cognitive skills (i.e., attention, executive functions, and memory; Miyake et al., Reference Miyake, Friedman, Emerson, Witzki, Howerter and Wager2000; Miyake and Friedman, Reference Miyake and Friedman2012; Marini et al., Reference Marini, Piccolo, Taverna, Berginc and Ozbič2020). Accumulating evidence from studies involving children with TD and DLD suggests that, among the cognitive skills involved in grammatical processing, working memory plays a major role (e.g., Baddeley et al., Reference Baddeley, Gathercole and Papagno1998; Gathercole et al., Reference Gathercole, Service, Hitch, Adams and Martin1999; Engel de Abreu et al., Reference Engel de Abreu, Gathercole and Martin2011; Marini et al., Reference Marini, Gentili, Molteni and Fabbro2014; Riva et al., Reference Riva, Cantiani, Dionne, Marini, Mascheretti, Molteni and Marino2017). According to an influential multicomponent model, working memory is a complex system made of different sub-components each dedicated to the storage and management of a specific type of representation (i.e., Visuo-Spatial Sketchpad, Episodic Buffer, Phonological Loop) under the supervision of a Central Executive (Baddeley, Reference Baddeley2003, Reference Baddeley2007). Together with the Central Executive, the Phonological Loop, made of a passive component (Phonological Short-Term memory, Pho-STM) and an active one (Verbal Working Memory, Verb-WM) which manipulates the information temporarily stored in the Pho-STM, is involved in lexical and grammatical processing. For example, vocabulary growth correlates with Pho-STM abilities as measured through serial span measures (e.g., Digit Forward recall task, Non-word repetition task) (e.g., Baddeley et al., Reference Baddeley, Gathercole and Papagno1998) in children with TD (Gathercole et al., Reference Gathercole, Service, Hitch, Adams and Martin1999; Leclerc and Majerus, Reference Leclercq and Majerus2010 ) and DLD (Archibald and Gathercole, Reference Archibald and Gathercole2006). Pho-STM memory and Verb-WM arguably operate synergically in sentence comprehension: Pho-STM holds linguistic material and makes it available for more complex or longer processing cycles; Verb-WM may instead provide the operational resources needed to perform grammatical computations such as those related to syntactic movement. Indeed, long-distance relations generated by the dislocation of a constituent from its original position likely involve working memory, at least until the dependency has been resolved (Gibson, Reference Gibson, Marantz, Miyashita and O’Neil2000; Lewis and Vasisht, Reference Lewis and Vasishth2005).

To obtain information about the different components involved, serial and composite span tasks can be used: Serial span tasks (e.g., the Digit Forward Recall and the Nonword repetition tasks) have been typically employed to assess the passive capacity of Pho-STM (storage only) whereas composite span tasks (e.g., the Digit Backward Recall and Last Word retrieval tasks) have been related to the functionality of the Central Executive to allocate Verb-WM resources (storage + processing). Results, however, still provide a blurred picture.

In a group of 6-year-old bilingual children, Engel de Abreu et al. (Reference Engel de Abreu, Gathercole and Martin2011) found that both Digit Forward and Backward Recall tasks correlated with accuracy in sentence comprehension. However, they measured accuracy considering the overall score obtained by administering the TROG-2 battery (Bishop, Reference Bishop2003). This choice does not allow us to determine whether sentences with an increasing degree of morpho-syntactic complexity also require an increasing amount of working memory resources. This issue was explicitly addressed by Montgomery, Magimairaj, and O'Malley (Reference Montgomery, Magimairaj and O’Malley2008). They looked at simple and complex sentences. Complex sentences were considered to include different types of non-local relations, such as anaphoras (pronouns, reflexives) and movement relations (passives). Comprehension accuracy was measured for simple and complex sentences separately. In children between 6 and 12 years old, the authors failed to find significant correlations between measures of Pho-STM and sentence comprehension. ^{Footnote 3}

The potential relation between Pho-STM, working memory, and syntactic complexity has been assessed also in studies that focused on clausal embedding. Arosio, Guasti, and Stucchi (Reference Arosio, Guasti and Stucchi2011) analyzed the comprehension of subject and object relative clauses in 9-year-old children. In object relative clauses, the object is moved across the subject, similarly to the movement step shown in (4). This factor has been thought to account for the higher processing load in both children (e.g., Adani, Reference Adani2011; Arosio et al., Reference Arosio, Adani, Guasti, Brucart, Gavarró and Solà2009) and adults (e.g., Gordon et al., Reference Gordon, Hendrick and Johnson2001; Grodner and Gibson, Reference Grodner and Gibson2005; King and Just, Reference King and Just1991; Staub, Reference Staub2010; Staub et al., Reference Staub, Dillon and Clifton2017; Traxler et al., Reference Traxler, Morris and Seely2002, Biondo et al. Reference Biondo, Pagliarini, Moscati, Rizzi and Belletti2022), also considering the involvement of memory resources in maintaining active long-distance dependencies (Gibson Reference Gibson, Marantz, Miyashita and O’Neil2000; Lewis and Vasisht Reference Lewis and Vasishth2005). The results presented in Arosio et al. (Reference Arosio, Guasti and Stucchi2011) showed a clear correlation between digit span and object relatives, i.e., the constructions with higher complexity.

The apparent contrast between the findings in Arosio et al. (Reference Arosio, Guasti and Stucchi2011) and the ones reported in Montgomery et al. (Reference Montgomery, Magimairaj and O’Malley2008) may depend on the different types of sentences that were used in the two studies, thus making any comparison based on grammatical complexity inconsistent. Indeed, in a systematic review of working memory and sentence processing in children, Boyle, Lindell, and Kidd (Reference Boyle, Lindell and Kidd2013) highlighted that “complexity” has usually been used to refer to either movement or embedding, two very different kinds of grammatical relations. Boyle, Lindell, and Kidd (Reference Boyle, Lindell and Kidd2013) treated movement or embedding as distinct factors and experimentally manipulated them in a sentence comprehension task administered to children aged 4 through 6 years old. They found a significant effect of movement, as sentences with object movement were harder in both moved (i.e., passives) and embedded clauses (i.e., object relatives) with a significant interaction between movement and embedding, meaning that object relatives were the hardest configuration to process. Turning to memory, however, Nonword repetition and Digit Backward Recall tasks, associated with Pho-STM and Verb-WM respectively, did not play any significant role ^{Footnote 4} . More recently, Delage and Frauenfelder (Reference Delage and Frauenfelder2020) assessed Pho-STM, Verb-WM, and sentence comprehension in children with TD and DLD aged 5 to 14 years old. Notably, the sentence comprehension task included complement clauses as well as subject and object relatives. In children with TD, measures of both Pho-STM and Verb-WM predicted better comprehension only when complex morpho-syntactic constructions were considered. However, again, movement and embedding were not analyzed separately. This limits the results in important ways, and no conclusions could be drawn about the impact of memory resources on object movement.

Taken together, previous studies on sentence comprehension have started unveiling a relationship between language processing and Verb-WM. However, the contribution of the different components of working memory and their involvement with respect to grammatical constructions with different degrees of derivational complexity is still not clear. The mixed pattern that emerges from the previous investigations is arguably related to the selection of the stimuli used in the experiments. Moreover, differences in the age range, in the size of the experimental groups, and in the measures used to assess working memory make it hard to come up with a fully coherent understanding of this complex relationship.

Overall, this study has two main goals: the first is to compare the developmental trend of sentences that do not involve movement (i.e., canonical SVO sentences) with alternative constructions (i.e., Clitic sentences and Passives) that are derived through a dislocation of the internal argument. This is built on a primitive notion of movement-based syntactic complexity, as proposed in Jakubowicz (Reference Jakubowicz2005, Reference Jakubowicz2011), Moscati and Rizzi (Reference Moscati and Rizzi2014), and Moscati et al. (Reference Moscati, Rizzi, Vottari, Chilosi, Salvadorini and Guasti2020); the second is to investigate the potential relation between Pho-STM and Verb-WM on constructions whose grammatical computation needs temporary storage of non-local dependencies created by a movement operation (i.e., Passives and Clitics).

Cross-sectional data were obtained by administering a grammatical comprehension task to a large cohort of 996 children aged 4 to 10. The participants’ accuracy in sentence comprehension was assessed with a picture-matching task, and separate values for each of the three grammatical constructions (i.e., canonical SVO sentences, passives, and clitic clauses) were collected. According to the existing literature, we hypothesized that movement would increase the processing load, making it harder for younger children to understand passives and clitics than SVO sentences. Here, we considered matrix clauses only, so as to exclude the potential confound introduced by clausal embedding. We also aimed to assess the relationship between sentence comprehension and working memory. We, therefore, collected two working memory measures: a serial span measure (i.e., Digit Span forward) that is usually associated with Pho-STM memory and a composite span measure (i.e., Digit Span backward) that is associated with Verb-WM.