3.1 Early production of clefts

Only a few studies investigate the production of clefts. Hupet & Tilmant (1989) designed an elicited production task and tested French speaking children from 4 to 10 years. Their results allowed to identify a significant subject/object asymmetry in cleft production. Children produced more clefts in the subject condition than in the object condition. Contrastive stress on the object was the strategy that children preferred to use in order to focus this constituent.

Soares (2006), working specifically on European Portuguese, examined the spontaneous production of constructions entailing the left periphery of the clause, such as clefts, in the speech of three children from 1;2 to 4;6 years. She explored the idea that syntactic complexity constrains the emergence order of the different constructions involving the C-domain. Her idea is that language acquisition is affected by developmental constraints such as working memory, which are sensitive to the complexity of derivations (Jakubowicz 2011). In particular, she proposed that embedding entails a dependence relation between the head of the embedded clause and a superordinate category which represents an additional computational load for children. From this work resulted the prediction that cleft constructions that do not involve embedding (namely, é que clefts) are expected to emerge earlier than other clefts.

Lobo, Santos & Soares-Jesel (2016) did a combined analysis of Soares (2006) and Santos (2006) data and examined the spontaneous production of clefts in the speech of six Portuguese children (1;6–4;6). They found that there was no statistically significant difference between the emergence of é que clefts and the emergence of standard clefts in spontaneous production, but they also concluded that the production of clefts which include a wh-clause, such as pseudoclefts, was extremely rare, whereas standard clefts and é que clefts were significantly more frequent in child speech. According to the authors, the production of clefts including a wh-constituent was delayed because children must learn that in these clefts an anaphoric dependency between the clefted constituent and the wh-constituent must be computed. Another relevant finding was the fact that subject clefts were more frequent than clefts presenting a clefted constituent other than a subject. The two types of asymmetries were confirmed by the authors with an elicited production task used to test children from 3 to 6 years: children and adults almost never produced clefts in conditions other than the subject condition;⁹ in addition, wh- clefts were also found to be rare in elicited production, contrasting with é que and standard clefts.

Puppo, Pivi & Cardinaletti (2015), in a study on the acquisition of Italian, found the same extremely marked asymmetry between elicited production of subject and object clefts (only those two types of grammatical relations were considered): children never produced object clefts, even though they produced some object relatives (at significantly lower levels than subject relatives, as expected).

Therefore, Hupet & Tilmant (1989), Puppo, Pivi & Cardinaletti (2015), and Lobo, Santos & Soares-Jesel (2016) have all found in the production of cleft structures an asymmetry between subject clefts and object clefts, even though extended to the clefting of other constituents in the latter study. The same study on Portuguese equally found a frequency asymmetry between é que and standard clefts, on the one hand, and pseudoclefts, on the other hand.

3.2 Early comprehension of clefts

The comprehension of a cleft involves syntactic, semantic and pragmatic knowledge, including the knowledge which allows the speaker to: i) interpret changes in word order and link moved elements to their base-generated position; ii) interpret the appropriate information structure, the type of focus and its presupposition. Different studies have investigated how some of these properties of clefts develop in young children. We will review some of them, concentrating mainly on the syntactic properties of clefts, since this is the focus of our research.¹⁰

The syntactic properties of clefts, namely the displacement of a constituent and the chain established between the clefted constituent and the copy left in the cleft sentence, have been considered in different studies, namely those concerned with subject-object asymmetries in the comprehension of clefts.

Bever (1970) used an Act-Out task to investigate children’s comprehension (2;0 to 4;11) of different kinds of structures in which the order of the arguments does not correspond to the canonical subject – verb – object order. The study included cleft sentences, and overall children had worse results in object cleft sentences (It’s the horse that the cow kisses) than in subject cleft sentences (It’s the horse that kisses the cow). These results were confirmed by Lempert & Kinsbourne (1978), who used an Act-Out Task to investigate whether children correctly interpret sentences with non-canonical word orders, including passives and what they call “inverted clefts”, which are it-clefts with a clefted object (e.g. It’s the truck that the wagon bumps). The authors compare the comprehension of these sentences with the comprehension of sentences with canonical word order, including active sentences and subject clefts (e.g. It’s the cow that bumps the horse). The children tested, aged 3;8 to 6;9 years-old, obtained better results with sentences with canonical word orders (actives and subject clefts), and worse results in object clefts and in passives. These authors explain children’s difficulties by hypothesizing that children initially follow an “agent first” heuristic. Therefore, in sentences with the linear word order N – V – N children would interpret the first N as the agent. Dick et al. (2004), who investigated the development of the comprehension of subject and object it-clefts, and also active and passive sentences, by English-speaking children and adolescents aged 5 to 17, comparing typically developing children with different groups of linguistically impaired children, have found similar subject-object asymmetries in the comprehension of it-clefts and longer reaction times in object clefts.

Thornton, Kiguchi & D’Onofrio (2016), however, show that children have grammatical knowledge of constraints that do not simply rely on linear word order. Following previous studies, the authors investigate whether children apply binding principles at levels of representation other than the surface syntax, that is whether they are able to reconstruct in two kinds of cleft sentences, involving the computation of Principle C (e.g. It was Spot that he brushed) and bound variable readings (e.g. It was her pig that every girl carried). To have the right interpretation, children had to be able to reconstruct the cleft constituent in its base position within the cleft sentence. The authors used a truth value judgment task with short stories, following Crain & Thornton (1998). The children tested, aged from 4;0 to 5;5 years-old, correctly rejected sentences that violated Principle C and accepted sentences with bound variable readings in a significant number of cases. The authors argue that these results show that children at 4 years old already have access to an abstract level of representation where reconstruction takes place and use c-command relations to compute the interpretation of cleft sentences.

More recently, Aravind, Hackl & Wexler (2017) question the nature of the subject-object asymmetry identified in the previous literature (namely Bever 1970; Lempert & Kinsbourne 1978; Dick et al. 2004) by investigating English-speaking children’s comprehension of it-clefts and pseudoclefts. Aravind, Hackl & Wexler (2017) aim at assessing children’s syntactic and pragmatic knowledge involved in the comprehension of clefts and argue that most previous experiments used to test the comprehension of clefts ask children to interpret cleft structures in infelicitous contexts. They also aim at determining whether a word-order-based interpretative strategy (an “agent first” heuristic) is indeed what explains children’s interpretation of clefts. For this reason, they compare it-clefts with pseudoclefts: whereas object it-clefts do not present the default SVO order, in the case of pseudoclefts it is the subject cleft that does not conform to SVO (19). As in the discussion carried by Aravind, Hackl & Wexler (2017), we only refer to the word order of subjects and objects with a lexical NP restriction.

The results obtained allow the authors to claim that only in infelicitous contexts do children show difficulties in the comprehension of clefts: in a first set of experiments, the subject-object asymmetry found in previous studies, with lower accuracy in the comprehension of object clefts, was replicated with it-clefts only in infelicitous contexts. When pseudoclefts were tested, a subject-object asymmetry was also found in infelicitous contexts, but in this case children revealed difficulties not in object pseudoclefts (which conform to an SVO word order, see 19b), but in subject pseudoclefts (which show a different word order, see 19c). According to the authors, “children’s unexpected success on object pseudoclefts and failure on subject pseudoclefts is predicted, however, on a view on which children adopt word-order-based interpretive heuristics when they can” (Aravind, Hackl & Wexler 2017: 23). The authors claim that these results do not support a general difficulty with the syntax of clefts (good comprehension was found in felicitous contexts), instead subject-object asymmetries in comprehension of clefts are taken to be merely superficial.

3.3 Subject – object asymmetries in the comprehension of structures involving A-bar movement

It is well-known that subject-object asymmetries have been found in child acquisition of different A-bar movement structures, including wh-questions, relative clauses, and topicalization structures (Corrêa 1995; Friedmann, Belletti & Rizzi 2009; Cerejeira 2009; Adani et al. 2010; Costa, Lobo & Silva 2011; Abalada 2012; Costa, Grillo & Lobo 2012; among many others). The fact that the derivation of (some) clefts involve A-bar extraction of a constituent to a left-peripheral position in the clause (see section 2) makes these structures a good testing ground for different explanations for subject-object asymmetries.

An explanation for these asymmetries, alternative to the “agent-first strategy” mentioned in the preceding section, has been proposed in Friedmann, Belletti & Rizzi (2009) and adopted in many subsequent studies. According to the authors, children’s immature grammatical system is more sensitive to intervention effects: children have trouble connecting the head X of the chain with the tail Z of the moved element when an element Y with at least a partially overlapping set of features intervenes (X c-commands Y and Y c-commands Z):

In the adult grammar, this kind of configuration is only problematic when the featural specification of the intervening element (Y) is not distinct from the featural specification of the moved element. This happens when we try to extract a wh-word from an indirect question, a weak island context, as in the example below:

In (21), the moved wh-word and who bear a structural similarity, since they are both specified with the [+Q] feature. Hence, the connection between the moved element and its copy fails. This has been captured in terms of Relativized Minimality, a locality principle in syntactic theory (Rizzi 1990; see Starke 2001 for a complete discussion).

Object root questions do not present a similar configuration. In (20), above, only the chain, but not the intervener, holds the feature [+Q], associated to the interrogative operator. However, according to Friedmann, Belletti & Rizzi (2009), in children’s grammar, the DP “the dog” may also intervene between a displaced wh-element such as “which monkey” and its copy, since children’s grammar would be characterized by a stricter version of the Relativized Minimality principle and would not admit an inclusion configuration in which the potential intervener partially shares features with the chain, as in (20) (also Belletti & Rizzi 2013). Therefore, children’s grammar (contrary to the adults’ grammar) requires that the chain and the intervening element present a disjoint featural specification. Thus object questions with d-linked wh-elements (as in 20) correspond to configurations that trigger intervention effects in children since the wh-chain and the intervening element share the feature [+NP] (associated to nominal expressions with a lexical restriction). Summarizing, the same locality principle (Relativized Minimality) would be operative in the adult and the child grammars, but children would adhere to a stricter version of Relativized Minimality. Whether this constraint is grammatical or the result of a processing mechanism is a matter of debate. Other studies have shown that adults are also sensitive to these constraints in tasks that are more demanding (e.g. self-paced reading) (Costa, Grillo & Lobo 2012, among others). Therefore, the difficulties may equally be seen as a processing constraint which relies on grammatical features and which is subject to development.

Several studies have investigated which type of features and configurations are relevant for intervention. The effects seem to be stronger when the morphological features shared by the moved element and the intervener overlap (including gender and number features) (Adani et al. 2010; Belletti et al. 2012). However, the effects of a lexical restriction on the moved element and on the intervener have been particularly explored, as explained above. Several studies found asymmetries between d-linked and non d-linked object wh-questions (Cerejeira 2009; Friedmann, Belletti & Rizzi 2009; Baião 2013; Baião & Lobo 2014), between restrictive and free object relatives (Friedmann, Belletti & Rizzi 2009), and between headed object relatives with a pro subject and headed object relatives with a lexical subject (Friedmann, Belletti & Rizzi 2009). In all these cases, when either the moved element or the intervener is a pronoun, therefore not a DP containing a lexically overt noun phrase (it does not have a [+NP] feature), the results in the comprehension of object extractions improve and no subject-object asymmetry was found in children’s comprehension.

The comprehension of object cleft sentences has been less frequently explored in the general framework of the discussion of intervention effects, possibly given the syntactic and pragmatic complexity of clefts. In what follows, we present the rationale and the research questions of the present study, which aims at exploring the existence of intervention effects in clefts, considering the syntactic diversity of different types of clefts in European Portuguese and using a protocol that maintains felicitous conditions for clefts to occur.

4.1 Comprehension of different types of EP clefts: Research questions and predictions

In this study, we consider the well documented subject-object asymmetries in structures with A-bar movement, but also the syntactic properties of different cleft structures available in European Portuguese. In particular, we aim at comparing adults’ and children’s comprehension of é que clefts, standard clefts and pseudoclefts, which we divide in two types according to their syntactic derivation and according to the assumptions made explicit in section 2 ((22) recovers the information on the syntactic structures of these different types of clefts): the derivation of é que clefts (22a) and of standard clefts (22b) involves A-bar movement of a constituent, which may be lexically restricted, as in (22a, b). When the clefted constituent moves, it crosses a DP that may also present a lexical restriction. If both constituents are lexically restricted, an intervention configuration arises. This is typically the case of object clefts; the derivation of a pseudocleft does not involve such movement, since in this case only a non-lexically restricted wh-element (quem ‘who’ in 22c) is extracted internally to the CP. In (22) potential interveners are represented in bold.

If the derivation of the different cleft structures that we consider here indeed coincides with the representations in (22), we expect to find intervention effects (of the type defined in the preceding section, along the lines of Friedmann, Belletti & Rizzi 2009), in é que and standard clefts (22a,b), but not in pseudoclefts (22c). Only in the case of object é que and object standard clefts the moved element (in both cases, the focused constituent) may present a lexical restriction and may cross a lexically restricted DP, the subject in the cleft clause (as illustrated in 22a and 22b). If both the A-bar chain and the intervener share a feature (the [+NP] feature), which is a subset of the set of features in the chain, we will have the type of featural inclusion configuration which, according to Friedmann, Belletti & Rizzi (2009), is possible for adults but justifies prolonged comprehension difficulties in children. On the contrary, and as explained in section 2, object pseudoclefts do not create the same configuration: in this case, the only A-bar chain involves a non-lexically restricted wh-element (quem in 22c).

If, on the contrary, the underlying structure of standard clefts is similar to the structure of pseudoclefts and the clefted material in a standard cleft is not extracted from the embedded clause, we would expect standard clefts to pattern with pseudoclefts, i.e. not displaying intervention effects when an object is clefted.

Finally, it might be that difficulties with object é que and standard clefts do not stem from intervention effects and may simply be explained by the fact that the child uses default assumptions concerning word order to interpret the sentence – this is the case if we assume that the child’s immature parser interprets the sentences according to an “agent-first strategy”, i.e. taking the first lexically restricted NP as the agent, and experiences difficulties in revising this interpretation (see Huang et al. 2013 on passives and Omaki & Lidz 2015 on the development of incremental sentence comprehension mechanisms). This may be seen as a heuristic in which children fall back in infelicitous contexts (Aravind, Hackl & Wexler 2017), but it can also be seen as a more general processing strategy explaining comprehension in felicitous contexts. If an “agent-first strategy” is used, we expect children to show difficulties in object é que and standard clefts (22a) and (22b), but not in object pseudoclefts (22c), since the latter present an SVO order. However, under the same assumption, we also expect children to experience difficulties in subject pseudoclefts (23), as explained in section 2. If indeed children use a word-order-based strategy, we expect a subject/object asymmetry in pseudoclefts, but in the reverse sense, i.e., with subject clefts showing lower accuracy rates.

We therefore consider the following research questions:

1. Are there subject-object asymmetries in Portuguese-speaking children’s comprehension of clefts?

2. Are these subject-object asymmetries, if found, exclusive of é que and standard clefts and do they negatively affect the interpretation of object clefts?

3. If subject-object asymmetries extend to pseudoclefts, does the asymmetry found in these structures go in the reverse sense, i.e. lower accuracy is found with subject pseudoclefts?

Assuming the analysis outlined in section 2 for EP clefts, and assuming that intervention effects (structurally defined, namely in terms of c-command) and not word-order-based strategies determine the interpretation of clefts, we predict:

1. subject-object asymmetries in the comprehension of both é que clefts and standard clefts: in this case, accuracy rates with object clefts are expected to be significantly lower than with subject clefts;

2. the inexistence of a subject-object asymmetry in the comprehension of pseudoclefts, since they do not correspond to a configuration in which featural intervention is expected.

In order to test the comprehension of standard clefts, é que clefts and pseudoclefts, we conducted an experiment using the truth value judgment task (Crain & Thornton, 1998) and manipulating two within-subjects independent variables: type of cleft (standard cleft, é que cleft and pseudocleft) and grammatical relation of the clefted constituent (subject vs. object). In what follows, we explain the method used and the results obtained.

4.3 Results

We begin by presenting the global results obtained in each condition: in Graphic 1, we present the means of individual proportions of correct answers obtained in the six items for each condition. In the graph, as well as in the presentation of the statistical analysis, SC stands for standard cleft, EQ for é que cleft, PC for pseudocleft. In addition, the extension _O stands for an object cleft and _S for a subject cleft (e.g., SC_O identifies the object standard cleft condition).

Graph 1

Global results: mean correct answers.

The observation of the graph allows two main observations, one pointing to a subject/object asymmetry and another pointing to an asymmetry between types of clefts and suggesting the possibility of an interaction between the two variables. In first place, the results globally show lower proportions of correct answers in the case of object clefting than in the case of subject clefting. However, this asymmetry seems to be obvious in the case of standard and é que clefts, but not in the case of pseudoclefts. The child groups, especially the 5-year-olds, seem to show better performance with object pseudoclefts than in other cases of object clefts (é que and standard).

A generalized linear mixed model (GLMM) executed in SPSS 22 was used to analyze the data. We wanted to verify the effect of the within subjects independent variables (Cleft type and Grammatical relation) and the effect of age as between subject factor. Since our hypotheses justified the expectation that a subject-object asymmetry would affect only children and would not be found in all types of clefts, we wanted to verify the effect of a Cleft type by Grammatical relation by Age interaction. Finally, we needed to accommodate the effects of a possible “yes-bias effect” affecting the judgment of children, which justified an Age by Expected answer (True vs. False) interaction.

Therefore, fixed effects entered into the model were Age_Group (4 years, 5 years, adults), Cleft type (SC, EQ, PC), Grammatical relation (Subject vs. Object) and a three-way interaction Age_Group by Cleft type by Grammatical relation interaction. In addition, a two-way interaction Age_Group by Expected_Answer (True vs. False items) was entered into the model. Subject was entered as a random factor. The model showed significant main effects of Age Group (F(2,2139) = 4.024, p = .018), and of the Age Group by Cleft type by Grammatical relation interaction (F(12,2139) = 5.118, p < .001). It also showed significant main effects of the Age Group by expected answer interaction (F(3,2139) = 12.904, p < .001).

We should first focus the yes-bias effect found in the data. As expected, the child groups, and particularly the 4-year-olds, are more likely to give the expected answer when this answer is True than when it is False. Sidak-corrected pairwise comparisons included in the model show significant differences between True and False items in the two child groups, but not in the adult group, which go in the expected direction: 4 years, True vs. False, t(2139) = 5.844, p < .001, 5 years, True vs. False t(2139) = 2.838, p = .005, adults, True vs. False n.s (t(2139) = .028, p = .978). However, the model equally showed the effect of the independent variables and interactions under scrutiny.

As for the global effect of Age, the model shows significant differences between the two child groups (4 vs. 5 years, t(2139) = –2.836, p = .014), but it showed no significant difference between 5 year-olds and adults. In this case, we need to look at the three-way interaction including age.

As for the interaction Age_group by Cleft type by Grammatical relation, which is the most relevant piece of the analysis allowing us to discuss our hypotheses, the Sidak-corrected pairwise comparisons included in the model allow us to confirm our predictions. First, we contrast subject and object clefts for each cleft type within each age group. In this case, a major contrast is found confirming a difference between pseudoclefts and the other types of clefts. In the case of pseudoclefts, in no group a significant difference was found between subject and object clefts. This contrasts with what happens in the case of é que clefts and standard clefts: in this case, in the two child groups and also in the adult group, we found a significant effect of the subject/object contrast. In what follows, we detail these results: (i) for the adult group, and in the case of é que clefts, S vs. O, t(2139) = 2.851, p = .004; in the case of standard clefts, S vs. O t(2139) = 3.019, p = .003; (ii) for the 5 year-old group, in the case of é que clefts, S vs. O t(2139) = 7.723, p < .001; in the case of standard clefts, S vs. O t(2139) = 7.229, p < .001; (iii) for the 4 year-old group, in the case of é que clefts, S vs. O t(2139) = 3.948, p < .001; standard clefts, S vs. O t(2139) = 4.082, p < .001.

In addition, a comparison between the different types of object clefts for each age group shows significant differences between object pseudoclefts and object é que and standard clefts in the 5 year-old group: object pseudocleft vs. object standard cleft t(2139) = 6.069, p < .001; object pseudocleft vs. object é que cleft t(2139) = 6.662, p < .001. In the case of the 4 year-olds, who show in pseudoclefts global results lower than the other age groups (see Graph 1), the difference does not reach significance.

We can therefore conclude that (i) a subject vs. object asymmetry is found only in the comprehension of é que and standard clefts (as opposed to pseudoclefts) and (ii) clefting an object in the case of a pseudocleft does not affect children’s comprehension in the same way as clefting an object with a standard cleft or an é que cleft. Moreover, the analysis shows that standard and é que clefts pattern together.