DISCUSSION

The present study investigated the ability of Greek-speaking TD children to comprehend complex constructions that entail wh-movement, such as wh-questions, RCs, and FRs. Several asymmetries were revealed across conditions.

Regarding wh-questions, although children as a group performed relatively highly across conditions, they faced more difficulties interpreting referential which-NP object questions. A subject/object asymmetry was observed in referential which-NP questions and in non-referential what questions, but not in non-referential who questions. This is not in line with previous findings on Greek, which report subject/object asymmetries in both referential and non-referential who-questions (see Stavrakaki, Reference Stavrakaki2001a). One could argue that the discrepancy between the findings of the two studies is due to the age difference between the children in our study (their mean age was 5;4) and those in Stavrakaki (Reference Stavrakaki2001a) (their mean age was 4;1). However, the fact that the pattern of asymmetries is identical in the two age groups of our study (i.e. in the four-year-olds, which is the age group that has similar mean age as the TD group in the Stavrakaki, Reference Stavrakaki2001a, study, and in the five-year-olds) indicates that the asymmetry between referential vs. non-referential who questions in our study is a genuine asymmetry, which did not surface in Stavrakaki's (Reference Stavrakaki2001a) study, probably because of the small sample size. Moreover, the wh-comprehension data revealed two important findings: (a) the subject non-referential who questions were better interpreted compared to the subject non-referential what counterparts, and (b) besides the asymmetry between object referential which-NP questions and object non-referential who counterparts, an asymmetry was also found between object who and object what questions.

With respect to RC comprehension, children presented subject/object asymmetries across relative types. Specifically, better performance was observed in subject compared to object dependencies in both right-branching and center-embedded RCs. Note that the subject/object asymmetry attested in the right-branching RCs is in contrast to the patterns observed in Stavrakaki (Reference Stavrakaki2001b), where better performance was attested in O-O relatives (our object right-branching RCs) compared O-S relatives (our subject right-branching RCs). However, no statistical analysis was performed in Stavrakaki (Reference Stavrakaki2001a) to confirm the O-O vs. O-S asymmetry. Additionally, the type of embedding had an impact on children's overall interpretation of RCs; a difference in the interpretation of right-branching and center-embedded RCs was observed with the latter being harder, especially in the object condition. A similar effect was also attested in Stavrakaki's (Reference Stavrakaki2001b) data, namely, S-O relatives obtained low accuracy scores compared to all other relatives (although this asymmetry was also not confirmed by a statistical analysis). Regarding object RCs in which the features of the intervening subject were manipulated, an asymmetry was observed between those in which the subject included a quantificational restrictor (objRC_Q) and those in which the subject included a quantificational phrase (objRC_Q+NP), with the latter being better comprehended by children.

Last, our data from the FR comprehension task revealed a subject/object asymmetry in FRs where a free relative restrictor and an NP were included, while no such asymmetry was attested between subject- and object-extracted FRs with a free relative restrictor. Moreover, our data revealed a dissociation between object FRs with an NP and those without an NP (FR_wh/whoever+NP-object vs. FR_wh/whoever-object), with the latter being better preserved.

Let us now turn to the predictions of RM to see whether the patterns obtained in our data can be efficiently accommodated by this account. Table 8 summarizes our findings along with Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) predictions.

Table 8.

Predictions from Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) account and findings

notes: RB = right-branching; CE = center-embedded; FR_wh/whoever: sentence with a free relative restrictor; FR_wh/whoever+NP: sentence with a free relative restrictor and an NP; RC _Q: relative clause in which the subject is a quantificational restrictor; RC_Q+NP: relative clause in which the subject is a quantificational phrase.

¹ However, although high accuracy scores were revealed in subject non-referential who and what questions, an asymmetry was found between them, in that performance on subject what questions was significantly lower than performance on subject who questions (see discussion in this section).

² Likewise, despite the high accuracy scores attained in object non-referential who and what questions, object what questions received significantly higher scores than object who questions (see discussion in this section).

³ Although low performance was observed in the RC condition with a quantificational phrase (obj RC_Q+NP), Friedmann et al.'s account predicts worse performance in this condition compared to the RC condition with a quantificational restrictor (obj RC_Q) (see discussion in this section).

As shown in Table 8, the predictions within the extended version of the RM approach, proposed by Friedmann et al. (Reference Friedmann, Belletti and Rizzi2009), are confirmed by our data to a great extent. There are, however, some findings which cannot be straightforwardly accommodated within this account. More specifically, the main claim that comprehension will be compromised when both the intervening subject and the moved A’-element in object-extracted constructions like wh-questions, RCs, or FRs, contain a lexical NP, was supported by the child Greek data. Nevertheless, the asymmetry between object who and object what questions is not predicted by Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) RM approach. Moreover, lexical NP restriction does not explain the pattern observed in object RCs with a quantificational subject, since lack of lexical NP restriction in these structures caused more difficulties to children than its presence.

Consistent with Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) predictions, our data from wh-questions revealed comprehension difficulties in object referential which-NP questions compared to subject referential which-NP counterparts and object non-referential who questions. This finding corroborates the extended version of RM, since object referential questions involve movement from a postverbal position (see structure (13) in the previous section), and both the moved element and the intervening subject DP are lexically restricted, thus resulting in intervention/minimality effects. However, if the presence of lexical NP restriction in the moved wh-phrase was, in fact, the only source of the comprehension difficulties that children have, no difference would be predicted between object non-referential who and what questions, since both constructions lack lexical NP restriction. Nonetheless, an advantage of object what over object who questions was attested in the data. This asymmetry remains unaccounted for on Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) account, since both who and what object constructions involve wh-movement in which the intervening subject DP is lexically specified, while the moved wh-word is not. This asymmetry has also been observed in agrammatic aphasia, and has been attributed to the mismatch in the grammaticalized [±animate] feature in what questions (Garaffa & Grillo, Reference Garraffa and Grillo2008). According to Garaffa and Grillo (Reference Garraffa and Grillo2008), the moved [–animate] object differs in animacy from the intervening [+animate] subject DP in object what questions, helping speakers construct the correct syntactic representations. In contrast, in object who questions, the same [+animate] feature is involved in both the moved wh-word and the intervening subject DP, thus leading to a chain formation blockage. Note that it has also been shown that in typical language acquisition, animacy mismatch enhances comprehension in object-extracted conditions (Arosio et al., Reference Arosio, Guasti and Stucchi2011). In line with this reasoning, we suggest that the φ-features of the moved and of the intervening element also play a significant role in the comprehension of constructions involving wh-movement. In Greek, the wh-word what is not specified for gender, person, and number features, whereas the intervening subject is fully specified for these features. Thus, we argue that the lack of feature specification in the moved wh-word, as shown in (14), facilitates children's comprehension of object what questions.

1. (14)

Note that an advantage of who over what questions was observed in the subject questions. This pattern is not related to minimality effects and might be due to the fact that, for some of the verbs, the use of the pronoun what was not preferred for animate agents in the subject questions; 64% of the errors in the subject what questions occurred with the verbs fotoγrafizο ‘photograph’, pleno ‘wash’, and kitao ‘look’ (see the ‘Appendix’ for the test sentences with these verbs), which may not be as tolerant of [–human] agents.

Turning to RCs, Friedmann et al. (Reference Friedmann, Belletti and Rizzi2009) would predict subject/object asymmetries across relative types, since in object-extracted RCs, the object has to move over an intervening subject, and both the moved element and the intervener share lexical NP restriction. This prediction was confirmed by our data, since subject/object dissociations were attested across relative types. Note, though, that center-embedded RCs were found to be harder to interpret compared to right-branching ones, especially in the object condition. Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) account does not make predictions for differences between different relative types, but this finding is in line with studies showing that center-embedded RCs are harder to parse compared to right-branching counterparts due to sentence complexity, memory limitations, or integration costs (Bates, Devescovi & D'Amico, Reference Bates, Devescovi and D'Amico1999; Corrêa, Reference Corrêa1995; Kidd & Bavin, Reference Kidd and Bavin2002).

With respect to FRs, our data are consistent with Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) account. A subject/object asymmetry was attested in FRs in which the moved DP included a free relative restrictor and a lexical NP (FR_wh/whoever+NP), suggesting that, in line with Friedmann et al.'s predictions, an intervener can block chain formation when it is similar in terms of lexical NP restriction to the moved element. In contrast, an intervening subject DP does not block chain formation in the case of FRs without a lexical NP (FR_wh/whoever); the dissimilarity in terms of lexical NP restriction between the moved element and the intervening subject DP leads to high performance in object-extracted sentences, further supporting Friedmann et al.'s account. The dissociation attested between object FRs with a lexical NP and object FRs without a lexical NP in the moved DP further supports the claim that the involvement of a lexically restricted intervener deteriorates children's performance.

Challenging the efficacy of that account, though, poor performance was attested in object RCs in which the intervening subject is a quantificational restrictor (objRC_Q), as well as those in which the intervening subject is a quantificational phrase (objRC_Q+NP). Note that Friedmann et al. (Reference Friedmann, Belletti and Rizzi2009) showed that in RC conditions where the intervening subject was an arbitrary pro, and thus had no lexical NP restriction, no minimality effects were obtained. However, they acknowledge the fact that the lack of intervention effects in this case may not be due to the absence of lexical NP restriction but to the null nature of the subject pro. In the light of this finding, we included in our study object RC conditions with a quantificational restrictor (objRC_Q) to assess whether there are minimality effects in the absence of lexical NP restriction in the intervening subject but in the presence of a quantificational operator, such as kapjos ‘some’. Moreover, we included object RC conditions with a quantificational phrase (objRC_Q+NP) to assess whether the presence of a lexical NP in the intervening quantificational subject creates additional comprehension difficulties for children. Contra Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) predictions, object RCs with a quantificational restrictor (objRC_Q), namely, structures that include subjects without lexical NP restriction, were found to be harder for children to interpret than those involving a quantificational phrase (objRC_Q+NP), namely, structures that include subjects with lexical NP restriction. The poor performance observed in our data in the condition of object RCs in which the subject was a quantificational restrictor (objRC_Q) suggests that the lexical restriction of the subject is not the critical factor for intervention effects. Instead, we argue that the feature specification of the intervening quantificational subject DP in both objRC_Q and objRC_Q+NP conditions, and the fact that both the moved and the intervening element in these conditions contain quantificational features, causes difficulties in children's comprehension. According to the RM approach to intervention effects developed in Grillo (Reference Grillo, Blaho, Schoorlemmer and Vicente2005, Reference Grillo2008, Reference Grillo2009), and subsequently advocated by Costa et al. (Reference Costa, Grillo and Lobo2012), in order for a speaker to distinguish between a moved and an intervening DP, s/he must activate, select, and maintain the activation of the feature structure of both DPs. This has a processing cost, which depends, among other things, on the complexity of the associated semantic representation and on the position of each feature in the hierarchy of morphosyntactic features. This entails that “the more features required to distinguish between the moved and the intervening DP, the higher the processing cost” (Costa et al., Reference Costa, Grillo and Lobo2012, p. 145). We conjecture that the poor performance observed in our RC conditions with an intervening quantificational subject was due to the features involved in the intervening quantificational subject DP, which induced in turn a structural (and possibly a semantic) complexity that increased processing load, because a quantificational element with a rich internal structure is intervening in the path of a quantificational movement (i.e. between a moved A’-element and its trace/copy). Specifically, the quantificational subjects kapjos ‘some’ and kapjos stratiotis ‘some soldier’ have a richer internal structure compared to the internal structure of subject DPs, such as o stratiotis ‘the soldier’, used in the non-quantificational RC conditions. Their internal structure includes argumental features such as case, number, and gender. Moreover, the element kapjos in Greek is an indefinite quantificational restrictor that has been argued to be ambiguous between a specific and a non-specific reading (Veloudis, Reference Veloudis1982). Thus, its internal structure specifies further features, such as indefinite specific or indefinite non-specific. If reduced processing resources, due to the additional features required to distinguish between the moved and the intervening quantificational DP in objRC_Q and objRC_Q+NP structures, make it impossible to activate (or maintain) the full array of features in these structures, their syntactic representation gets impoverished, resulting in an underspecified representation of the intervening DP, which generates a minimality effect, since it does not allow the construction of the relevant chain between the moved wh-element and its trace/copy. We assume, based on Grillo (Reference Grillo, Blaho, Schoorlemmer and Vicente2005, Reference Grillo2009), that the impoverished representation of an objRC_Q structure such as (15) looks like (16) (similar representations hold for objRC_Q+NP structures).

1. (15)

1. (16)

An anonymous reviewer raised the issue that since it is known that some quantificational structures may be difficult for children, one needs to know whether the difficulties observed in the sentences with quantificational subjects in the current study are due to the feature specification of these quantifiers or to a general delay in their comprehension. Although difficulties have been observed with the ‘inferred’ meaning of existential quantifiers (e.g. that ‘some’ implies ‘not all’; Noveck, Reference Noveck2001; Papafragou & Musolino, Reference Papafragou and Musolino2003, for the Greek quantifier meriki ‘some’), what is crucial for our purposes is that existential quantifiers are among the first ones to be acquired cross-linguistically (Hanlon, Reference Hanlon and Kessel1988) and that difficulties do not arise with their ‘semantic/logical’ meaning (i.e. with the meaning that some means ‘at least one’). Even though there are no studies on the acquisition of kapjos ‘some’ in Greek, Pouscoulous, Noveck, Politzer, and Bastide (Reference Pouscoulous, Noveck, Politzer and Bastide2007) show that four-year-old French-speaking children have perfect mastery of the ‘semantic/logical’ meaning of certains ‘some’ (i.e. they know that certains means ‘at least one’), even though they have not mastered the ‘inferred’ meaning at adult-like rates. Given that in our experiment only the ‘semantic/logical’ meaning of kapjos ‘some’ was available (e.g. for the sentences δikse mu ton zoγrafo pu kapjos kiniγa ‘show me the painter that someone is chasing’ and δikse mu ton zoγrafo pu kapjos stratiotis kiniγa ‘show me the painter that some soldier is chasing’, there was one soldier depicted in the figure), it is reasonable to conclude that the difficulties we observed in the RC conditions with quantificational subjects are due to their impoverished syntactic representation, because of the features involved in the intervening quantificational subject DP in these conditions.

If our explanation regarding the poor performance in object RCs with a quantificational intervening subject is on the right track, there is still an issue that needs further clarification, namely, why object RCs with a quantificational restrictor (objRC_Q) were harder to comprehend than object RCs with a quantificational phrase (objRC_Q+NP) or other object RCs examined in the study. Recall that the proportion of correct responses for objRC_Q was 42%, as opposed to 63·1% for objRC_Q+NP, 65% for object right-branching RCs, and 52·3% for object center-embedded RCs. It has been proposed that kapjos is considered to be an anti-specific indefinite which is referentially vague, i.e. it is used when the speaker does not have a fixed value in mind (Giannakidou & Quer, Reference Giannakidou and Quer2013). Experimental data confirm this claim, in that Greek native speakers do not prefer to use kapjos in specific contexts (Giannakidou, Papadopoulou & Stavrou, Reference Giannakidou, Papadopoulou and Stavrou2012). Hence, Greek children may have experienced difficulties with the interpretation of kapjos in our experimental sentences because a specific reading of the quantificational restrictor kapjos is required in them. In other words, the use of an anti-specific noun in specific positions, such as the preverbal subject position in our experimental conditions, creates a semantic/pragmatic conflict for children, possibly because of their reduced processing resources. As a result of this conflict, children performed very poorly in RCs with the indefinite quantificational restrictor kapjos in the preverbal subject position. This line of reasoning predicts that performance should improve in RC conditions in which the anti-specific indefinite kapjos appears in the postverbal subject position (e.g. δikse mu ton zoγrafo pu kiniγa kapjos instead of δikse mu ton zoγrafo pu kapjos kiniγa ‘show me the painter that someone is chasing’), which is a position more consistent with anti-specificity (Anastasia Giannakidou; personal communication). Moreover, if this line of reasoning is in the right direction, it raises the question why structures involving a quantificational restrictor and an NP (objRC_Q+NP) turned out to be easier for children compared to structures involving only a quantificational restrictor (objRC_Q). We conjecture that structures involving a quantificational phrase, although referentially vague, also generate a domain restriction, which creates more informative sentences (Anastasia Giannakidou; personal communication). Hence, these structures are easier to comprehend compared to objRC_Q structures because the NP aids children to attribute a specific reading to kapjos in a position that enforces a specific reading in any case, such as the preverbal subject position. To sum up, children's low performance on object RC conditions with a quantificational restrictor or a quantificational restrictor and an NP was attributed to the feature specification of the intervening quantificational subject DP, which triggers an impoverished/underspecified representation due to reduced processing resources that children have. Nonetheless, it was argued that the position of the intervening quantificational subject, namely, the preverbal subject position, also contributed to the very low performance observed in object RCs with a quantificational restrictor (which, in fact, was lower compared to the performance observed in the other object RC conditions), since it imposed a specific reading to a quantificational element that has an anti-specific meaning. We leave open for future research the question whether performance will improve in object RCs with the quantificational restrictor kapjos in postverbal subject positions.

The last issue we would like to address concerns the effect of case marking in facilitating children's comprehension of various structures involving wh-movement, such as wh-questions, RCs, and FRs. We found that case affected children's performance, since more errors were observed in the following instances: (a) in referential questions which included feminine interrogative pronouns (i.e. those without overt/morphological case marking), compared to referential questions which included masculine ones (i.e. those with overt case) in both subject and object dependencies; (b) in object RCs with a quantificational DP in which the quantificational subject restrictor was marked for feminine gender, and thus was not overtly marked for nominative, compared to object RCs in which the quantificational subject restrictor was marked for masculine gender, and thus was overtly marked for nominative case; and (c) in object FRs in which the free relative pronoun was marked for feminine gender, and thus was not overtly marked for accusative case, compared to object FRs in which the free relative pronoun was marked for masculine gender, and thus was overtly marked for case. These findings provide evidence for the claim that case facilitates children's comprehension of subject/object dependencies in languages, like Greek, that mark their arguments with morphological case. In fact, it was shown that case facilitates not only the comprehension of subject/object dependencies in RCs, as also shown in Stavrakaki (Reference Stavrakaki2001b), but also the comprehension of subject/object dependencies in other structures that involve wh-movement, such as wh-questions and FRs. Recall that Stavrakaki (Reference Stavrakaki2001a) argued that children's performance improved considerably in RCs in which the head and the subject of the RC were marked for case (accusative and nominative), as opposed to RCs in which the object head and the subject of the RC were not marked for case, since they were both neuter DPs. As noted before, though, it is not obvious that it was case and not gender that facilitated performance in the relatives with non-neuter nouns, because their heads and their subjects had different gender features (one was masculine and the other feminine). In our study, the head and the subject of object RCs were marked with identical gender values (i.e. either both masculine or both feminine). Similarly, the wh-pronoun in the referential wh-questions and the free relative pronoun in the FRs with a free relative pronoun and an NP were marked with identical gender values (i.e. both masculine or both feminine). Thus, our study provides new and more robust evidence compared to previous studies that case facilitates the interpretation of object dependencies, since higher accuracy was observed in the structures that involved arguments with masculine gender, where case is overtly marked.

An anonymous reviewer raised the question of in what way Case features are active by the time the sentences are processed, since Case features are supposed to be eliminated during the course of the derivation, as a result of checking. We follow the well-known distinction between abstract (syntactic) Case and morphological case (see Harley, Reference Harley1995; Marantz, Reference Marantz1991). In the present study we are concerned with morphological case, which is the realization of abstract Case at the level of Phonetic Form (PF), hence accessible to the speaker at the time the sentences are processed. Moreover, note that within current assumptions in the Minimalist Program, checking of uninterpretable Case (and φ-) features leads to their valuation at PF (see Chomsky, Reference Chomsky and Kenstowicz2001, and the literature building on the notion of valuation as a result of checking).

If the claim that morphological case facilitates children's comprehension is on the right track, there are two remaining issues that need further discussion. First, why were more errors attained in the sentences with overt case marking (i.e. those with masculine pronouns) compared to the sentences without overt case marking (i.e. those with feminine pronouns) in the case of FRs with a subject dependency? We conjecture that the pattern observed in FRs with subject dependencies was due to the case matching required in these structures (for discussion on case matching in Greek FRs, see Alexiadou & Varlokosta, Reference Alexiadou, Varlokosta and Alexiadou2007; Philippaki-Warburton & Stavrou, Reference Philippaki-Warburton and Stavrou1986; among others). In particular, the fact that the relative pronoun was marked with accusative, as the object argument of the main verb, although it was associated with the subject position of the FR, possibly created a conflict for the children, leading them to more errors in the FR clauses with DPs marked with masculine gender. In contrast, the unmarked for case relative pronoun in the FR clauses that contained DPs with feminine gender value facilitated children's association of the relative pronoun with the subject position of the FR clause. Case matching is also the reason why subject/object asymmetries were not attested in FR clauses that contained DPs with masculine gender value, as opposed to those that contained DPs with feminine gender value. Again, the fact that the relative pronoun was marked with accusative case in the FRs with masculine DPs, although it was associated with the subject position of the FR, possibly diminished children's performance in FRs with a subject dependency, thus minimizing the difference in performance between subject and object dependencies.

The second issue that calls for further discussion is whether there are indeed genuine minimality effects in non-canonical structures in a language that marks its arguments with case, given the evidence we found that case acts as a facilitator to reduce these effects. We argue that there are genuine minimality effects, despite the fact that case facilitates the comprehension of non-canonical structures. First, subject/object asymmetries arose in right-branching and center-embedded RCs, although both arguments in these structures were marked with overt case (nominative for subjects, accusative for objects). Second, even though one could argue that the lack of subject/object asymmetries in non-referential who questions was not due to the absence of lexical NP restriction, as we claimed within Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) RM approach, but was a result of the presence of case in the moved and in the intervening argument, the subject/object asymmetry observed in the referential questions with masculine wh-pronouns (i.e. pronouns marked with overt case) indicates that, despite the presence of case, minimality effects do arise in wh-questions.

To recapitulate, in the present study we assessed how children comprehend subject and object dependencies in structures that involve wh-movement (wh-questions, RCs, FRs), in order to investigate whether Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) predictions regarding minimality effects can be confirmed in Greek, a language that displays overt/morphological case marking in its arguments and does not linearly show intervention in non-canonical sentences. Driven by the aforementioned findings, we argued that Friedmann et al.'s (Reference Friedmann, Belletti and Rizzi2009) RM approach offers an adequate account for children's performance to a great extent. More specifically, lexical NP restriction seems to have an impact on the comprehension of object dependencies in structures that involve wh-movement. However, lexical NP restriction per se cannot provide an account for the entire dataset we presented here. In fact, we showed that the internal structure (i.e. the feature specification) of the moved element and the intervener affect children's comprehension, which is in line with recent findings by Adani et al. (2010), Costa et al. (Reference Costa, Grillo and Lobo2012), and with the version of the RM approach developed in Grillo (Reference Grillo, Blaho, Schoorlemmer and Vicente2005, Reference Grillo2008, Reference Grillo2009). Thus, we conclude that although intervention effects emerge in the presence of lexical restriction, they also depend on the feature specification of the moved or the intervening element, as shown in the case of what questions and in the case of object RCs with a quantificational subject. Last, we argued that although case facilitates children's comprehension in languages that mark their arguments with overt/morphological case, genuine minimality effects still arise in non-canonical structures.