1.1.1 SV word order with unaccusative verbs

Unaccusative verbs are verbs that take only one internal argument. The DP occurring with an unaccusative verb is the theme in the sentence, for example: in the sentence “the leaf fell”, “the leaf” is not the agent of the act, rather the theme of it. The basic word order of sentences with unaccusatives is verb-subject, VS (nafal ha-ale, “fell the leaf”), that is – the order in which the DP is a complement to the verb. The subject-verb, SV order (ha-ale nafal, “the leaf fell __”), is attained by movement of the DP from object position to subject position (Perlmutter 1978; Levin & Rappaport Hovav 1995). We refer to this movement, from object to subject position, as A-movement. In Hebrew, unlike in English, this movement is not obligatory with unaccusatives, so sentences can occur in both V_unaccS and SV_unacc orders. See Figure 1 for such an example for the basic order (Figure 1a), and the order after the A-movement of a DP from object to subject position (Figure 1b).² (Figure 1b also shows movement of the verb to I, in grey).

Figure 1

The syntactic trees of (a) the basic sentence in V_unaccS order and (b) the sentence after A-movement, in SV_unacc order for the sentence ha-ale nafal “the-leaf fell”.

One of the most influential hypotheses on language acquisition literature is the Maturation of Syntax hypothesis by Borer & Wexler (1987). Based on children’s difficulty with verbal passive, a structure that is also derived by A-movement, Borer and Wexler proposed that A-movement matures late in the acquisition process, hence children younger than five years cannot produce sentences with such movement (see also Hirsch & Wexler 2007, who proposed that A-movement matures only around age seven, based on a test of raising construction comprehension, administered to English-speaking children). Under this framework, Wexler and his collaborators argued that when young children produce sentences with an unaccusative verb in SV order, they are actually analysing these verbs as unergative verbs, that is, verbs in which the DP is base-generated in subject position and does not move there from object position (Babyonyshev et al. 2001 for Russian; Machida et al. 2004 for Japanese). Contrarily, many other findings from studies examining comprehension or production in a variety of languages (English, European Portuguese, French, Hebrew, Italian, and Japanese) demonstrate that young children do distinguish unaccusative verbs from other verbs, and analyse A-movement correctly (Pierce 1992a; Snyder et al. 1995; Sano et al. 2001; Adragão & Costa 2004; Lorusso et al. 2005; Friedmann 2007; Shimada & Sano 2007; Costa & Friedmann 2009; 2012; Friedmann & Costa 2011). Studies found that whereas it is difficult for young children to produce sentences in VS order with verbs other than unaccusative verbs (e.g., “jumped”, “laughed”) and they prefer producing these verbs in SV order (Pierce 1992a; Lorusso et al. 2005), they produce unaccusatives both in VS and in SV order before the age of three (Adragão & Costa 2004; Friedmann 2007; Costa & Friedmann 2009; 2012). Other studies found that children use correctly morpho-syntactic means that are unique to unaccusatives, like the possessive dative li (for-me) in Hebrew (Friedmann 2007), which is a marker of an internal argument (Borer & Grodzinsky 1986), and the auxiliary verb in Italian (different auxiliaries are used for unaccusative and non-unaccusative verbs, Burzio 1986), even before the age of three years (Guasti 2002).

1.1.2 Passives

An additional structure derived by A-movement is verbal passive. Verbal passives are derived by syntactic movement of the object to subject position (e.g., in the sentence “The teddy bear was hugged __ [by the boy]” the DP “the teddy bear” moved from its post-verbal position to the pre-verbal subject position). In some studies, some passive constructions have been reported to be difficult in the first stages of acquisition of English and German, and verbal passives (which include A-movement) are acquired later than adjectival passives, which do not include A-movement (Maratsos et al. 1985; Borer & Wexler 1987; Pierce 1992b; Israel et al. 2000; Abbot-Smith & Behrens 2006; Hirsch & Wexler 2007). Considering the evidence described above, according to which children do not have a general difficulty with A-movement, it seems that their difficulty with passive does not stem from late maturation of A-movement, and may be ascribed to other sources, unrelated to A-movement, such as a difficulty with by-phrases or with passive morphology. This is supported by studies that found that English- and Japanese-speaking children have no problem understanding and producing short passives without by-phrases or passive morphology (Fox & Grodzinsky 1998; Sano et al. 2001, see also Machida et al. 2004). A study of the acquisition of passive comprehension in 11 languages with different typologies (Armon-Lotem et al. 2016) found, in most languages examined, that at age 5, children understand passives without a by-phrase at 80% and above success rate, better than passives with a by-phrase. In most of the 11 languages, five-year-olds were also able to understand passives with a by-phrase but in other languages, including Hebrew, passives with a by-phrase were still difficult (leading to reversed roles errors).

In Hebrew, verbal passives are rarely used (Berman & Sagi 1981; Berman 1997; Ayal-Smikt 2001; Jisa et al. 2002), and Hebrew-speaking children more commonly use arbitrary pro constructions, which maintain the argument structure of the transitive verb (lakxu oti, pro_arb-took._pl me ‘(they) took me’), and unaccusative verbs which only include one argument (Berman 1997).

1.1.3 Raising

Raising constructions are also derived by A-movement, of the subject of the embedded clause moving to the subject position of the main clause (Landau 2011; Danon 2015; Melnik 2017). Raising constructions may or may not include an experiencer argument (e.g., ‘Silvia seems (to Anna) to be smart’). Not many studies examined the acquisition of raising constructions, and the ones that did (mainly testing English-speakers) portray an inconsistent picture. Hirsch & Wexler (2007) and Mateu (2020) report that young English-speaking children’s production of raising constructions is minimal. As for comprehension, whereas Hirsch & Wexler (2007), Orfitelli (2012), and Mateu (2020) report that children do not understand these structures before the age of six/seven, Becker (2006) showed that children as young as 3–4 years old already succeed in understanding raising seem constructions without an overt experiencer. Froud et al. (2010) showed that 3–6 year olds do not have a general problem with raising constructions, but rather only in cases in which the A-movement crosses an experiencer DP: they understood raising sentences without an experiencer, but had difficulty understanding raising sentences with an experiencer (e.g., “Pooh seems to Piglet to be holding an umbrella”). This finding suggests that it is not A-movement that is difficult for young children, but possibly the crossing dependency.

Thus, most studies of the acquisition of A-movement indicate that even in early ages there is no difficulty acquiring A-movement itself, as sentences with unaccusative verbs in SV order and verbal passives without a by-phrase are acquired early and without difficulties. Some other properties of certain sentences with A-movement may be acquired later, such as verbal passives with a by-phrase, and possibly also raising structures with an experiencer.

1.2.1 Topicalization

The object of topicalized sentences moves from the VP-internal complement position to Spec-CP (Haegeman 1994). For example, as shown in Figure 2, the Hebrew sentence “acc this girl, grandma drew” is derived by movement of the object, “this girl”, together with the accusative case marker, to the beginning of the sentence.

Figure 2

Topicalization: A-bar-movement of the object to the beginning of a sentence, to Spec-CP, for the sentence et ha-yalda ha-zo safta ciyra “acc the-girl the-this grandma drew”.

Only few studies examined the acquisition of topicalization. Studies of the production of topicalization structures in spontaneous speech reported age of emergence between 1;8 and 2;5 (European Portuguese: Soares 2003a; b; Adragão & Costa 2004; Japanese: Otsu 1994; German: Spinner & Grinstead 2006). Studies examining the comprehension of topicalized sentences found that European-Portuguese-speaking children aged 3;3–6;1 can understand OSV topicalization sentences well, but still struggle with OVS topicalization sentences (Adragão & Costa 2004). Sano (2004) tested 3–6 year old Japanese-speaking children, and found that even the youngest children understood topicalized sentences when the object was presented in a previous utterance as the topic of the discourse, but struggled greatly without this introduction. Biran & Ruigendijk (2015) found that Hebrew-speaking children aged 3–6 years performed at chance in the comprehension of OVS topicalized sentences.

1.2.2 Relative clauses

Sentences containing a relative clause are derived by A-bar-movement of a DP from within the embedded clause to Spec-CP position of the main clause³ (as demonstrated in Figure 3). The movement in subject relatives is from the subject position of the embedded clause (e.g., “The grandma that _ kisses the girl”). The movement in object relatives is from the object position of the embedded clause (“The grandma that the girl kisses _”).

Figure 3

A-bar-movement of an object to the relative head position in an object relative clause for the phrase ha-safta she-ha-yalda menašeket, “The-grandma that-the-girl kisses”.

Two timelines for the acquisition of relative clauses can be seen in the literature: some studies report early acquisition of relative clauses, e.g., Armon-Lotem (2005) and Berman (1997) report that first relative clauses appear in the spontaneous speech of children acquiring Hebrew around age 2–2;6, and mastered around age 4. Other studies refer to much later acquisition, around age 6 (e.g., McKee et al. 1998; Friedmann & Novogrodsky 2004; Günzberg-Kerbel et al. 2008; Friedmann et al. 2009). Additionally, whereas some studies refer to similarities in the acquisition of subject- and object relatives, others consistently report subject-object asymmetry in relative clauses and wh-questions (de Villiers et al. 1994; Correa 1995; Friedmann & Novogrodsky 2004; Arosio et al. 2009; Friedmann et al. 2009; Belletti & Contemori 2010; Biran & Ruigendijk 2015; Haddad-Hanna & Friedmann 2015).

These two differences, with respect to ages and with respect to subject-object asymmetry, we suggest, emerge from different studies exploring different types of relative clauses. As we will also show in our study below, whereas relative clauses are indeed acquired early, and with no subject-object asymmetry, a specific configuration is problematic for children until around age 6, which is a configuration derived by A-bar-movement of a lexically-restricted DP across an intervening lexically-restricted DP (Friedmann et al. 2009; Belletti et al. 2012; Rizzi 2018). These intervention constructions can be found in certain object relatives, topicalization structures, and wh-questions, and studies that test the comprehension and production of these structures with intervention configuration find late acquisition and subject-object asymmetry. Indeed, Friedmann et al. (2009), for example, showed that Hebrew-speaking children aged 3;6–5;0 understood well object relatives in which one of the DPs is not lexically restricted, not differently from subject relatives, and avoided this intervention configuration in their production of relative clauses. In contrast, the same children found it difficult to understand and produce object relatives (and object questions) that involved an intervention configuration. Therefore, relative clauses are acquired early and much before age 6, but only around age 6 is the intervention configuration mastered.

Additional approaches attributed the difficulty in relative clauses to partial construction of the syntactic tree in the early stages of acquisition (Radford 1990; Clahsen et al. 1993/4), and others attribute it to a more general complexity in the sentence and/or to pragmatic factors in the comprehension tasks (Sheldon 1974; Goodluck & Tavakolian 1982; Hamburger & Crain 1982; Correa 1995).

In studies examining the production of DP (“direct”) object relatives in infancy using structured elicitation tasks, it was reported that many early relative clauses include a resumptive pronoun, even in languages that do not allow them in adult language (Crain et al. 1990; McKee et al. 1998; Varlokosta & Armon-Lotem 1998; see also Armon-Lotem et al. 2006, for differences between relative clauses containing various resumptive pronouns, and Guasti & Cardinaletti 2003 for children acquiring Italian and French who do not tend to overuse resumptive pronouns). Friedmann et al. (2009), however, found no difference in comprehension of object relatives with and without resumptive pronouns. In Hebrew, a resumptive pronoun is obligatory in PP (“indirect”) object relatives (where it is a part of a prepositional phrase: ze ha-yeled še-ha-yalda histakla al-av, “that’s the-boy that-the-girl looked at-him”), and optional in DP object relatives (ze ha-yeled še-ha-yalda ra’ata [oto], “that’s the-boy that the-girl saw [him]”). According to some researchers, the dependency between the resumptive pronoun and the head of the relative clause is not derived through syntactic movement (Shlonsky 1992); others suggest that some resumptive pronouns are a syntactic last resort phenomena and others have phonological bases and are inserted in the phonological spellout stage (Rasin 2016).

1.2.3 Wh-questions

Wh-questions are derived by wh-movement, which is another instance of A-bar-movement. The wh-phrase moves to the beginning of the sentence, to Spec-CP (see Figure 4). The wh-phrase can move from subject position, creating a subject question (“Who _ built the building?”), or from object position, creating an object question (“What did the child build _?”), and so on.

Figure 4

A-bar-movement of the object in a Hebrew object question et eizo yalda safta ciyra?, acc which girl grandma drew? ‘Which girl did grandma draw?’

In the beginning of the third year of their lives, and even earlier, children produce wh-questions, although at a young age many of their questions are frozen expressions (e.g., “what’s that?”) or ungrammatical (Thornton 1990; Thornton & Crain 1994; Dabrowska 2000; Guasti 2000; Hamann 2000; Soares 2003a; Armon-Lotem 1996; 2008). In an in-depth examination of the stages of acquisition of wh-questions in Hebrew, Armon-Lotem (1996; 2008) identified four stages distinguishable by the characteristics and morphology of verbs used, between the ages 1;6–3;0: verbless questions and formulaic questions; questions in the present tense with non-frozen verbs, inflected for gender and number; questions with inflected verbs in the present and past tense; adultlike questions.

In the spontaneous speech of young children, questions with the wh-elements what, where, and who‎ appear at the average age 2;2–2;4 – before some adjunct wh-questions (why and how), which appear at 2;9–2;11. Lexically-restricted wh-questions (which and whose) are rare even at age 3;0 (Bloom et al. 1989; Rowland et al. 2003).

As for intervention and for the symmetry between subject and object structures, the picture resembles the one reported for relative clauses. Wh-questions are produced early, and subject- and object questions are acquired at the same time (Stromswold 1995 for English). However, the comprehension of object questions, specifically of wh-questions with intervention configurations, i.e., lexically-restricted object questions with which in which the two DPs are lexical (e.g., Which elephant did the-boy hug?), is acquired only around 5–6 years (Hebrew: Friedmann et al. 2009; Friedmann & Novogrodsky 2011; Friedmann & Szterman 2011; Palestinian Arabic: Haddad-Hanna & Friedmann 2015).

Errors in which the wh-phrase is left in its original position (e.g. “grandma cooked what?”) in languages where this structure is ungrammatical are very rare (Guasti 2000). Four-years-old English speakers are able to perform long distance movement in complex wh-questions, and even show sensitivity to syntactic barriers of different kinds (de Villiers et al. 2008).

The derivation of wh-questions in some languages involves movement of the auxiliary from I to C. It was found that children acquiring German, Italian, and Swedish perform this movement successfully. Young children acquiring English, in contrast, perform this movement inconsistently, or show difficulties in Do Support (Clahsen et al. 1993/4; Santelmann & Adger 1998; Guasti 2000; 2002; Rowland & Pine 2000), with marked difference in performing subject-aux inversion between why-questions (in which they show difficulties) and other wh-questions (in which they succeed, Thornton 2008).

2.2.1 Differences between structures with and without movement

The correct repetition of the 60 children in each of the structures is summarized in Figure 6.

Figure 6

The rate of structurally correct repetitions for each structure.

We start by comparing structures with and without movement, using minimal pairs of a basic sentence, and the sentence derived from it through one movement.⁵ In general, sentences without movement were repeated correctly significantly more often (93%) than sentences with movement (61%), t(59) = 7.08, p < .001 (throughout the study we used an α level of 0.05 to determine significance). SV_transitive sentences without V-C were repeated significantly better (92.8%) than the corresponding VS_transitive sentences with V-C (27.2%), t(59) = 13.50, p < .001. SV_unergative sentences without V-C were repeated significantly better (93.3%) than the corresponding V_unergativeS sentences with V-C (35.8%), t(59) = 11.32, p < .001. The repetition of SV_transitive sentences (92.9%) was significantly better than that of sentences with A-bar-movement: subject relatives (72.3%), object relatives (70.3%), and topicalization (69.8%), t(59) ≥ 4.54, p < .001. Thus, in all comparisons between minimal pairs of sentences with A-bar-movement or V-C movement and the parallel structures without movement, the correct repetition was significantly better for the structure without movement. In contrast, A-movement from object position was not different from SV sentences without such movement: no difference was found between SV_unaccusative and SV_unergative sentences, t(59) = 0.25, p = .81.

2.2.2 Structures with similar acquisition patterns

Next, we compared sentences in each group of sentences involving the same type of movement: sentences without movement, sentences with A-bar-movement, and sentences with V-C (A-movement was not included in this analysis, as there was only one kind of structure with A-movement). No significant difference was found between sentences without movement that included transitives or unergatives (93% and 94% average, respectively), t(59) = 0.63, p = .22. No significant differences were found between the various kinds of A-bar-movement (72% subject relatives, 70% object relatives, 70% topicalization), F(2,118) = 0.30, p = .75. Regarding structures with V-C, performance was low for both sentences with transitive and with unergative verbs, but the VS sentences in which the unergative verb moved were repeated better (36%) than sentences in which it was a transitive verb that moved (27%), t(59) = 2.30, p = .01.

To examine the internal consistency of structures with each kind of movement, Cronbach’s alpha coefficients were calculated. It was found that each kind of movement (or lack thereof) is characterized by high internal consistency: SV sentences without movement yielded a Cronbach’s alpha of 0.90; the three A-bar-movement structures yielded Cronbach’s alpha of 0.88; and the two structures with V-C movement yielded a Cronbach’s alpha of 0.82. In light of this, we could perform analyses with the rate of a kind of movement including different structures derived by the same movement.

To assess the organization of the various kinds of structures we performed a factor analysis and a correlation matrix. Three factors were found: no movement (S-V_transitive-O, S-V_unergative), A-bar-movement (subject relatives, object relatives, topicalization), and V-C movement (V_transitive-S-O, V_unergative-S) (Table 2).

As shown in Table 2, the analysis of the children’s performance in repetition displays internal consistency and organization of the types of structures that agrees completely with the linguistic classification of types of syntactic movement, a finding supporting the psychological reality of this classification. Alongside the internal consistency of each kind of movement, a significant difference was found between the three movement types, F(2,118) = 111.76, p < .001. The rates of correct repetitions for the three types of structures were significantly different: there were significantly more correct repetitions of structures with A-movement (94%) than structures with A-bar-movement (71%), t(59) = 5.98, p < .0001; there were significantly more correct repetitions of structures with A-movement (94%), than there were of structures with V-C (32%), t(59) = 14.34, p < .0001; and there were significantly more correct repetitions of structures with A-bar-movement (71%), than structures with V-C (32%), t(59) = 9.15, p < .0001.

2.2.3 Order of acquisition of various movement types

Next, we examined whether it is possible to identify a fixed order of acquisition of the various structures. Based on the internal consistency findings, we combined the various structures in each of the types of movement. The results, organized by movement types, are summarized in Figures 7a and 7b, where each child is presented in a row. We defined mastery of a movement type as repetition that was at least 80% correct. Movement types that each child already mastered are colored blue in the figures. In Figure 7a the rows are ordered by age. As the reader can see, the age of a child does not predict mastering of the various structures: for example, a 2;3 year old child was able to repeat all structures, whereas a 3;7 year old child could only successfully repeat one type of movement – A-movement.

Figure 7

Patterns of success in repeating the three movement types (%correct). Blue cells indicate a structure that the child repeated at least 80% correctly. (a): The 60 participants are arranged by age. (b): The same data as (a), presented by the stage of movement acquisition: Guttman scale of the various movements.

The seemingly chaotic results in Figure 7a, however, become perfectly organized if instead of by age, we order the children by the movement structures they have already mastered, as in Figure 7b. For this purpose, we tested whether the findings follow a Guttman scale. In questions conforming to a Guttman scale, one may, from success in an item that is higher in the scale, infer success in all items ranked below it (Guttman 1944; 1950). For example, success in solving a hard question on a test indicates success solving simpler questions (for example, success in addition of four-digit numbers indicates the examinee will succeed in addition of two-digit numbers). Applying this approach to the acquisition of syntactic structures, we examined whether we can discover a hierarchy of structures: Do children who master a certain structure necessarily also master another structure?

Indeed, we discovered that the acquisition of syntactic movements in Hebrew forms a perfect Guttman scale. When rearranging the children by the movement types they master (i.e., repeated successfully at least 80% of the sentences) as in Figure 7b, rather than by age (as in Figure 7a), the Guttman scale and the order of acquisition of movements in Hebrew are immediately visible: for each child, if they master V-C, they master A-bar-movement and A-movement, and if they master A-bar-movement, then they master A-movement.

This Guttman scale can be translated into an order of acquisition of the various movement structures: first, mastery of SV with unaccusatives is established, and only (but not all) children mastering this structure can successfully repeat structures including A-bar-movement which, in turn, constitute a necessary perquisite for the emergence of V-C movement (Figure 7b). No Guttman scale was found between simple SV sentences and sentences with A-movement, suggesting that they are acquired together.

We found four profiles, describable by the movements a child has mastered, based on a criterion according to which successful repetition of 80% or more of a certain kind of structure is considered success in repetition and indicates that the structure has been mastered (Figure 8):

Figure 8

Order of acquisition of various kinds of movement emerging from the repetition task.

1. 2 children could not repeat (did not reach 80% correct) any of the structures (not even simple SV).

2. 23 children could only repeat simple sentences and/or sentences with A-movement (18 could repeat both, 3 could only repeat sentences with A-movement, and 2 could only repeat simple sentences).⁶

3. 27 children could repeat simple sentences, sentences with A-movement, and sentences with A-bar-movement,⁷ but could not repeat sentences with V-C.

4. 8 children could repeat simple sentences, sentences with A-movement, sentences with A-bar-movement, and sentences with V-C.

Namely, out of the 60 participants, 56 children acquired A-movement, 35 children acquired A-bar-movement (in addition to A-movement) and 8 children acquired V-C (in addition to A-and A-bar-movement). A remarkable property of the children’s repetitions is that they displayed a perfect Guttman scale – there were no children who mastered a late movement and not a prior movement. For instance, no child successfully repeated A-bar-movement but could not repeat A-movement.

2.2.4 Age effect

To examine whether age has an effect on the acquisition of the various structures within the age range tested, we used Pearson correlations between the age of a participant and the rate of correct repetitions of the various structures, and point biserial correlations that determine correlations between the continuous variable of age and the binary variable of performance, with two values defined for it: “success” – 80% correct repetition or more, and “failure” – less than 80% correct repetitions.

Age did not correlate with any of the movement structures (p > .80). It only correlated with the success rate of repeating simple structures without movement, r_pb = 0.27, p = .04, r_pb = 0.31, p = .02. The correlation between age and simple structures was caused by the six children younger than 2;4. Calculating correlation without these children no longer yielded a correlation with age even for simple structures. In conclusion, age was not a good predictor for the mastery of complex syntactic structures in this age range (2;2–3;10).

2.2.5 Error types

3.2.1 The order of emergence of various types of movement and embedding

The sentence production pattern of 56 children can be ascribed to one of the following five profiles (Figures 9 and 10):

Figure 9

The different profiles of acquisition in the samples.

Figure 10

Production profiles of various types of movement and embedding, from which it is possible to draw the order of appearance of these structures. Ages are given in months. Blue cell: the structure appeared in the child’s sample. Unshaded empty cell: no grammatical sentence appeared of this structure. Grey shaded cell: The sample did not include unaccusative verbs.

1. None of the structures – (no SV, no A,A-bar,V-C movement, no embedding).

2. Simple SV sentences and/or structures with A-movement. As we report in detail below - there was no Guttman scale between these structures.

3. Simple SV sentences, structures with A-movement, and wh-questions.⁹

4. Simple SV sentences, structures with A-movement, wh-questions, other structures with A-bar-movement (relatives clauses and/or topicalization), and finite clausal embedding without A-bar movement.

5. Simple SV sentences, structures with A-movement, structures with A-bar-movement, clausal embedding without movement, and V-C with lexical verbs.

These findings allow us to identify an order in the appearance of the various structures in the young children’s spontaneous speech: simple structures with no (order-changing) movement and structures with A-movement appear first, following these, structures with A-bar-movement appear– first represented only by wh-questions, and then (finite) sentential complements appear, together with relative clauses and topicalization, and finally verb movement to C (with lexical verbs) appears. The order of acquisition that arises from the analysis of young children’s spontaneous speech is identical to the order that arose in the repetition task.

3.2.2 Age effect

The analysis of acquisition of the various structures by age indicated that different children proceed through the acquisition of syntactic structures at different paces. Similarly to the findings in the repetition task, although there is a very clear order of acquisition of structures for all children, manifesting a Guttmann scale between the various types of structures, these structures are acquired at different ages for different children. As seen in Figures 7, 10 and 11, each stage includes children in very different ages, and children in the same age belong to different stages of acquisition. Children who did not produce any structures were the youngest (up to 2;1 years old), but there were young children who produced a variety of structures. Similarly, children who produced all movements and embedding were older, but there were older children who still could not produce all structures.

Figure 11

Acquisition ranges of the main structures (from age of emergence to the structure’s stabilization).

Light-blue cell: the structure appeared in the child’s sample. White cell: the structure did not appear.

Black windows frame the range of ages of appearance until stabilization.

Age had no effect, apart from the peripheries – each structure has a youngest age in which it can be found, before which no child produced it (an age we termed “emergence age”), and for a large part of the structures we could identify a “stabilization age” from which on it appears in 80% or more of the samples, thus most children older than it have already acquired it. The main point here is that a stage during which children acquire a specific structure may span over several years, from the age of emergence (in at least one child’s speech) to the age stabilization (in which most children already produce it). Within this range the variance between children is tremendous, and very few correlations with age can be seen, as shown in Tables 5 and 6.

Table 5 summarizes the following factors regarding each of the different structures: first age of emergence in the samples, age from which on the structure stabilizes, the percentage of samples containing the structure from the stabilization age on, and the rate of samples that include the structure within the acquisition range (as a measure of independence from age). Appearance rates of structures with an unaccusative verb were calculated out of the number of samples that included at least one unaccusative verb (46 samples).

3.2.3 Detailed analysis by structure

4.4.1 A-movement

The maturation hypothesis states that the principles of syntax develop with age (Borer & Wexler 1987; Wexler 1994). One specific structure to which Borer and Wexler suggest maturation applies is A-movement, which, according to them matures only at age 4–5 years (and according to the most recent claim, based on an understanding of raising constructions, not before age 7, Borer & Wexler 1987; Hirsch & Wexler 2007). Whereas the present study corroborates the basic claim that syntax matures with age, its findings contradict the claim that A-movement only matures at age 4–5 (or 7): 93% of the children younger than 3;10 succeeded in the repetition of sentences that involve A-movement (unaccusative SV). In spontaneous productions as well, 28 of the 30 children younger than age 4 who produced any unaccusative verb with an argument in their samples, produced sentences with unaccusatives in SV order. Could the children’s success be a result of an incorrect analysis of unaccusatives as unergatives (as Babyonyshev et al. 2001, suggested)? The different production patterns of structures with unaccusative and unergative verbs contradict this hypothesis. In the children’s spontaneous productions, structures with unaccusatives were produced in SV order as well as in VS order, whereas structures with unergatives were rarely produced in VS order. This indicates that from the onset of the production of sentences with unaccusatives, children distinguish them from unergatives, and there is no foundation to the claim that early production of unaccusatives in SV order stems from children analysing them as unergatives. Another indication for the analysis of unaccusatives in SV order as unaccusatives rather than unergatives is the children’s use of possessive datives in this structures, which are markers of internal arguments (Borer & Grodzinsky 1986). In conclusion, very young Hebrew-speaking children master A-movement. This finding joins existing evidence for this ability at a young age (Pierce 1992a; Déprez & Pierce 1993; Fox & Grodzinsky 1998; Sano et al. 2001; Adragão & Costa 2004; Lorusso et al. 2005; Friedmann 2007; Shimada & Sano 2007; Costa & Friedmann 2009; 2012; Friedmann & Costa 2011).

Another finding with respect to unaccusatives was that children are sensitive to the rules that govern the distribution of pronouns and lexical DPs, and produce, as in adult language, lexical DPs before and after an unaccusative verb (equally), but produce pronouns, kinship terms, proper names, and demonstratives only before the unaccusative verb.

Finally, A-movement was presented in the children’s samples almost exclusively by sentences with unaccusatives, and only one structure that may be a raising construction was produced. This finding is consistent with what has been reported on the rare use of raising structures with the verb seem in the spontaneous language of English-speaking children (Hirsch & Wexler 2007). In addition, no verbal passive appeared in any of the children’s samples, and this finding matches the reports in the literature regarding the scarce use of passives in Hebrew (Berman & Sagi 1981; Berman 1997; Ayal-Smikt 2001; Jisa et al. 2002).

4.4.2 A-bar-movement

We examined a variety of structures derived by A-bar-movement: wh-questions, relative clauses, and topicalization, each of various kinds. The analysis of the speech samples indicated that the three structures, albeit derived from the same movement type, are not acquired together. Wh-questions were the first A-bar-structure to appear in children’s spontaneous speech. Wh-questions started to appear in the earliest samples: from age 1;6 they appeared in about half of the samples up to age 2;5, and from that age on they appeared in almost all (92%) the samples. These findings are in accordance with Armon-Lotem’s (1996; 2008) findings on the early acquisition of questions in Hebrew. Relative clauses and topicalization structures were acquired together (as shown in both the repetition task and the spontaneous speech samples), and later than wh-questions but still at a very young age, starting at around age 2;6, with great variance between the children after this point, and stabilizing around age 3;7–4;0 (consistent with previous findings in literature: Berman & Herzberg 1983; Berman 1997; McKee et al. 1998; Soares 2003a; b; Adragão & Costa 2004; Friedmann et al. 2011; Biran & Ruigendijk 2015; Lustigman & Berman 2016; Berman 2018). In the repetition task, in which we examined subject relatives, object relatives, and topicalization, no significant differences were found between these structures, and high internal consistency was found, indicating that relative clauses and topicalization structures, all derived by this A-bar movement are acquired together. (Notice that wh-questions were not tested in repetition).

4.4.3 Embedding

One of our main research questions, which led us to include embedded sentences in the analysis, was whether the order of acquisition of the different constructions is related to movement or to the tree structure. A striking finding was that finite embedded clauses without A-bar movement appeared at the same stage as relative clauses and topicalization structures. This finding suggests that movement types cannot fully account for the stages of acquisition.

We also examined the order of acquisition of various types of embedded structures as reflected in the order they appeared in spontaneous speech.

Nonfinite complement clauses of modals were the first to appear, before finite embedded clauses. They appeared already at age 1;6, and until age 2;5 were the only embedding structures in the children’s spontaneous speech (in line with Bowerman 1979; Kaplan 1983; Bloom et al. 1989; Diessel 2004; Owen & Leonard 2006; Berman 2018; Berman & Lustigman 2020). Then, from age 2;5, 89% of the children had already produced finite embedded clauses, and from age 3;3 all have. In the longitudinal data, finite embedding first appeared at ages 1;11–2;3. This order of nonfinite clauses before finite ones may be ascribed to the part of the tree that they involve, suggesting that nonfinite embedding to modals does not require higher nodes of the tree, but finite embedding does. Nonfinite embedding was acquired in two stages: the first cases of nonfinite embedding were all embedded to modal verbs (mainly want and going-to), and these already appeared in the stage in which children produced SV structures, before the acquisition of wh-questions and relatives/topicalization. Nonfinite embedding to other verbs appeared later, together with finite embedding, relative clauses, and topicalization.

Declarative and interrogative embedding appeared together. The co-appearance of declarative and interrogative embedding differs from Bowerman (1979), who reported declarative before interrogative embedding in English. (This can be attributed to the fact that in English the embedded interrogative clause is structurally different from the corresponding root question.) The finding also differs from Kaplan (1983), who took a different measure of acquisition (stabilization) and reported interrogative before declarative embedding in Hebrew. We also found that embedded clauses that are complements preceded embedded adjuncts, and that sentential complements of verbs preceded complements of nouns. Relative clauses, which are also clauses embedded to nouns, preceded noun clausal complements.

Finally, we found that finite sentential complements and relative clauses appear together, similarly to Armon-Lotem (2005 for Hebrew, Penner 1992; Müller & Penner 1996, for Swiss German; see Steel et al. 2013 and Nir 2016 for review and discussion).

4.4.4 Verb movement to C

Both the repetition results and the spontaneous speech analyses demonstrated that V-C structures are still difficult in the ages we tested, and are acquired relatively late. In repetition, it was the last structure to be acquired, and only 8 of the 60 children succeeded to repeat it consistently; in the spontaneous speech only 7 of the 56 samples of children produced any instance of V-C. This is in line with previous research in Hebrew, and in European Portuguese, another language in which V-C is optional, reporting difficulties in V-C at least until age 6 (Zuckerman 2001; Friedmann & Novogrodsky 2003; Friedmann 2007; Costa & Friedmann 2009; 2012; Friedmann & Costa 2011).

An interesting finding emerged from the spontaneous speech samples, was that of the 7 children who produced a (non-unaccusative) VS structure that could be an instance of V-C, the 4 younger ones were at the stage in which they have acquired wh-questions but no relative, topicalization, and embedding yet. The VS sentences of all these four children were with the verb “do”. Only the three older children (aged 5;8–6;1) used other, lexical verbs, and all three were already at the third stage – they all also produced relative clauses/topicalization and embedding. The longitudinal analysis yielded the same picture: in the first stages the children only produced V-C structures with “do”, and once they acquired relatives, topicalization, and embedding, additional verbs appeared in their V-C structures.

It might be that “do” in these structures moves to a lower portion of the tree than V-C structures with a lexical verb or that V-C with “do” is not part pf a register that needs to be anchored in higher functional nodes. (See a related discussion of the Dutch doen, which appears early in V2 position, and suggested not to reflect the acquisition of V2, Jordens 1990; Roeper 1991; Hollebrandse & Roeper 1996; Zuckerman 2001).