Scalar implicatures in Chinese children with reading difficulties

Abstract

This study investigates the derivation of scalar implicatures in Chinese children with reading difficulties (RD). Twenty-four children with RD (mean age 9 years and 8 months), 20 age-matched typical readers (mean age 9 years and 10 months), 20 six-year-old children and 20 five-year-old children were tested with the comprehension of sentences with scalar items yixie ‘some’ and suoyou ‘all’. The pattern of children with RD was similar to that of six-year-old children but differed from that of age-matched typical readers in the comprehension of the sentences containing yixie that were pragmatically underinformative in the context. Interestingly, many children with RD and younger children, who accepted the sentences containing yixie that were pragmatically underinformative, rejected the sentences containing yixie that were true in a context supporting the literal (semantic) interpretation. These results support the view that the computation of scalar implicatures may be impaired in children with RD, due to a complex interplay of factors such as (at least) the lexical knowledge of the scalar term and processing/pragmatic limitations.

Keywords

comprehension implicature poor readers pragmatic development

Growing evidence has shown that children with reading difficulties (RD) have significant impairments in oral language (Byrne, 1981; Cantiani, Lorusso, Perego, Molteni, & Guasti, 2015; Fiorin, 2010; Joanisse, Manis, Keating, & Seidenberg, 2000; Nation & Snowling, 2000; Scarborough, 1991; Waltzman & Cairns, 2000; Wilsenach, 2006; Wise, Sevcik, Morris, Lovett, & Wolf, 2007; among many others). In language comprehension, children with RD make more errors or take longer than age-matched typical readers in interpreting affirmative and negative sentences (Hu, Vender, Fiorin, & Delfitto, 2018; Vender & Delfitto, 2010), passive sentences (Leikin & Assayag-Bouskila, 2004; Reggiani, 2010) and relative clauses (Arosio, Panzeri, Molteni, Magazù, & Guasti, 2017; Bar-Shalom, Crain, & Shankweiler, 1993; Byrne, 1981; Casalis, Leuwers, & Hilton, 2012; Mann, Shankweiler, & Smith, 1984; Smith, Macaruso, Shankweiler, & Crain, 1989; Stein, Cairns, & Zurif, 1984).

In a number of recent studies, children with RD also showed a delay in the area of pragmatics (Cardillo, Garcia, Mammarella, & Cornoldi, 2018; Kasirer & Mashal, 2017; Lam & Ho, 2014; Vender, 2017). One of the most studied pragmatic phenomena is the derivation of a special kind of conversational inference that ensues from the use of scalar quantifiers like some. According to the standard Gricean analysis, these are pragmatic inferences that are derived by a hearer when a speaker utters a sentence in which a scalar term (like some) is used, as in (1). Given that a stronger linguistic alternative to some is available (all), and that the Gricean Maxim of Quantity exhorts cooperative speakers to be optimally informative (Grice, 1957, 1975, 1989), the fact that a speaker utters (1) instead of (2) induces the listener to believe that the stronger alternative (2) does not hold. This, in turn, results in the derivation of the scalar implicature (SI) in which the meaning of the weak quantifier some is strengthened, as in (3).

(1) John ate some of the cookies in the box.

(2) John ate all of the cookies in the box.

(3) John ate some but not all of the cookies in the box.

The derivation of SI in children with RD has been investigated in recent years, and different scalar items such as quantifiers, connectives and aspectual verbs have been tested (Arosio, Pagliarini, Perugini, Barbieri, & Guasti, 2016; Vender, 2009, 2017). In fact, different studies yielded contrasting results: Vender (2017) reported that Italian dyslexic children (M = 9;9 [9 years and 9 months], aged 8;7–11;5) were impaired in comparison to controls in the interpretation of scalar items belonging to different scales, including quantifiers (some vs. all, and most vs. all), frequency adverbs (sometimes vs. always, and often vs. always) and connectives (or vs. and); specifically, dyslexic children performed similarly to younger children, whereas age-matched controls performed as accurately as adults. Conversely, Arosio et al. (2016) observed that children with RD (M = 9;3, aged 7;5–12;3) at a comparable age to the group tested by Vender were at ceiling in comprehending the scalar quantifier some, although they tested a different scalar term (Italian qualche instead of Italian alcuni tested by Vender 2017).

The aim of the present study is to contribute further data to this debate, in order to deepen our understanding of how children with RD derive pragmatic inferences. Since previous research was restricted to the two studies mentioned above, both testing Italian (Arosio et al., 2016; Vender, 2009, 2017), we decided to investigate a typologically different language, i.e., Mandarin Chinese. In principle, the mechanism for derivation of SIs in Mandarin Chinese should not differ from that of Italian with respect to the semantic and pragmatic abilities involved in the computation. Interestingly, however, Katsos et al. (2016) compared five-year-old TD children’s comprehension of quantifiers in typologically different languages, including Mandarin Chinese and Italian, by means of the same task translated across languages. A different performance with respect to the comprehension of the quantifier ‘some’ was revealed in five-year-old Mandarin-speaking children as compared to Italian children of the same age.

By testing Mandarin Chinese, we aim to provide new insights into the mixed results obtained with Italian children with RD, by evaluating them from a broader cross-linguistic perspective. To this end, we assessed the ability to compute SIs in a group of children with RD who spoke Mandarin Chinese, and compared them with a group of age-matched typical readers and two groups of children at the earlier stages of language development.

The article is organized as follows. We first briefly introduce the phenomenon of SI computation as well as the previous acquisition studies concerning SI in typically developing (TD) children and children with RD. Then we present our experimental study and offer a general discussion.

Understanding scalar implicatures: theoretical and acquisition background

Children’s ability to derive scalar inferences has been hotly debated in recent years. This interest is partly due to the fact that this phenomenon can shed some light on the interaction between semantics and pragmatics in children (as well as in adults).

As outlined above, a SI is triggered by the use of an informationally weak term belonging to an informational scale, whereby asserting a lower-ranking alternative implicates that the higher-ranking one does not hold. This is by virtue of the fact that, if the most informative alternative held, not using it would lead to a violation of the Maxim of Quantity, according to which cooperative speakers make their contribution as informative as required. As we have seen, a typical trigger for SIs is the quantifier some, which is ordered in a scale with other elements like many, most, all. These elements are ordered in a scale based on their informational strength (Horn, 1972), and a proposition containing a logically stronger term entails a proposition containing a weaker one. In the given scale some<many<most<all, a sentence with the stronger term (e.g., all) entails a sentence with the weaker one (e.g., some); referring to the examples above, (2) entails (1). In other words, (1) is semantically compatible with (2), but (2) is more informative than (1), as it is true in a subset of situations in which (1) is true. As we said, (1) is generally interpreted as in (3), because the speaker is expected to abide by the conversational maxims: if she intended to express the richer proposition according to which John ate all of the cookies, she should have uttered the sentence in (2). Her choice to use (1) instead of (2) implies that the speaker’s intended meaning of (1), if the speaker has full knowledge of the facts, corresponds to the proposition expressed by (3).

At the theoretical level, the steps involved in the derivation of SIs require (i) the computation of the basic meaning of the scalar entry, (ii) the activation of the scalar alternatives and (iii) the negation of the most informative interpretation (if relevant to the purposes of the interchange), in accordance with the Maxim of Quantity. The question of whether and how these steps are also reflected at the processing level is controversial (cf. Chemla & Singh, 2014a, 2014b). On the one hand, there is evidence that scalar inferences are cognitively demanding even for adults in some tasks (e.g., Bott & Noveck, 2004; De Neys & Schaeken, 2007), and that these additional steps in the derivation reflect in a delayed processing in online tasks (Huang & Snedeker, 2009). On the other hand, some authors have argued that these steps are carried out during the standard semantic derivation of the meaning of a sentence, at least in contexts that support the SI derivation, thus not necessarily requiring additional (costly) steps that reflect in a delayed processing of the inference (e.g., Breheny, Ferguson, & Katsos, 2013; Foppolo & Marelli, 2017).

The question whether the steps involved in the derivation of a SI are cognitively demanding in terms of processing resources is crucial in acquisition, even more so when facing atypical populations. One way to understand the underlying process that leads to the ‘some but not all’ interpretation is exemplified in Foppolo, Guasti, and Chierchia (2012), considering the quantifier some as an example. (i) At the first step, the basic meaning of some as an existential quantifier is accessed in the lexicon. (ii) At the second step, all is activated upon hearing some and the hearer acknowledges the fact that the speaker could have used all, while recognizing the fact that the use of some in a circumstance in which all applies would be underinformative, thus violating the Maxim of Quantity. (iii) From this, the hearer is entitled to infer that all does not apply, generating the SI according to which some is interpreted as some but not all. Although experimental studies have provided some evidence for these processing steps, the precise role of each step in the derivation of SIs remains open. In particular, the way in which alternatives are activated and recognized as relevant is still debated (for recent overviews, see Papafragou & Skordos, 2016; Skordos & Papafragou, 2016).

A variety of studies in the acquisition literature have documented that preschool children do not compute SIs as adeptly as adults (e.g., Foppolo et al., 2012; Katsos & Bishop, 2011; Papafragou & Musolino, 2003), while school-age children behave in an adult-like way (e.g., Doitchinov, 2005; Guasti et al., 2005). For instance, TD Italian-speaking children at age 5 have no problems in the comprehension of the stronger quantifier tutti ‘all’, as in Tutti i puffi sono andati in barca ‘All the Smurfs went on a boat’, while they fail to reject underinformative qualche ‘some’ statements, e.g., Qualche puffo è andato in barca ‘Some Smurfs went on a boat’, in a situation in which all the Smurfs went on a boat (Foppolo et al., 2012). By contrast, Italian-speaking children at age 7 have been found to be capable of behaving like adults in rejecting underinformative qualche ‘some’ statements in the situation in which all the Smurfs went on a boat (Guasti et al., 2005). Different hypotheses have been formulated to account for this state of affairs, and the debate also extends to theoretical accounts of SIs (cf. Chemla & Singh, 2014a, 2014b for an overview).

One account, which we might dub the processing account, starts from the assumption that the derivation of SIs is cognitively demanding, thus exceeding the children’s limited resources (Pouscoulous, Noveck, Politzer, & Bastide, 2007). Other accounts, instead, explain children’s difficulties by appealing to their immature pragmatic system; under these approaches, that we might dub pragmatic, children show a delayed ability with scalar inferences either because they are pragmatically more tolerant than adults (Katsos & Bishop, 2011), or because they fail to recognize the contextual relevance of alternatives (Skordos & Papafragou, 2016). A third account, that we might call the lexicalist account, locates children’s difficulty with the derivation of scalar inferences in their yet immature lexicon; according to these accounts, children fail to access or retrieve the lexical scalar alternatives, since the scale is not represented in the lexicon, or the scalar elements are not stably associated with one another in a scale yet (Barner, Brooks, & Bale, 2011; Foppolo et al., 2012; Horowitz, Schneider, & Frank, 2018; Tieu, Romoli, Zhou, & Crain, 2016).

To date, while a large number of empirical studies converge on the fact that TD children have difficulties with scalar inferences, a consensus has not been reached with respect to the source of this delay. This uncertainty is compounded by the fact that experimental findings from atypical populations, such as individuals with autism spectrum disorders (ASD) and specific language impairment (SLI), are fairly mixed, so it is unclear how these findings might contribute to the theoretical debate (Arosio et al., 2016; Hochstein, Bale, & Barner, 2018). As far as children with RD are concerned, the literature is sparse and the results are varied. However, we think that testing children with RD is a worthwhile enterprise, as it could help disentangle the different approaches to SIs, providing evidence for one of the theoretical alternatives reviewed above. The next section is devoted to introduce the sparse literature on the computation of SIs in children with RD.

The computation of scalar implicatures in children with reading difficulties

As anticipated in the introduction, reading disorders are only the major and most evident manifestation of dyslexia: dyslexic individuals, indeed, can suffer from deficits affecting both their linguistic competence and their processing abilities (McLoughlin, Leather, & Stringer, 2002; Nicolson & Fawcett, 2008; Vender, 2017).

There is a lack of in-depth studies regarding the acquisition of SIs in children with RD, despite the fact that pragmatics is an interesting test case for defining the areas in which individuals with RD might lag behind their TD peers, and for rehabilitation purposes. Only two recent studies have assessed it in Italian children with RD (Arosio et al., 2016; Vender, 2017). The main goal of both studies was to explore whether children with RD exhibited difficulties with SIs.

Vender (2017) tested 20 Italian dyslexic children (M = 9;9), comparing them with 20 age-matched children (M = 9;8) and 16 younger children (M = 6;9), by using sentences containing the quantifier alcuni ‘some’, i.e., sentences such as Alcuni bambini hanno ricevuto le caramelle ‘Some children received the candies’. A truth value judgment task was used, with four items per condition. The results showed that 54% of the dyslexic children and 35% of the younger TD children correctly rejected the some-underinformative statements in which all the children received the candies, while their age-matched control children did so 97% of the time. The author explained these results by arguing that the computation of SIs requires the construction and comparison of two alternative descriptions of the sentence (respectively with the existential and the universal quantifier), an operation which is remarkably demanding in terms of cognitive costs and that exceeds the limited processing capacities of younger and dyslexic children. However, the study only reported the group’s average responses, without reporting each individual’s performance. Thus, the issue whether the difficulties of SIs can be generalized to all the children with RD, or are only restricted to some individuals, deserves further attention.

These results contrasted with the findings reported by Arosio et al., (2016). In that study, 24 children with RD (M = 9;3), 24 age-matched TD children (M = 9;3) and 24 vocabulary-matched TD children (M = 8;0) were tested with intransitive sentences containing qualche ‘some’, such as Qualche mela è nelle scatole ‘Some apples are in the boxes’, and tutti ‘all’, such as Tutte le mele sono nelle scatole ‘All the apples are in the boxes’. They used a shortened version of the Cavegirl and Boxes task, namely, a binary judgment task as in Katsos, Roqueta, Estevan, and Cummins’s (2011) paper. Each condition included three items. In the task, each sentence was combined with a visual display including a certain number of boxes that contained the mentioned object(s) (e.g., apples). Accuracy rates of all the groups were above 92% across conditions. This result shows that children with RD had no problem with the comprehension of the logical meaning of all and some, and that, crucially for the present purposes, they were able to recognize the infelicity of some-underinformative statements, at the same rate of school-aged TD children.

To summarize, contrasting results are reported in understanding the scalar quantifiers some by children with RD: severe deficits in Vender (2017) vs. at ceiling performance in Arosio et al. (2016). However, this contrast might be due to the different classes of existential quantifiers investigated and to important differences in the experimental designs used in the two studies.

First of all, the two studies used different Italian quantifiers (alcuni in Vender, 2017 and qualche in Arosio et al., 2016), a factor that could have influenced children’s performance. The fact that children are affected by the choice of different types of scalar expressions has been reported in previous studies across languages (e.g., Pouscoulous et al., 2007 for French children’s production of French existential quantifiers quelques vs. certains ‘some’). In formal semantics, existential quantifiers such as qualche/quelques can be differentiated from existential quantifiers such as alcuni/certains. The difference between these two classes of quantifiers has often been interpreted as related to their capacity of ‘truly’ introducing discourse-referents in discourse, and it has been conveniently formalized in Partee (1988) and especially in the framework of Discourse Representation Theory (DRT; Kamp & Reyle, 1993), in terms of the distinction between ‘proportional’ quantifiers on one side, and ‘cardinality’ quantifiers on the other side. Proportional quantifiers, such as ogni ‘every’ in Italian, express a specific relation between the two sets corresponding to the nominal restriction of the quantifier and to the predicative restriction. For instance, in a sentence such as Every Italian is blonde, the quantifier every encodes a specific relation between set A (the set of the Italians) and set B (the set of blonde people). Crucially, no independent set is introduced here as a discourse-referent, since A and B are bound by the quantifier. Moreover, there is no way to interpret every (i.e., the quantifier) as expressing an intersective property of A and B: what every expresses is simply that A is included in B, without reference to any independent property of the A’s that are B’s. The same holds for a quantifier like most. What matters to us here is that existential quantifiers such as qualche/quelques can also be interpreted as proportional quantifiers. A sentence such as Some Italians are blonde is accordingly interpreted as simply expressing a relation between A (the set of Italians) and B (the set of blonde people). However, notice that with existential quantifiers, the property expressed by the quantifier could also be expressed as a ‘cardinal’ property of the A’s that are B’s: the set of blonde Italians has at least cardinality one. In this way, a discourse-referent is plausibly introduced (as the intersection of A and B) and the quantifier can be viewed as a cardinal property predicated on this referent (i.e. the class of blonde Italians contains at least one individual). Quantifiers that exhibit this behavior are ‘cardinality’ quantifiers. As discussed in Kamp and Reyle (1993, pp. 452–461), quantifiers such as many sometimes behave as a proportional and sometimes as a cardinality quantifier. In some cases, whether the sentence is true seems to depend only on the nominal restriction of many: the sentence Many inhabitants of this town are Mexican is more promptly judged as true than the sentence Many Mexicans live in this town. This suggests that many behaves proportionally, since it is sensitive only to A and not to the intersection between A and B. However, in a context where 10% of the houses in our street are insured and exactly those houses burned down, it is certainly less acceptable to state that Many houses in our street were insured than stating that Many houses in our street burned down. In this case, whether the use of many is felicitous seems to depend on the intersection between A and B.

On these grounds, one might ask whether a given existential quantifier, in a particular language, is interpreted as a proportional or as a cardinality quantifier. When an existential quantifier Q is cardinal, a discourse-referent is introduced (corresponding to the set of the A’s that are B’s) and the cardinal property expressed by Q is then predicated on this discourse-referent. From this perspective, as already mentioned above, the existential quantifier qualche plausibly qualifies as proportional in Italian: given the sentence Qualche ospite è arrivato ‘Some guests arrived’, pronominal anaphora remains marginal both with a singular and a plural pronoun (??È nell’atrio ‘He is in the hall’ and ??Sono nell’atrio ‘They are in the hall’), suggesting that there is no set counting as an independent discourse-referent. Conversely, the existential quantifier alcuni is arguably cardinal. It promptly warrants pronominal anaphora (Alcuni ospiti sono arrivati. Sono nell’atrio. ‘Some guests arrived. They are in the hall.’) and its felicity seems to be sensitive to the properties of the intersection between A and B. Take for instance a situation in which 20% of the Mexican population is both blonde and sick: it is plausibly more felicitous to utter the sentence Alcuni messicani sono biondi ‘Some Mexicans are blonde’ than to utter the sentence Alcuni messicani sono malati ‘Some Mexicans are sick’ (in the latter case, it is certainly more accurate to say that many Mexicans are sick). Alcuni qualifies thus arguably as a ‘cardinality’ quantifier.

Now, it is quite reasonable to argue that cardinality quantifiers might involve more complex processing than proportional quantifiers. For instance, while calculating the relevant SI associated with the last sentence, an Italian speaker needs: (i) to introduce the set of sick Mexicans as a discourse-referent, (ii) to be endowed with the piece of lexical knowledge according to which alcuni expresses a well-defined cardinal property of this set (i.e., the set has at least cardinality one), and (iii) to exclude the informationally richer alternatives. Conversely, an Italian speaker who calculates the SI with the proportional quantifier qualche only needs to know which relation is expressed by qualche and to compare it with informationally richer relations along the relevant scale (many, most, all). Clearly, it might well be the case that learning these two classes of quantifiers is not a unique undifferentiated process, and that this might contribute to explain the children’s different behavior with qualche (a proportional quantifier) and alcuni (a cardinality quantifier), both when they have to assess the semantic value of a sentence containing an existential quantifier and when they have to derive its pragmatic meaning by calculating a SI. In fact, we propose that the difference between proportional existential quantifiers and cardinality existential quantifiers is what explains the different results obtained by Arosio et al. (2016) and Vender (2017) with, respectively, qualche and alcuni. Similarly, we will argue below that the different results obtained with different classes of existential quantifiers in Mandarin Chinese are arguably related to the very same difference.

Of course, we should also emphasize that the two studies used verbs of different categories, i.e., transitive verbs in Vender (2017) and intransitive verbs in Arosio et al. (2016), which might lead to different experimental demands. We know from previous studies that children’s performance with SIs varies tremendously across different experimental tasks, and that children are sensitive to training and task manipulations (Foppolo et al., 2012; Guasti et al., 2005; Katsos & Bishop, 2011; Noveck, 2000; Papafragou & Musolino, 2003; among many others). Even in adults, there is considerable variation across scales among the rates at which scalar expressions give rise to upper-bounding inferences (van Tiel, van Miltenburg, Zevakhina, & Geurts, 2016).

As we said, testing a language typologically different from Italian is crucial for an independent assessment of the competence with the pragmatic meanings of scalar terms in children with RD and, eventually, to better explain the conflicting results obtained in Italian. All the more so when considering that, first of all, Mandarin Chinese displays two variants of ‘some’, as briefly outlined in the next section and, secondly, Mandarin-speaking children showed a non-adult-like comprehension of this scalar term even when no inference was required, as attested by the cross-linguistic investigation by Katsos et al. (2016). In this way, we also hope to be able to tease apart children’s knowledge and understanding of scalar quantifiers from the processing abilities required to derive a scalar inference.

‘Some’ in Mandarin Chinese

Our study was conducted in Mandarin Chinese. In this language, there are two scalar items which are roughly equivalent in meaning to English some: yixie and youxie, as exemplified in (4–5).

(4) Yixie ren na le pingguo.

one-some person take ASP apple

Some people took an apple.

(5) Youxie ren na le pingguo.

have-some person take ASP apple

Interpretation 1: Some people took an apple.

Interpretation 2: There were some people who took an apple.

Although yixie and youxie are often treated equally as English some (Politzer-Ahles, Fiorentino, Jiang, & Zhou, 2013; Tsai, 2004; Wu & Tan, 2009; Zhao, 2012), syntactically and semantically they are not equivalent to each other. According to Tsai (2003, 2004), you ‘have’ can be analyzed as a determiner, occupying a D position, while yi ‘one’ is a numeral, occupying a Num position (from Tsai 2003, p. 5). The tree diagrams of youyixie, youxie and yixie are illustrated in (6a, b, c), respectively.

Youxie is considered to be equal to youyixie when the numeral yi ‘one’ is omitted. On these morphosyntactic grounds, it is reasonable to propose that yixie (whereby a numeral composes with a noun) qualifies as a cardinality quantifier (such as alcuni in Italian), whereas youxie (where only the determiner-like element is overtly realized) qualifies as a proportional quantifier (such as qualche in Italian). In a nutshell, we propose that (5) is interpreted as true whenever a certain relation holds between the two relevant sets (the set of people and the set of those who took an apple), whereas (4) is interpreted as true whenever the discourse-referent that is introduced (the set of people who took an apple) satisfies the cardinality property expressed by yixie (this set contains at least one element).

Regarding the syntax/semantics interface, there are certainly other properties that should be explored. For instance, it is generally cardinality quantifiers that give rise to there-be constructions. However, in Mandarin Chinese it is the proportional quantifier youxie that gives rise to a construction roughly equivalent to an English there-sentence, as illustrated in Interpretation 2 of (5). Here, we will put these complications aside, especially because they are plausibly related to a set of distinct syntactic properties of the Chinese existential quantifiers that are not shared by English. For instance, as far as their syntactic positions are concerned, an asymmetry occurs with youxie, but not with yixie. That is, youxie can only occur in the subject position, whereas yixie can appear either in the subject position or in the object position, as in (7).

(7) Zhang San na le yixie/*youxie pingguo.

Zhang San take ASP some apple

Zhang San took some apples.

In particular, what we intend to emphasize here is that, given these syntactic and semantic differences between the two scalar items youxie and yixie, we might expect children to exhibit different patterns when acquiring the two scalar items, on the model of what we proposed for Italian above. The question is whether this prediction is borne out or not.

As far as we are aware, some experimental studies on the acquisition of youxie ‘some’ have been carried out, while the acquisition of yixie ‘some’ has not yet been explored. Su and Su (2015) tested the comprehension of sentences containing youxie ‘some’ with a group of TD children (N = 16, M = 6;6, SD = 1.6) and a group of young children with ASD (N = 14, M = 6;6, SD = 1.6). The data showed that the TD children accepted logical interpretations (i.e., they accepted the sentences in the some-underinformative condition) 21% of the time, and that the children with ASD did so 36% of the time, but no significant difference was found between the two groups. Regarding the some-true and the some-false conditions, both groups provided high proportions of correct judgments (above 93%). The results confirmed the findings from the aforementioned cross-linguistic studies: namely, some young children have difficulties in deriving the pragmatic inference of some. However, even young children appear to perfectly master the semantic meaning of the existential quantifier, as revealed by their virtually ceiling performance in the some-true and some-false conditions, where no inferential process was involved.

In addition, Katsos et al. (2016) tested the comprehension of sentences containing youyixie ‘some’ by TD Mandarin-speaking children (N = 34, M = 5;5, aged 5;0–5;11), using the Cavegirl and Boxes task (Katsos et al., 2011). This existential quantifier is morphologically more complex than youxie, since it overtly features not only the determiner you but also the numeral element yi. We propose that this morphological complexity is mirrored by semantics; in fact, the presence of the numeral yi suggests that the interpretation of youyixie involves a process of intersective reduction, at the end of which the cardinality quantifier is interpreted as the property of a free discourse-referent. In other words, youyixie is a cardinality quantifier, on a par with yixie. The data provided by Katsos et al. (2016) showed that the children provided target responses in the some-underinformative condition (i.e., they rejected underinformative statements) 33.8% of the time, in the some-true condition 63.7% of the time, and in the some-false condition 77.5% of the time. Apart from children’s behavior in the some-underinformative condition, where children performed far worse than with youxie, these results are also interesting in another respect: namely, Mandarin-speaking children made lots of errors in the some-true and some-false conditions. This finding was robust in Mandarin Chinese, but it is at odds with other findings regarding Indo-European languages in which children’s performance on some-true and some-false statements is typically at ceiling. However, these results are arguably no longer so puzzling if one considers not only that youyixie, as a cardinality quantifier, should be generally harder to master, as suggested by our discussion above, but also that the degree of morphological complexity of this quantifier has been taken to correspond to an interpretive process according to which the proportional interpretation of the quantifier gets reinterpreted in terms of the cardinal property of a discourse-referent. In these conditions, we predict in fact that it is not only the pragmatic reading that becomes more complex, but that lexical knowledge itself involves a complex semantic process, which consists in identifying an intersective set as the discourse referent and in predicating a specific cardinality property (i.e., the set contains at least one element) of this discourse-referent. This is exactly the kind of difficulty that emerges from Katsos et al.’s data. As we will see, these results are fully confirmed by those of our experiment, to be discussed below.

Given these previous studies on Mandarin-speaking children, in our study we decided to investigate yixie ‘some’. Based on the discussion above, and on the fact that yixie only features a numeral in addition to the classifier, we think that it is reasonable to propose that yixie is a cardinality quantifier.

The current study

In the present study, we investigate the comprehension of sentences containing yixie ‘some’. As shown above, cross-linguistic studies have shown that children seem to respond differently to different scalar items, and it would be interesting to see whether Chinese children also have difficulties in deriving the pragmatic inference of yixie ‘some’. Previous studies on adults’ processing reported that the interpretation of yixie ‘some’ is costly and effortful (Chen & Guo, 2012; Zhao, 2012). For instance, Zhao (2012) reported that adults took longer to interpret sentences such as Yixie nühai you bianzi ‘Some girls have their hair in braids’ pragmatically, than to interpret them logically (cf. also Bott & Noveck, 2004 for related findings in French). Given this result, we might expect children to display some difficulty in interpreting yixie ‘some’ pragmatically.

In addition, previous results on youyixie ‘some’ showed, as discussed above, that young children made many errors in the some-true and some-false conditions, a situation not attested in many other languages (Katsos et al., 2016). It would be interesting to see whether Chinese children also make errors in the some-true and some-false conditions when comprehending sentences containing yixie ‘some’.

All in all, the present study has a threefold objective. First, we aim to investigate the comprehension of the scalar quantifiers yixie ‘some’ and suoyou ‘all’ in Chinese children with RD, and to test their ability to interpret yixie ‘some’ pragmatically. This would allow us to assess the children’s knowledge of the scalar quantifiers in the first place, and their competence with SIs. Second, our participants included two groups of younger children in order to investigate how children develop their knowledge of the scalar quantifiers involved in the scale over time, as well as their ability to derive SIs, by looking at their errors in different conditions. Third, we attempt to explore whether there are similarities between children with RD and children at the earlier stages of language development: on the basis of the literature reviewed above (Vender, 2017), we predict that children with RD exhibit significant limitations in the interpretation of sentences requiring complex processing and/or pragmatic analysis, performing similarly to younger TD children.

Method

Participants

Eighty-four children participated in the experiment, including a group of 24 poor readers, a group of 20 typical readers of the same chronological age, a group of 20 six-year-old children and a group of 20 five-year-old children. They were recruited in Zhejiang and Fujian, China, and the medium of instruction at schools was Mandarin Chinese. In addition, we tested 20 adult controls aged 19;0–21;11, all native speakers of Mandarin Chinese.

The reading skills of children in the poor reader and typical reader groups were assessed using a literacy test which has been used in prior research (Hu et al., 2018). Children were placed in the poor reader group if their scores on the literacy test were at least 1.5 SD below the average mean for their grade, and if they had difficulties in reading and writing Chinese according to daily observations in their teachers’ reports. They were also assessed on the Chinese version of the combined Raven’s Progressive Matrices test (Zhang & Wang, 1985), obtaining a standardized score equal to or above 80; two poor readers were excluded from the initial sample of 24 participants, due to a score below 80. Moreover, they were tested for phonological awareness and morphological awareness, which have been claimed to be essential for learning Chinese (McBride & Wang, 2015). Phonological awareness was assessed with onset detection, rhyme detection and tone detection tests, and morphological awareness was assessed with homophone awareness and homograph awareness tests (Hu et al., 2018). A summary of the children’s descriptive characteristics is presented in Table 1. One-way ANOVAs revealed no differences between the poor readers and the typical readers regarding age, F(1, 40) = .33, p = .57, and the Raven scores, F(1, 40) = 2.08, p = .16, while one-way ANOVAs revealed significant differences between two groups in literacy, F(1, 40) = 21.64, p < .001, phonological awareness F(1, 40) = 10.67, p < .01, and morphological awareness, F(1, 40) = 12.04, p < .01.

Table 1.

Participant characteristics.

	Poor readers	Typical readers	TD age 6	TD age 5
Number	22	20	20	20
Gender/boys	17 boys	9 boys	10 boys	12 boys
Age/years	9;8 (1.16)	9;10 (1.03)	6;5 (0.31)	5;6 (0.24)
Raven IQ	100.54 (11.88)	105.70 (11.21)	–	–
Literacy	0.65 (0.16)	0.83 (0.07)	–	–
Phonological awareness	0.58 (0.22)	0.78 (0.17)	–	–
Morphological awareness	0.56 (0.19)	0.73 (0.13)	–	–

Note. For age, the numbers are mean years for each group, e.g., 9;8 means 9 years and 8 months; for the Raven IQ, the numbers are mean of standardized scores for each group; for literacy, phonological awareness and morphological awareness, the numbers are mean of accuracy for each group.

None of the children tested had any report of brain damage, sensory impairments, or serious emotional or behavioral problems. Their vision was normal or corrected to normal.

Materials

The experiment involved five conditions: sentences containing yixie ‘some’ that were true, false, or pragmatically underinformative in a context; sentences containing suoyou ‘all’ that were logically true or false within a context. Examples of yixie ‘some’ and suoyou ‘all’ are given in (8–9).

(8) Yixie nühai na le pingguo.

some girl take ASP apple

Some girls took an apple.

(9) Suoyou nühai na le pingguo.

all girl take ASP apple

All girls took an apple.

The test comprised 20 items, eight of which were in the some-underinformative condition. In addition, there were four warm-up items. Each item was combined with the different types of object in order to avoid any effects from previous mention of a quantifier and object combination. A full list of experimental items is given in the Appendix 1.

The task administered was a PowerPoint version of the truth value judgment task (Crain & Thornton, 1998). Similar to the prior work by Su and Su (2015) and Vender (2017), the stories were presented by one experimenter, using Microsoft PowerPoint on a laptop. Before the experiment started, the experimenter introduced a group of characters to participants, including a puppet (a small bear), five girls and five boys. The participants were invited to listen to stories and look at a number of visual displays showing the characters performing some actions. Here is an example of how the experimenter introduced the story: ‘In this scenario, there are five girls and five apples. The first girl took an apple. The second girl took an apple. The third girl took an apple. The fourth girl took an apple. The fifth girl took an apple’ (for a sample of the visual stimuli, see Figure A1 in Appendix 2). At the end of the action, the small bear gave a statement on what he saw and then the participants had to judge the accuracy of the puppet’s description. In the example provided, the small bear could describe the situation as ‘All/Some girls took an apple’. This sentence could be uttered in different situations, as exemplified in Figure 1, that made it true, false or, in the case of some, underinformative. If participants rejected a statement, they were invited to explain why they judged it wrong. This question was asked to check whether they rejected the statement for reasons unrelated to falsity or informativeness (Katsos et al., 2011).

Figure 1.

Examples of final outcomes displayed on screen in each condition.

Procedure

All the children were tested in their school during school-time, and the tasks were administered in three sessions: in the first two sessions, which lasted roughly 20–30 minutes each, we administered the preliminary measures reported in the Participants section. In the third session, which was about 15–30 minutes, we administered the task with SIs.

Results

The percentages of target responses in each condition for each age group are provided in Table 2. It is evident that there is some variation across the groups. Specifically, the poor readers rejected the sentences in the some-underinformative condition 73% of the time, while the age-matched typical readers did so 100% of the time, as expected given their age; moreover, the six-year-old children and the five-year-old children rejected them 55% and 30% of the time, respectively. With respect to the some-true condition, the poor readers, the six-year-old children and the five-year-old children only accepted the truth-conditionally correct statements 79%, 60% and 40% of the time, respectively, while typical readers never rejected them, as expected. On some-false, all-true and all-false conditions, all participants responded correctly above 95% of the time. Adults performed at ceiling in all the conditions. Thus, we did not include adults’ data for further analyses.

Table 2.

Percentage (%) of target responses across conditions and groups, with standard deviation in parentheses.

	Poor readers	Typical readers	TD age 6	TD age 5
some-underinformative	73 (46)	100 (0)	55 (51)	30 (47)
some-true	79 (41)	100 (0)	60 (50)	40 (50)
some-false	100 (0)	100 (0)	100 (0)	98 (7)
all-true	100 (0)	100 (0)	100 (0)	100 (0)
all-false	100 (0)	100 (0)	98 (7)	95 (22)

To better understand the pattern of children’s responses, we further inspected children’s responses in the two conditions in which they showed a more variable performance. With respect to the some-underinformative condition, one first observation is that children were consistent in their strategy and were bimodally distributed, as already observed in previous studies (Guasti et al., 2005): 27% (6 out of 22) poor readers, 45% (9 out of 20) six-year-old children and 65% (13 out of 20) five-year-old children accepted all the critical underinformative statements, therefore always avoiding the computation of the implicature. Only one five-year-old child showed an inconsistent behavior, accepting 3 of the 8 some-underinformative items, and rejecting the others. By contrast, none of the age-matched typical readers accepted the underinformative statements (Figure 2). Regarding children’s rejection of some-underinformative statements, they were asked to provide an appropriate correction: all children responded appropriately by invoking the strong quantifier all, e.g., by saying that since all girls took an apple, ‘some girls took an apple’ was a bad description of the story, indicating that they were rejecting the underinformative statements for the correct reason. With respect to the some-true condition, 23% (5 out of 22) poor readers, 40% (8 out of 20) six-year-old children and 60% (12 out of 20) five-year-old children consistently rejected the statements (at least 2 out of 3 times), while all the typical readers consistently accepted it.

Figure 2.

Children’s target (pragmatic) responses in the some-underinformative condition across groups.

Following Skordos and Papafragou (2016), we categorized participants according to their performance on each trial type as either Passers (if they scored 0.75 or greater), or Failers (if they scored 0.50 or less). Children that provided inconsistent answers were excluded (this was the case of one five-year-old child). Indeed, children were highly consistent and almost all provided the same type of response across items of the same condition. We thus conducted nonparametric statistics on the data, adopting the Bonferroni correction to adjust p-values for multiple comparisons when appropriate.

Two separate Fisher’s Exact Tests on 2 × 4 contingency tables revealed a significant difference between the number of children that consistently rejected the some-underinformative statements and accepted the some-true statements in the four groups (both ps < .001).

We further explored these results by running separate Fisher’s Exact Tests on 2 × 2 contingency tables comparing the number of children who consistently provided a correct or pragmatic response (i.e., accepting the some-true statements and rejecting the some-underinformative statements) with the number of children that consistently failed to do so across groups. We found a statistically significant difference between the poor and the typical readers in the rate of derivation of SI (p = .0432), but not in the rate of correct responses in the some-true condition (p = .217); no significant difference was revealed between the poor readers and the six-year-old children in both conditions (both ps < .01); significant differences were observed between the poor readers and the five-year-old children in both conditions (both ps < .05). By comparing the six-year-old children with the other groups, a significant difference was only revealed with the typical readers in both conditions (both ps < .01), but not between the six-year-old and the five-year-old children (both ps > .05).

The results in the some-underinformative condition reflect children’s difficulty with the derivation of SI at younger ages, as already attested in previous studies (e.g. Foppolo et al., 2012); also, it shows that the poor readers, albeit older, still accepted logical interpretations, patterning like the six-year-old children and, crucially, differently from their TD peers.

The finding about children’s rejection of some-true statements, albeit surprising if compared with the findings in other languages, is in line with what was previously found by Katsos et al. (2016) on Mandarin Chinese. One possibility to explain this result, already discussed in the literature, is that children expected a numeral to be used in this case (Degen & Tanenhaus, 2016; Foppolo et al., 2012), and this might have offered grounds for rejection. However, we can rule out this possibility, based on children’s justifications. Interestingly, they always justified their rejection by saying that ‘some girls took an apple’ was a bad description of the story because some girls included all five girls, and in the relevant scenario only three girls took an apple, while the other two girls did not take an apple. What is going on is thus not simply the fact that children expected ‘three’ instead of ‘some’. As their justifications clearly indicate, children seem to be able to identify the discourse-referent in the relevant scenario, i.e., the set of girls who took an apple (corresponding to A∩B), but are arguably not able to apply the correct cardinality property to this set (the set must contain at least one element). It is as if they identify the discourse-referent (the girls who took an apple) and are led to interpret it universally (as the obvious default option), since they cannot calculate the predicative limitation that comes with it. This is exactly the kind of difficulty that we may expect under the hypothesis that a sentence containing a cardinality existential quantifier is true if and only if the discourse-referent corresponding to the intersective set (the girls who took the apple) satisfies the cardinal property of containing at least one element. It is also important to bear in mind that this kind of error was found only in younger children, and that it was not so common in the poor readers, who correctly interpreted the quantifier in the majority of the cases (79%).

As pointed out by one of the reviewers, another possible explanation for this behavior is more methodological and pertains to the material presented in this condition: in the relevant scenario, three girls took an apple and the other remaining two girls did not take anything. This might have created an (unfulfilled) expectation that the other girls were taking, for example, another kind of fruit, generating some possible confusion in some of the poor readers and the younger children (cf. also Crain & Thornton, 1998 for a discussion about the felicity conditions in the truth value judgment task).

To further inquire into the responses of younger and RD children, we checked whether there is an overlap between children who accepted statements in the some-underinformative condition and those who rejected statements in the some-true condition. Indeed, 4 out of the 5 poor readers, 7 out of the 8 six-year-old children and 11 out of the 12 five-year-old children that rejected some-true actually accepted some-underinformative as well, thus signaling a strong relation between the knowledge of the basic meaning of the scalar term, and the derivation of its strengthened meaning. What this suggests is that one of the main sources of difficulty in the some-underinformative condition is that children cannot activate the some/all scale with cardinality quantifiers, since children are still not able to master the piece of lexical/semantic knowledge required to interpret the cardinality quantifier as a specific property of the intersective discourse-referent.

In order to better understand children’s performance on the some-underinformative condition, we analyzed the data by excluding those children who consistently rejected the statements in the some-true condition; namely, we used the some-true condition as an exclusion criterion. The remaining children included 17 poor readers, 12 six-year-old children, 7 five-year-old children and 20 typical readers. As shown in Table 3, the remaining group of poor readers, the six-year-old children and the five-year-old children rejected the sentences in the some-underinformative condition 94%, 83% and 71% of the time, respectively.

Table 3.

Percentage (%) of target responses across conditions and groups, with exclusion of children who did not respond to the some-true condition correctly, with standard deviation in parentheses.

	Poor readers	Typical readers	TD age 6	TD age 5
some-underinformative	94 (24)	100 (0)	83 (39)	71 (46)
some-true	100 (0)	100 (0)	100 (0)	100 (0)
some-false	100 (0)	100 (0)	100 (0)	100 (0)
all-true	100 (0)	100 (0)	100 (0)	100 (0)
all-false	100 (0)	100 (0)	100 (0)	88 (33)

A Fisher’s Exact Test on a 2 × 4 contingency table revealed a significant difference between the number of children that consistently rejected the some-underinformative statements in the four groups (p = .0383). This effect was further explored by running separate Fisher’s Exact Tests on 2 × 2 contingency tables comparing each group to each other. By adopting the above-mentioned exclusion criterion, no difference was revealed in any of the comparisons (all ps > .05), except for a marginal difference between the five-year-old children and the typical readers (p = .0598).

To sum up, there were three main findings. First, the pattern of the poor readers was similar to that of the six-year-old children but differed from that of age-matched typical readers in the some-underinformative condition when all children were included in the pairwise comparisons. Second, some poor readers and some younger children did not accept the sentences in the some-true condition. By excluding these children, the pattern of the poor readers did not differ from that of the age-matched typical readers in the some-underinformative condition, suggesting that the delay observed in the previous comparison might be (also) due to some difficulty with the lexical knowledge of the scalar term involved in the computation. Third, there was an improvement from age 5 to age 6 both in the some-underinformative and the some-true conditions, which were found to be particularly problematic for the younger children. Interestingly, there was a significant overlap between the some-underinformative and the some-true conditions in the poor readers, the six-year-old and the five-year-old groups: many children who accepted ‘some’ in the some-underinformative condition also rejected ‘some’ in the some-true condition.

Discussion

Our study shows that some children with RD exhibit deficits in the comprehension of the scalar quantifier yixie ‘some’ and in the derivation of the SI associated with it. In fact, as shown by the individual analysis, 6 out of 22 children with RD consistently accepted the critical underinformative statements, while none of the typical readers of the same age accepted them. This indicates that (at least some) children with RD have problems with some of the steps involved in the calculation of SIs. These results are compatible with Vender’s (2017) findings, since they provide a confirmation that some poor readers exhibit serious deficits in the computation of SIs. Our results are also consistent with the findings from previous studies on adults with dyslexia (Griffiths, 2007), demonstrating a possible correlation between dyslexia and non-pragmatic readings of scalar terms.

Although it is not surprising that, overall, children with RD perform differently from their TD peers, we want to focus on two interesting patterns observed in our data.

First, by comparing poor readers with TD children at an age that has been found to be crucial for pragmatic development, namely age 5 and 6, we found that the children with RD pattern alike to the TD children at age 6 and perform better than the younger children. Second, we found a systematic correspondence between the failure to accept some in a condition in which it was appropriate to describe a subset situation, and the acceptance of some in a condition in which it was infelicitously used to describe a superset situation. All the children, however, correctly rejected some when it was used to describe a situation in which the set was empty (i.e., in some-false condition). We think that these results, all together, are quite interesting in light of the different hypotheses proposed to explain children’s failure. As outlined in the introduction, one possibility is that the children that fail with SIs might not have acquired the meaning of some yet, and that this fact might lead to problems in retrieving the lexical alternative all within a lexicalized scale, a proposal compatible with lexicalist accounts (Barner et al., 2011; Foppolo et al., 2012). As we have seen, this hypothesis can be fleshed out in an interesting and precise way. When the existential quantifier is a cardinality quantifier, it can be argued that what is at stake semantically is not simply the relation between two sets; rather, the process of semantic interpretation involves two distinct phases: the first consisting in the identification of an intersective discourse-referent and the second consisting in the attribution of the cardinal property expressed by the quantifier to this discourse-referent. If children, at a certain phase of development, get stuck after the first phase, we predict in fact that (i) they will resort to the universal reading of the discourse-referent as a default option, and (ii) they will not be able to activate the some/all scale in the computation of the SI. Arguably, this is exactly what we found: a subset of our Chinese children interpreted yixie ‘some’ as all and had difficulty calculating the pragmatic reading in the some-underinformative condition.

Of course, this kind of difficulty might be compounded by other independent factors. For instance, some children might be more tolerant or less capable of recognizing contextual relevance, which is compatible with a general pragmatic delay in pragmatic competence. Moreover, the steps involved in the derivation of the scalar inference might require an extension of children’s processing resources, which is in line with a processing account. In fact, what we have proposed is simply that the complex semantics of cardinal existential quantifiers is likely to add further demands on these processing resources.

However, we believe that our findings are not easily reconcilable with the view according to which difficulties in computing SIs are exclusively due to processing (or pragmatic) limitations: if children’s problems only stemmed from a limited availability of processing resources, or from an immature pragmatic system, it would remain unclear how to model the interaction between a delay in the pragmatic module or limited processing resources, as surfacing in the some-underinformative condition, and the difficulty of properly interpreting cardinal some, as surfacing in the some-true condition. In fact, no pragmatic computation was required in the some-true condition.

We can actually build on the second major finding of the present study to propose an explanation of children’s performance with SIs in Mandarin Chinese. As already reported for youyixie ‘some’ in Katsos et al. (2016), we found that some (5 out of 22) children with RD and many children at age 5 and 6 have problems in correctly understanding the truth-conditions of ‘some’ in Mandarin Chinese (yixie in our study and youyixie in Katsos et al.’s study). It is important to highlight that when these children were asked to justify their rejection of some in the some-true scenario, they always made reference to the contextually-relevant universal set, by saying, for instance, that some girls included five girls, and in the story ‘only three girls took an apple, and the other two girls did not take an apple’. This non-adult-like behavior of Chinese children has never been reported for English, Italian or French (e.g., Foppolo et al., 2012; Guasti et al., 2005; Katsos et al. 2016). This finding suggests that certain children who are still not able to compute SIs also do not exhibit adult lexical knowledge of cardinal some: they are not able to interpret this quantifier as the cardinal property ‘the set contains at least one element’ (as in adult language), and in the absence of a well-defined cardinal specification, the relevant set is referred to (by default) as a whole. Accordingly, we propose that these children reject ‘some’ in the some-true condition because the cardinality quantifier remains actually semantically underspecified. In these conditions, no some/all scale can be activated and this explains why they accept ‘some’ in the some-underinformative condition. This conjecture also explains why they correctly reject some in a context in which none of the individuals are doing the action denoted by the predicate (i.e., in the some-false condition).

At first sight, this result is not consistent with the findings reported by Su and Su (2015), as discussed earlier. However, note that in our study, yixie ‘some’ was used (a cardinality quantifier), while in their study, youxie ‘some’ was tested (a proportional quantifier). As we have seen, Arosio et al. (2016), who investigated children’s behavior with the Italian proportional quantifier qualche ‘some’, found that proportional quantifiers essentially yield an adult-like performance in the computation of SIs in children. In other words, children tend to perform at ceiling in the some-underinformative condition whenever ‘some’ is a proportional quantifier (as is the case of qualche/quelques in Italian/French). These results contrast with those in Katsos et al. (2016), where the Chinese cardinality quantifier youyixie was tested, and the finding was that children exhibited serious deficits in non-pragmatic conditions. Significantly, however, these deficits concern the some-false condition in Katsos et al.’s experiment, a datum that partially contrasts with our results (i.e., the children performed at ceiling in the some-false condition), and that suggests that differences in the experimental design might have played a role as well.

Summarizing, our results suggest some firm conclusions, while also raising intriguing questions for future research. They suggest that processing (or pragmatic) limitations play an important role in computing sentences involving scalar quantifiers, at least for some individuals, confirming and extending the insights in Vender (2017), and as is made particularly evident by the observation that the Chinese children with RD perform worse than the age-matched controls and similarly to the younger TD children. At the same time, though, our findings strongly suggest that these processing (or pragmatic) limitations are not the whole story: they interact with a number of so-far largely neglected lexical factors, from which we have tentatively disentangled the role arguably played by the contrast between proportional and cardinality quantifiers, and the role played by an incomplete lexical knowledge of the quantifier itself. Indeed, the data reviewed above suggest that the acquisition of adult-like lexical knowledge of different classes of quantifiers might follow distinct learning paths, and that immature lexicons might reflect in the way lexical alternatives are accessed while computing SIs, confirming the insights in Foppolo et al. (2012) and in Foppolo and Marelli (2017). At the same time, the residual differences between our results and those discussed in the literature point to a possible effect of the difference in experimental designs, suggesting that conclusions in this domain are premature and that further research is required.

Finally, the line of analysis tentatively proposed here will have to be confirmed by comparing the behavior of ‘cardinality’ yixie and ‘proportional’ youxie in Mandarin Chinese more systematically and, more generally, by extending the investigation of cardinal and proportional ‘some’ to more languages where semantically distinct existential quantifiers are attested.

Conclusion

While this study is the first to examine the derivation of SIs in Chinese children with RD, it comes with a few limitations. One limitation, as mentioned, is that we did not compare yixie ‘some’ and youxie ‘some’ with the same group of children. Another limitation is that children with RD were poor readers assessed on the basis of literacy tests and teachers’ reports, instead of children with diagnosed dyslexia who would have been the ideal population for this investigation. Therefore, it might not allow us to generalize the findings of the present study to dyslexic children across languages.

We leave these issues to future research and conclude by emphasizing that the present study provides an important confirmation for the hypothesis that children with RD exhibit a delayed knowledge of scalar quantifiers and SIs, as compared with age-matched controls, and are in fact on a par with six-year-old children. These results support the view that processing and/or pragmatic competence are impaired in some children with RD. In addition, the overlap between the comprehension of some-underinformative and some-true statements for children with RD and younger children was an original finding of the present study, and it has been revealed as one crucial factor in modulating children’s behavior with underinformative statements. We propose that this is related to lexical/semantic factors, among which is the neglected difference between proportional and cardinality ‘some’. Arguably, these two classes of quantifiers trigger different patterns of behavior in children. The line of analysis that we have tentatively proposed in the present contribution will have to be confirmed by future research.

Footnotes

Appendix 1: Experimental items

Appendix 2 Acknowledgements

The authors are grateful to the children and the adults who participated in the study; to Gu Xiaoming, Hu Zhangjian and Shan Wei who helped us out with the data collection; to Hu Zhizhi, Huang Guoli, Xu Jun and Zhang Wenjun for their help in obtaining permission to work in kindergarten and primary school; and to Gaetano Fiorin who kindly made figures available for us. We also thank Angel Chan, Stephen Crain, Maria Teresa Guasti, Hu Chanchan, Lin Yunqiang, Maria Luisa Lorusso, Marco Marelli, Chloë Marshall, Chiara Melloni, Stephen Politzer-Ahles, Anne Reboul, William Snyder, Yi (Esther) Su, Rosalind Thornton, Kenneth Wexler, Xu Dan, Zhang Peixin, Zhang Yunqiu and Zhu Rui for their insightful suggestions. Authors’ contribution is as follows: All authors conceived the experimental question, developed the experimental tasks, and drafted the article; Shenai Hu recruited and tested the children, and the other authors supervised the testing; Francesca Foppolo and Shenai Hu performed the statistical analyses.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by the Fundamental Research Funds for the Central Universities under Grant Agreement No. 20720171054 and the European Union’s Seventh Framework Program for research, technological development and demonstration under Grant Agreement No. 613465.

ORCID iD

Shenai Hu

References

Arosio

Pagliarini

Perugini

Barbieri

Guasti

M. T.

(2016). Morphosyntax and logical abilities in Italian poor readers: The problem of SLI under-identification. First Language, 36, 295–315.

Arosio

Panzeri

Molteni

Magazù

Guasti

M. T.

(2017). The comprehension of Italian relative clauses in poor readers and in children with Specific Language Impairment. Glossa: A Journal of General Linguistics, 2, 9.

Barner

Brooks

Bale

(2011). Accessing the unsaid: The role of scalar alternatives in children’s pragmatic inference. Cognition, 118, 84–93.

Bar-Shalom

E. G.

Crain

Shankweiler

(1993). A comparison of comprehension and production abilities of good and poor readers. Applied Psycholinguistics, 14, 197–227.

Bott

Noveck

I. A.

(2004). Some utterances are underinformative: The onset and time course of scalar inferences. Journal of Memory and Language, 51, 437–457.

Breheny

Ferguson

H. J.

Katsos

(2013). Investigating the timecourse of accessing conversational implicatures during incremental sentence interpretation. Language and Cognitive Processes, 28, 443–467.

Byrne

(1981). Deficient syntactic control in poor reader: Is a week phonetic memory code responsible. Applied Psycholinguistics, 2, 201–212.

Cantiani

Lorusso

M. T.

Perego

Molteni

Guasti

M. T.

(2015). Developmental dyslexia with and without language impairment: ERPs reveal qualitative differences in morphosyntactic processing. Developmental Neuropsychology, 40, 291–312.

Cardillo

Garcia

R. B.

Mammarella

I. C.

Cornoldi

(2018). Pragmatics of language and theory of mind in children with dyslexia with associated language difficulties or nonverbal learning disabilities. Applied Neuropsychology: Child, 7, 245–256.

10.

Casalis

Leuwers

Hilton

(2012). Syntactic comprehension in reading and listening: A study with French children with dyslexia. Journal of Learning Disabilities, 46, 210–219.

11.

Chemla

Singh

(2014a). Remarks on the experimental turn in the study of scalar implicature, Part I. Language and Linguistics Compass, 8, 373–386.

12.

Chemla

Singh

(2014b). Remarks on the experimental turn in the study of scalar implicature, Part II. Language and Linguistics Compass, 8, 387–399.

13.

Chen

B. F.

Guo

T.-M.

(2012). Hanyu jicha cixiang ‘yixie’ de yuyi: Yuyong jiedu [The semantic-pragmatic interpretations of the Chinese scalar item ‘yixie’]. Waiyu Xuekan / Foreign Language Research, 6, 63–68.

14.

Crain

Thornton

(1998). Investigations in universal grammar: A guide to experiments on the acquisition of syntax and semantics. Cambridge, MA: The MIT Press.

15.

Degen

Tanenhaus

M. K.

(2016). Availability of alternatives and the processing of scalar implicatures: A visual world eye-tracking study. Cognitive Science, 40, 172–201.

16.

De Neys

Schaeken

. (2007). When people are more logical under cognitive load: Dual task impact on scalar implicature. Experimental Psychology, 54, 128–133.

17.

Doitchinov

(2005). Why do children fail to understand weak epistemic terms? An experimental study. In Kesper

Reis

(Eds.), Linguistic evidence: Empirical, theoretical, and computational perspectives (pp. 123–144). Berlin, Germany: Mouton de Gruyter.

18.

Fiorin

(2010). Meaning and dyslexia: A study on pronouns, aspect and quantification (Doctoral thesis). Utrecht Institute for Linguistics OTS, LOT, The Netherlands.

19.

Foppolo

Guasti

M. T.

Chierchia

(2012). Scalar implicatures in child language: Give children a chance. Language Learning and Development, 8, 365–394.

20.

Foppolo

Marelli

(2017). No delay for some inferences. Journal of Semantics, 34, 659–681.

21.

Grice

H. P.

(1957). Meaning. The Philosophical Review, 66, 377–388.

22.

Grice

H. P.

(1975). Logic and conversation. In Cole

Morgan

J. L.

(Eds.), Syntax and semantics (Vol. 3, pp. 41–58). New York, NY: Academic Press.

23.

Grice

H. P.

(1989). Studies in the way of words. Cambridge, MA: Harvard University Press.

24.

Griffiths

C. C. B.

(2007). Pragmatic abilities in adults with and without dyslexia: A pilot study. Dyslexia, 13, 276–296.

25.

Guasti

M. T.

Chierchia

Crain

Foppolo

Gualmini

Meroni

(2005). Why children and adults sometimes (but not always) compute implicatures. Language and Cognitive Processes, 20, 667–696.

26.

Hochstein

Bale

Barner

(2018). Scalar implicature in absence of epistemic reasoning? The case of autism spectrum disorder. Language Learning and Development, 14, 224–240.

27.

Horn

L. R.

(1972). On the semantic properties of logical operators in English (Doctoral dissertation). UCLA, Los Angeles.

28.

Horowitz

A. C.

Schneider

R. M.

Frank

M. C.

(2018). The trouble with quantifiers: Exploring children’s deficits in scalar implicature. Child Development, 89, e572–e593.

29.

Vender

Fiorin

Delfitto

(2018). Difficulties in comprehending affirmative and negative sentences: Evidence from Chinese children with reading difficulties. Journal of Learning Disabilities, 51, 181–193.

30.

Huang

Y. T.

Snedeker

(2009). Online interpretation of scalar quantifiers: Insight into the semantics–pragmatics interface. Cognitive Psychology, 58, 376–415.

31.

Joanisse

M. F.

Manis

F. R.

Keating

Seidenberg

M. S.

(2000). Language deficits in dyslexic children: Speech perception, phonology, and morphology. Journal of Experimental Child Psychology, 77, 30–60.

32.

Kamp

Reyle

(1993). From discourse to logic: Introduction to modeltheoretic semantics of natural language, formal logic and discourse representation theory. Dordrecht, The Netherlands: Kluwer.

33.

Kasirer

Mashal

(2017). Comprehension and generation of metaphoric language in children, adolescents, and adults with dyslexia. Dyslexia, 23, 99–118.

34.

Katsos

Bishop

D. V. M.

(2011). Pragmatic tolerance: Implications for the acquisition of informativeness and implicature. Cognition, 120, 67–81.

35.

Katsos

Cummins

Ezeizabarrena

M.-J.

Gavarró

Kuvač Kraljević

Hrzica

Noveck

(2016). Cross-linguistic patterns in the acquisition of quantifiers. Proceedings of the National Academy of Sciences, 113, 9244–9249.

36.

Katsos

Roqueta

C. A.

Estevan

R. A. C.

Cummins

(2011). Are children with specific language impairment competent with the pragmatics and logic of quantification? Cognition, 119, 43–57.

37.

Lam

K.-H.

S.-H.

(2014). Pragmatic skills in Chinese dyslexic children: Evidence from a parental checklist. Asia Pacific Journal of Developmental Differences, 1, 4–19.

38.

Leikin

Assayag-Bouskila

(2004). Expression of syntactic complexity in sentence comprehension: A comparison between dyslexic and regular readers. Reading and Writing: An Interdisciplinary Journal, 17, 801–822.

39.

Mann

V. A.

Shankweiler

Smith

S. T.

(1984). The association between comprehension of spoken sentences and early reading ability: The role of phonetic representation. Journal of Child Language, 11, 627–643.

40.

McBride

Wang

(2015). Learning to read Chinese: Universal and unique cognitive cores. Child Development Perspectives, 9, 196–200.

41.

McLoughlin

Leather

Stringer

(2002). The adult dyslexic: Interventions and outcomes. London, England: Whurr.

42.

Nation

Snowling

M. J.

(2000). Factors influencing syntactic awareness skills in normal readers and poor comprehenders. Applied Psycholinguistics, 21, 229–241.

43.

Nicolson

Fawcett

(2008). Dyslexia, learning and the brain. Cambridge, MA: The MIT Press.

44.

Noveck

I. A.

(2000). When children are more logical than adults: Experimental investigations of scalar implicature. Cognition, 78, 165–188.

45.

Papafragou

Musolino

(2003). Scalar implicatures: Experiments at the semantics–pragmatics interface. Cognition, 86, 253–282.

46.

Papafragou

Skordos

(2016). Scalar implicature. In Lidz

Snyder

Pater

(Eds.), The Oxford handbook of developmental linguistics (pp. 611–630). Oxford, UK: Oxford University Press.

47.

Partee

B. H.

(1988). Many quantifiers. In Powers

de Jong

(Eds.), ESCOL 89: Proceedings of the Eastern States Conference on Linguistics (pp. 383–402). Columbus: Ohio State University.

48.

Politzer-Ahles

Fiorentino

Jiang

Zhou

(2013). Distinct neural correlates for pragmatic and semantic meaning processing: An event-related potential investigation of scalar implicature processing using picture-sentence verification. Brain Research, 1490, 134–152.

49.

Pouscoulous

Noveck

I. A.

Politzer

Bastide

(2007). A developmental investigation of processing costs in implicature production. Language Acquisition, 14, 347–375.

50.

Reggiani

(2010). Dyslexia and the acquisition of syntax: Passive and control (Doctoral dissertation). University of Verona, Italy.

51.

Scarborough

H. S.

(1991). Early syntactic development of dyslexic children. Annals of Dyslexia, 41, 207–220.

52.

Skordos

Papafragou

(2016). Children’s derivation of scalar implicatures: Alternatives and relevance. Cognition, 153, 6–18.

53.

Smith

S. T.

Macaruso

Shankweiler

Crain

(1989). Syntactic comprehension in young poor readers. Applied Psycholinguistics, 10, 429–454.

54.

Stein

C. L.

Cairns

H. S.

Zurif

E. B.

(1984). Sentence comprehension limitations related to syntactic deficits in reading-disabled children. Applied Psycholinguistics, 5, 305–322.

55.

Y. E.

L.-Y.

(2015). Interpretation of logical words in Mandarin-speaking children with autism spectrum disorders: Uncovering knowledge of semantics and pragmatics. Journal of Autism and Developmental Disorders, 45, 1938–1950.

56.

Tieu

Romoli

Zhou

Crain

(2016). Children’s knowledge of free choice inferences and scalar implicatures. Journal of Semantics, 33, 269–298.

57.

Tsai

W. T. D.

(2003). Three types of existential quantification in Chinese. In Li

Simpson

(Eds.), Form, interpretation and functional structure: Perspectives from Asian languages (pp. 161–179). London, England: Curzon/Routledge.

58.

Tsai

W. T. D.

(2004). Tan ‘youren’ ‘youderen’ he ‘youxieren’ [On ‘youren’, ‘youderen’ and ‘youxieren’]. Hanyu Xuebao/Chinese Linguistics, 8(2), 16–25.

59.

van Tiel

van Miltenburg

Zevakhina

Geurts

. (2016). Scalar diversity. Journal of Semantics, 33, 137–175.

60.

Vender

(2009). Scalar implicature and developmental dyslexia. In Marrero

Pineda

(Eds.), Linguistics: The challenge of clinical application (Proceedings of the 2nd International Conference of Clinical Linguistics) (pp. 254–261). Madrid, Spain: Euphonia Ediciones.

61.

Vender

(2017). Disentangling dyslexia: Phonological and processing impairment in developmental dyslexia. Frankfurt, Germany: Peter Lang.

62.

Vender

Delfitto

(2010). Towards a pragmatic of negation: The interpretation of negative sentences in developmental dyslexia. Generative Grammar at Geneva, 6, 1–27.

63.

Waltzman

D. E.

Cairns

H. S.

(2000). Grammatical knowledge of third grade good and poor readers. Applied Psycholinguistics, 21, 263–284.

64.

Wilsenach

A. C.

(2006). Syntactic processing in developmental dyslexia and in specific language impairment: A study on the acquisition of the past participle construction in Dutch (Doctoral dissertation). Utrecht Institute for Linguistics OTS, LOT, The Netherlands.

65.

Wise

J. C.

Sevcik

R. A.

Morris

R. D.

Lovett

M. W.

Wolf

(2007). The relationship among receptive and expressive vocabulary, listening comprehension, pre-reading skills, word identification skills, and reading comprehension by children with reading disabilities. Journal of Speech, Language, and Hearing Research, 50, 1093–1109.

66.

Tan

(2009). Hanyu ertong yuyan zhong de dengji hanyi–Yixiang shiyan yanjiu [Scalar implicature in Chinese child language: An experimental study]. Waiguoyu / Journal of Foreign Languages, 32(3), 69–75.

67.

Zhang

Wang

(1985). Raven standard progressive matrices. Beijing, China: The National Revision Collaborative Group.

68.

Zhao

(2012). Hanyu dengji hanyi jiagong de shenjing jizhi yanjiu [The study on the neuromechanism underlying the scalar implicature processing in Chinese] (Doctoral dissertation). Zhejiang University, Hangzhou, China.