Acquisition of Mandarin long passives by children with developmental language disorder

ABSTRACT

This study investigated the comprehension and production of long passives (i.e. bei-constructions with an overt agent) in Mandarin-speaking children with developmental language disorder (DLD). Seventeen preschool children with DLD (1 female; mean age: 61 months old) and 23 typically developing (TD) children (6 females; mean age: 62 months old) participated in a sentence-picture matching task (for comprehension) and an elicited production task. Their nonverbal working memory (NVWM) was measured with the fourth edition of the Wechsler Preschool and Primary Scale of Intelligence. Results showed that children with DLD were less accurate and more likely to choose the picture with reversed thematic roles than their TD peers on passives in the sentence-picture matching task; in the elicited production task, they produced fewer target responses than TD children in passives. For NVWM, although that of the DLD group was lower than that of TD children, most children in the DLD group were within the average range. Furthermore, their performance on passives in the comprehension and production tasks was significantly correlated with their NVWM, which adds to the body of work suggesting links between complex syntax and working memory. However, the fact that NVWM could be preserved in the face of difficulties with passives suggests that this link may be due to NVWM enhancing performance during tasks with a high visual component, while it may not be underlyingly responsible for syntactic impairments in children with DLD.

KEYWORDS:

• Children with DLD
• Mandarin
• long passives
• bei-construction
• nonverbal working memory (NVWM)

Introduction

Developmental Language Disorder (DLD), or specific language impairment, refers to a language impairment that does not stem from a known biomedical aetiology (Bishop et al., 2017). Children with DLD show impairments in language abilities although they are normal in nonverbal intelligence (Leonard, 2014). The impairments are present in phonology (Georgiou & Theodorou, 2022; Quam et al., 2021), semantics (Haebig et al., 2017; Mainela-Arnold et al., 2010), pragmatics (Andrés-Roqueta & Katsos, 2020; Bühler et al., 2018), as well grammatical aspects (Chen & Durrleman, 2022; Dube et al., 2019; Sasaki et al., 2021). Further, children with DLD illustrate difficulties in social communication (Adams et al., 2018; Snowling et al., 2020) and reading ability (Botting et al., 2006; Kalnak & Sahlén, 2022). Among the grammatical deficits reported for this population are difficulties in the acquisition of complex structures that involve argument movement, such as relative clauses (Delage et al., 2007; Frizelle & Fletcher, 2014; Novogrodsky & Friedmann, 2006; Rakhlin et al., 2016; Yu et al., 2021), object questions (Dai & He, 2021; Friedmann & Novogrodsky, 2011), object clitic pronouns (Durrleman & Delage, 2016; Paradis et al., 2003; Tuller et al., 2011), and passives (Bishop et al., 2000; Leonard et al., 2006; Marinis & Saddy, 2013; Norbury et al., 2002; van der Lely, 1996).

Passives in DLD

This study focuses on the acquisition of long passives, where both arguments are realised however the mapping of thematic roles is inverted from active (1a). Long passives (1b), which contain a lexical logical subject, differ from short passives (1c) in which an overt lexical logical subject remains implicit. The inverted relation between grammatical and thematic roles in long passives poses problems for children with DLD, and also for young typically developing (TD) children, who acquire passives later than actives (Marchman et al., 1991; Pinker et al., 1987).

(1a) Active: The tiger scratched the lion.

(1b) Long passive: The lion was scratched by the tiger.

(1c) Short passive: The lion was scratched.

Comprehension of English long passives poses difficulties for children with DLD. van der Lely (1996) found that nine- to thirteen-year-old children with DLD were significantly less accurate and made more errors of choosing reversed thematic roles in long passives than five- to eight-year-old TD controls using the Test of Active and Passive Sentences (TAPS), a sentence-picture matching task. Bishop et al. (2000) reported similar findings to van der Lely (1996) that seven- to thirteen-year-old twin children with DLD made more reversed-thematic-role errors in the comprehension of long passives than TD children using the TAPS. The findings were later replicated, again with the TAPS, by Norbury et al. (2002), in six- to thirteen-year-old children with DLD. Marinis and Saddy (2013) conducted the revised TAPS task and the Self-paced Listening and Picture Verification task. They found that children with DLD performed worse not only on passives but also in actives as compared to their TD peers in both tasks, and they displayed more reversal errors than their TD peers. To sum up, consistent findings were revealed in previous studies, namely that children with DLD obtained lower accuracy on passives in the sentence-matching task than their TD peers and made the error of selecting the picture in which the thematic roles were reversed.

The difficulties children with DLD experience with passives go beyond the sphere of comprehension into that of production, as shown by Leonard et al. (2006). Using sentence completion tasks,¹ Leonard et al. (2006) investigated the production of passives by four- to six-year-old English-speaking as well as Cantonese-speaking children with DLD. The authors found that English-speaking children with DLD were less proficient than both age-matched and language-matched TD children but those who speak Cantonese only performed worse than the age-matched TD children in the production of passives.

The later acquisition in TD children and the delayed acquisition of passives in children with DLD is arguably due to the nature of the syntactic movement involved in the derivation of passives (Baker et al., 1989; Borer & Wexler, 1987; Chomsky, 1981). Syntactic movement has been argued to increase complexity (Moscati & Rizzi, 2014), which has been linked to the computational load involved as predicted by the Derivational Complexity Metric (DCM; Jakubowicz, 2005, 2011). According to the DCM, structures involving movement are more complex than those without movement, and require more computational resources to process, such as working memory, where children with DLD demonstrate deficits compared with their TD peers (Archibald & Gathercole, 2007; Baddeley, 2003; Briscoe & Rankin, 2009; Coady & Evans, 2008; Larson & Ellis Weismer, 2022; Montgomery et al., 2010; Talli & Stavrakaki, 2020; Vugs et al., 2017). In light of their soliciting increased working memory, movement structures are expected to be particularly challenging for children with DLD (Jakubowicz, 2011; Stanford et al., 2019; Torrens & Yagüe, 2018).

Nonverbal working memory

Working memory involves the temporary storage and manipulation of information (Baddeley, 2003). The verbal domain of working memory (i.e. verbal working memory; VWM) deals with the temporary storage and manipulation of auditory – linguistic information, while the nonverbal domain of working memory (i.e. nonverbal working memory; NVWM) manipulates and stores spatial, visual, and possibly kinaesthetic information (Baddeley, 2003). In line with the predictions of the DCM, deficits of children with DLD in their working memory are correlated with their weak processing ability of syntactic structures (Delage & Frauenfelder, 2020; Durrleman & Delage, 2016; Ellis Weismer et al., 2017; Fortunato-Tavares et al., 2015; Marinis & Saddy, 2013). However, the vast majority of studies concerned with the relationship between working memory and syntactic processing in children with DLD have focused on VWM rather than NVWM. Indeed, VWM has been argued to play an important role in the syntactic processing of children with DLD (Conti-Ramsden et al., 2015; Montgomery & Evans, 2009; Montgomery et al., 2021), and training on VWM has been shown to be beneficial for the acquisition of syntactic structures (Delage et al., 2021; Henry et al., 2022; Stanford et al., 2019). However, tasks measuring VWM (e.g. non-word repetition task, sentence repetition ask, etc.) render it difficult to disentangle effects related to linguistic processing from those related to memory processes (Ellis Weismer et al., 2017), especially in children with DLD who show deficits in linguistic processing. Therefore, the examination of NVWM could minimise the influence of linguistic processing on the measurement of working memory.

NVWM is significantly correlated with VWM (Leonard et al., 2007) as well as with language abilities (Montgomery & Evans, 2009; Vugs et al., 2016) and predicts language abilities in children with DLD, especially the receptive modality (Baddeley, 2003; Ellis Weismer et al., 2017; Karasinski & Ellis Weismer, 2010). However, no study to date has examined the relation between NVWM and the processing of complex structures (e.g. passives) in these children, while linguistic complexity has been hypothesised to recruit more working memory resources (Jakubowicz, 2005).

Furthermore, controversial findings have been reported on whether children with DLD display deficits in NVWM. Some studies have reported that NVWM is significantly weaker in children with DLD than in TD children (Bavin et al., 2005; Henry et al., 2012; Larson & Ellis Weismer, 2022; Marton, 2008) and that it is impaired in children with DLD (Henry et al., 2012; Vugs et al., 2013). However, other studies revealed that children with DLD demonstrate similar performance to their TD peers in NVWM tasks (Baird et al., 2010; Blom & Boerma, 2020; Ellis Weismer et al., 2017) and argued that NVWM is preserved in children with DLD (Archibald & Gathercole, 2007). The inconsistent findings of previous studies might be caused by the differences between these studies in task demands and task materials (Ellis Weismer et al., 2017). For instance, Henry et al. (2012) required the participants to recall the spatial location of nonsense diagrams, Archibald and Gathercole (2007) demanded the participants to recall the number and location of squares, and Ellis Weismer et al. (2017) requested the participants to match abstract shapes to shapes that they had previously been shown. Moreover, studies examining NVWM of children with DLD who speak Chinese languages (e.g. Mandarin) are scarce.

The present study investigates NVWM of Mandarin-speaking children with DLD using the two sub-tests that measure visual-spatial working memory in the Wechsler Preschool and Primary Scale of Intelligence-fourth edition (WPPSI-IV; Li & Zhu, 2014), including the Picture Memory test and the Zoo Locations test. The WPPSI-IV is an individually administered measure of cognitive intelligence by certified psychologists (or professionals from related fields with formal training in standardised psychological or educational testing) for children aged 2;6 through 7;7 (Soares & McCrimmon, 2013), which is a suitable tool to measure NVWM of preschool children with DLD of this study.

This study aims to explore on the one hand NVWM skills of Mandarin-speaking children with DLD, and on the other hand, whether links between NVWM and the processing of complex structures indeed arise, through an examination of Mandarin long passives, a complex structure involving syntactic movement.

Passives in Mandarin Chinese

In Mandarin Chinese (Mandarin), an SVO language (see (2)), passives involve the movement of the object from its canonical post-verbal position, such that the patient is realised before the agent (see (3)). Mandarin passives, which are similar to get passives in English (Huang et al., 2009), are mainly expressed using the morpheme bei.²

Depending on the realisation of the lexical logical subject, Mandarin passives are also divided into long passives and short passives (see (4)).

Table

Default active	Long Passive	Short passive
(2) 狮子打 了 老虎。 shizi da-LE laohu. lion hit-LE tiger ‘The lion hit the tiger’.	(3) 老虎 被 狮子 打了。 laohu BEI shizi da-LE. tiger BEI lion hit-LE ‘The tiger was hit by the lion’.	(4) 狮子被 打了。 shizi BEI da-LE. lion BEI hit-LE ‘The lion was hit”.

The structure of a long passive, which can be analysed similarly to the English tough construction, involves both complementation and movement (Chiu, 1995; Ting, 1998). In a long passive sentence, the morpheme bei is regarded as a main intransitive verb that selects an NP as its subject and a clausal category as its complement. The clausal category involves A’-movement of the embedded null object operator (NOP) to the specifier of the clausal category, from where it is then predicated on the subject of the passive sentence (see Figure 1).

Figure 1. Derivation of long passive (adopted from Huang et al., 2009).

Similar to long passives, the formation of short passives also involves complementation and movement. In this study, the long bei-constructions that include overt agents were examined for three reasons: first, short passives have been argued to potentially be parsed as adjectival passives (Borer & Wexler, 1987); second, examining long passives allows us to determine how children handle the non-canonical realisation of both arguments; third, this structure allows for a clean comparison between the processing of actives and passives. The term ‘passives’ is used to refer to long bei-constructions in the following of this study.

Studies examining the acquisition of Mandarin passives are rare and have mainly been concerned with factors such as referentiality and aspect: Using a visual-world paradigm, Huang et al. (2013) showed that five-year-old TD children were more likely to comprehend passives as actives when the passive morpheme bei appeared after the referential noun (e.g. ‘‘Seal BEI it eat” →The seal is eaten by it) than before the referential noun (‘‘It BEI seal eat” →It is eaten by the seal). By investigating the aspectual properties of ba-³ and bei- constructions in a new multimedia longitudinal child language corpus – Tong corpus, Deng et al. (2018) reported that TD children began to use bei-constructions (i.e. passives) when they were around two years old, and they used passives (as well as ba-constructions) more frequently in the situation of perfective aspect that requires telic predicates than in imperfective aspect. The findings of the present study will, on the one hand, contribute to our understanding of the acquisition of passives by TD children (recruited for this study) as compared to active sentences; and if TD children and/or children with DLD show worse performance on passives than on actives on the other hand, they will confirm that forming Mandarin long passive sentences is more complex than actives, plausibly due to the movement involved.

The current study

No work thus far has studied both the comprehension as well as the production of passives in Mandarin-speaking children with DLD. In addition, previous studies are controversial about whether children with DLD are impaired in NVWM, and no study has examined NVWM of children with DLD who speak Chinese to check if there is a relation between the comprehension and/or the production of passives in children with DLD and their NVWM. Thus, this study addresses these gaps in the literature by investigating the comprehension and production of passives in Mandarin-speaking children with DLD. Our results will shed light on whether Mandarin-speaking children with DLD show deficits in NVWM, and explore the links between NVWM and the performance of children with DLD on passives. Three research questions are posed in this study.

First, do children with DLD display lower performance on the comprehension and production of Mandarin passives than actives, as well as lower performance on passives as compared to age-matched TD peers? The formation of a passive sentence involves movement (Chiu, 1995; ; Ting, 1998), which makes it more complex than forming an active sentence according to the DCM (Jakubowicz, 2005, 2011). Therefore, children with DLD would be expected to show lower performance on the comprehension and production of passives than actives, and worse performance on passives than their TD peers, however, they may show similar performance to TD children on actives.

Second, do children with DLD show deficits in NVWM? Since NVWM is significantly correlated with VWM (Leonard et al., 2007), and children with DLD show deficits in VWM (Archibald & Gathercole, 2007; Baddeley, 2003; Briscoe & Rankin, 2009; Coady & Evans, 2008; Larson & Ellis Weismer, 2022; Montgomery et al., 2010; Talli & Stavrakaki, 2020; Vugs et al., 2017), preschool children with DLD in this study might show deficits in NVWM, as reported in previous studies (Henry et al., 2012; Vugs et al., 2013).

Third, are there links between the performance of children with DLD on the comprehension and production of passives and their NVWM? Strong links between NVWM and morphosyntactic processing of children with DLD have been reported (Ellis Weismer et al., 2017). Therefore, the performance of children with DLD on the comprehension and production of passives can be predicted to be associated with their NVWM.

Methodology

Participants

A total of 40 four- to six-year-old children were recruited in this study: 17 with DLD recruited from special education schools, kindergartens, and hospitals, and 23 TD controls recruited from kindergartens.

Measurements of both global language abilities and NVI were administered to all the participants. For global language abilities, receptive vocabulary was tested with the Peabody Picture Vocabulary Test-Revised (PPVT; Sang & Miao, 1990); language comprehension (LC) and production (LP)⁴ were tested using the Rating Scale for Pre-school Children with Language Disorder-Revised (RSPCLD-R; for children under six years old; Lin et al., 2008) or the Rating Scale for School Children with Language Disorder-Revised (RSSCLD-R; for children above six years old; Lin et al., 2009). Three scores, namely, that of PPVT-R, LC, and LP from the RSPCLD-R (or RSSCLD-R) were collected for each participant. At least two of the three scores for children with DLD were 1.25 SD below their age norms (following Tomblin et al., 1997); while all scores of TD children were equal to or above their age norms. Their NVI was measured with the Chinese version of the WPPSI-IV (Li & Zhu, 2014). Only two participants scored below 80, and we opted to include them because there are studies that consider 70 as the cut-off score of normal intelligence (e.g. Swillen et al, 1997; Wodka et al., 2013). All participants’ NVI was higher than 77.

Moreover, parents, teachers, or therapists were interviewed to ensure that no participants had hearing loss, neurological or psychiatric disorders, behavioural disorders, or emotional abnormalities. Finally, all parents signed consent forms for their children’s participation, which was approved by the Medical Ethics Committee of Xi’an TCM Hospital of Encephalopathy. Participants’ descriptive characteristics are shown in Table 1.

Table 1. Participants’ descriptive characteristics.

Group	N	MoA (SD) (Range)	PPVT(SD) (Range)	LC(SD) (Range)	LP (SD) (Range)	NVI (SD) (Range)
DLD	17 (1F)	61.4 (9.6) (48.5–80.0)	48.2 (11.4) (30–71)	21.8 (4.0)(13–26)	26.3 (5.5)(16–36)	95.3 (9.5)(77–107)
TD	23 (6F)	62.8 (7.4) (49.2–75.2)	69.6 (22.1)(29–116)	31.7 (2.7)(27–37)	39.7 (2.2)(37–45)	111 (10.7)(93–129)

Notes: N = Number of Participants, F = Female, MoA = Months of Age.

Results of independent-sample t-tests showed that the DLD and TD groups were matched in age (t(38)=-0.536, p = .595, g = 0.172). The TD group, however, scored significantly higher than the DLD group on language and intelligence (t_ppvt(38) = 3.980, p < .001, g = 1.164; t_LC(38) = 8.875, p < .001, g = 3.002; t_LP = 9.469, p < .001, g = 3.394) and the NVI (t(38) = 4.842, p < .001, g = 1.549).

Comprehension of passive sentences

Sentence-picture matching and sentence repetition are two classical paradigms used to measure receptive language. In this study, comprehension of the passive sentence was measured using a sentence-picture matching task. We conducted the sentence-picture matching task rather than the sentence repetition task in that the sentence repetition paradigm itself is claimed as a working memory task. Since we also explored the links between NVWM and the processing of passives by children with DLD, it would thus create an experimental confound with the working memory task if we used the sentence repetition task.

The global protocol of the sentence-picture matching task to assess passives was adopted from Durrlemann et al. (2017), which investigated the comprehension of French passives by children with autism spectrum disorder. However, we created the sentences and pictures in this study from scratch.

Materials

Actives and passives were examined in two separate sections. Actives have the structure ‘subject (agent)+verb±le (the perfective aspect marker) + object (patient)’, and passives have the structure ‘subject (patient)+BEI+object (agent)+verb ±le’. The verbs and nouns used in this task and the following production task were drawn from those used commonly in daily life and were acquired before the age of three (Li, 1995). There was one practice item and five test items in each section. The practice item was the same as the test items but was not included in the score.

The same verbs (and pictures) were used for actives and passives respectively. Following Durrlemann et al. (2017), there were four pictures for each test item. Each picture depicted three characters, with two being involved in the action and one being an observer. Besides the target picture (P1), the agent and patient were reversed in P2, the ‘observer’ substituted the agent in P3, and there was no action between the figures in P4. All pictures were cartoon pictures set against a white background, in which the figures were easy to recognise (see Figure 2).

Figure 2. Example of the comprehension task.

Procedures

Participants were tested individually in a quiet room. Actives were tested before passives with a twenty-minute interval between them. Test items were presented in a certain order in each section with a filler item between every two to three of them.

At the beginning of each section, which lasted about two minutes, an experimenter explained the task. Then, the experimenter displayed the pictures on a screen using Microsoft PowerPoint (on a laptop or iPad) and gave instructions on the practice item to participants to familiarise them with the task procedure (with passive as an example), ‘Which picture is “The brother was bit by the sister”? Please point to it.’ The experimenter praised participants for this practice item when they pointed to the target picture, and showed them the target picture when they pointed to other pictures. The test items and fillers were displayed after the practice item without feedback. We piloted the study on ten TD children around three-year-old, which confirmed that three-year-old TD children knew well all the vocabulary and could follow the instructions of this task (also those in the following priming picture-description task). An assistant recorded the picture numbers selected by the participants.

Data treatment

Correct answers corresponded to participants pointing to the target picture (i.e. choosing P1; even if they corrected the first choice by selecting the correct image quickly afterwards), and non-target pictures were the result of pointing to the wrong picture (i.e. choosing other pictures), or giving no response. Then the data of the DLD and TD groups were analysed from two aspects: 1) the accuracy of each sentence structure; 2) the proportions of non-target pictures chosen by the participants.

Production of passive sentences

Since passives are non-canonical structures and only constitute a small proportion of children’s production (Kline & Demuth, 2010), it is time-consuming to obtain passives from daily conversation. Furthermore, eliciting passives through video/picture story narration or picture-description tasks without given cues of passives may not get enough data within a limited time. Since children with DLD demonstrate similar syntactic priming effects to TD children (Miller & Deevy, 2006), an elicited production task utilising the syntactic priming effects was conducted to measure the production of actives and passives in this study. The production task was conducted individually two to three days before the comprehension task since the latter task may involve the target sentences from the former and thus could have otherwise influenced results.

Materials

Two sections were used to test participants’ production of actives and passives. Each section had five test items and one practice item that was the same as the test items but was not included in the score. The sentence structures of actives and passives are the same as those in the comprehension task.

Six more verbs were used besides the verbs (and pictures) used in the comprehension task. Half of them were used in the priming part, and another half in the target part.

There were two pictures for one test item (see Figure 3). The left picture was the priming picture described by the experimenter, and the right one was the target picture that should be described by the participant.

Figure 3. Example of the production task.

Procedures

The whole procedure, which lasted 5 minutes, was similar to the comprehension task with an initial explanation of the task, later practice, and the final test. In a test item (with passive as an example), the experimenter described the priming picture first, ‘In this picture, “The goat was hit by the tiger”? (Pointing to the priming picture)’. Then the experimenter elicited a response from participants by saying, ‘How about this picture? (If the participant did not give a response) The bear … ’. The priming sentence of a test item should be repeated no more than twice. An assistant recorded the test process and kept a written record of participants’ responses.

Data treatment

The responses transcribed by the assistant were cross-checked with the recordings before analysis. Then the participants’ production of actives was divided into target responses (i.e. SVO structure) and non-target responses (including ba-constructions and other responses such as saying ‘I don’t know’ or giving no response); their production of passives was divided into the target responses (i.e. long passives) and non-target responses (including the SVO structures, ba-constructions, short passives, and other responses). Then, the data were analysed from two aspects: 1) comparisons of the proportions of the target and non-target responses; 2) analysis of non-target responses.

Measurement of NVWM

The participants’ NVWM was drawn from the two sub-tests for working memory index (WMI) in the WPPSI-IV, including the Picture Memory test and the Zoo Locations test.

Picture memory test

The Picture memory test is a core WMI subtest that measures participants’ NVWM (Li & Zhu, 2014; Soares & McCrimmon, 2013; Wechsler, 2012).

Materials

There are 35 test items and two practice items (samples A and B) in this test. Each test item has one target picture on the first page, and 4 to 12 optional pictures (including the target picture) on the second page. All pictures in this test are in colour.

Procedure

Participants in this study (four- to six-year-old children) began with sample B, and then started from the seventh test item. In each item, the examiner asked the participants to watch the target picture on the first page for 5 seconds, then turned to the next page and required the participants to point out the target picture from the optional pictures. The test would be terminated if they selected non-target pictures for successive four test items.

Data treatment

Participants obtained 1 point for choosing the target picture, and 0 points for the non-target pictures and the test items that were not done.

Zoo Location test

The Zoo Locations test is a supplemental WMI subtest to measure participants’ NVWM (Li & Zhu, 2014; Soares & McCrimmon, 2013; Wechsler, 2012).

Materials

The zoo Location includes one practice item and 20 test items. For each item, there are 1 to 7 picture cards (representing animals) and 2 to 9 squares (representing different locations in a zoo) on one page.

Procedure

In each item, the examiner showed the participants where specific animals ‘live’ in a zoo, and asked the participants to remember the locations (represented by the squares) of animal(s) in the zoo. Then, the examiner handed the animal(s) to the participants and required them to place the cards in the correct locations from memory. The test would be terminated if they reproduced wrongly the animal locations for successive three test items.

Data treatment

Participants obtained 1 point if they reproduce all the animal locations for one test item, otherwise, they received 0 points.

Generation of standard WMI score

By inputting the raw scores of the two subtests in the Web-based Scoring and Reporting Platform, the standard WMI scores were generated. The standard WMI score in WPPSI-IV ranges from 40 to 160,⁵ and the score range from 90 to 109 was considered average since the vast majority of children tested scored in this range (Wechsler, 2012).

Data analysis

Data was conducted with intra-group and inter-group comparisons, as well as correlation analyses on the links between participants’ performances on the comprehension and production of the two structures and their NVWM. The normality of variables was checked using the Shapiro-Wilk tests before data analyses. All analyses were performed using the stats package (R Core Team, 2021) and effect sizes⁶ were evaluated using the effectsize package (Ben-Shachar et al., 2020) in R (R Core Team, 2021). Figures were produced using the ggplot2 package (Wickham, 2016). The p-values for all post hoc comparisons presented the adjusted significance.

Results

Results of the comprehension task

The data in the comprehension task were non-normally distributed (see Table 2). Therefore, Mann-Whitney tests were conducted to examine the inter-group differences, and Wilcoxon tests were used to compare the differences between actives and passives for each group of children.

Table 2. Tests of normality (Shapiro-Wilk tests) for the comprehension task.

	Actives		Passives
	DLD	TD	DLD	TD
accuracy (P1)	.805**	.324***	.876*	.406***
P2	.710***	.324***	.876*	.406***
P3	.560***	–	–	–
P4	.262***	–	–	–

Notes: – means no participantchose this picture; * refers to p < .05, ** refers to p < .01, *** refers to p < .001.

Accuracy of the two structures

The mean and SD of the accuracy for each sentence structure are shown in Table 3.

Table 3. Accuracy (%) in the comprehension task.

Group	Actives	Passives	Wilcoxon test
DLD	80.0 (22.4)	62.4 (32.3)	Z = 2.169*, r = 0.37
TD	98.3 (5.8)	95.7 (12.0)	Z = 1.134, r = 0.17
Mann-Whitney test	U = 103.0**, r = 0.51	U = 73.5***, r = 0.61

Notes: *refers to p ≤ .05, **refers to p ≤ .01, ***refers to p ≤ .001.

As shown in Table 3, results of Mann-Whitney tests demonstrated that the DLD group obtained significantly lower accuracy than the TD group on both actives and passives. Furthermore, results of the Wilcoxon test showed that the DLD group got significantly higher accuracy on actives than on passives.

Results of this section showed that the DLD group performed worse than the TD group on the comprehension of both actives and passives. Moreover, they demonstrated better performance on actives than passives.

Errors made by the participants

Each group’s error types are shown in Figure 4.

Figure 4. Proportion of each picture in the comprehension task.

Mann-Whitney tests revealed that the DLD group made significantly more errors in choosing P2 than the TD group in passives (U = 73.5, p = .001, r = 0.61), while no differences were found between the two groups in choosing the non-target pictures in actives.

Error analyses illustrated that the DLD and TD groups only selected P2 (reversed agent and patient) for the non-target picture for passives and the DLD group made significantly more errors than the TD group.

Results of the production task

Because data of the production task were non-normally distributed (see Table 4), Mann-Whitney tests and Wilcoxon tests were conducted to examine the inter-group and intra-group differences respectively for the results of the production task.

Table 4. Tests of normality (Shapiro-Wilk tests) for the production task.

	Actives		Passives
	DLD	TD	DLD	TD
long passives	.835**	.782***	.837**	.541***
short-passives	.723***	.867**	.618***	.333***
SVO structure	.728***	.496***	.726***	.412***
ba-constructions	.490***	.215***	.262***	–
other responses	.730***	.412***	.753***	–

Notes: – means no participant produced that type of sentences; *refers to p < .05, **refers to p < .01, ***refers to p < .001.

Proportions of the target responses

The means and SDs of the target responses for each sentence structure are shown in Table 5.

Table 5. Proportions (%) of target responses in the production task.

	DLD	TD	Mann-Whitney test
Actives	89.4 (20.1)	99.1 (4.2)	U = 134.5, r = 0.40
Passives	47.1 (40.6)	88.7 (21.6)	U = 79***, r = 0.54
Wilcoxon test	Z = 2.953**, r = 0.51	Z = 1.841, r = 0.27

Notes: **refers to p ≤ .01.

Results of the Mann-Whitney tests revealed that the DLD group produced significantly fewer target responses than the TD group in passives, while they did as well as the TD group in actives (see Table 5).

Results of the Wilcoxon tests showed that the DLD group produced significantly more target responses in actives than in passives, however, no significant difference was found between the target responses between actives and passives for the TD group (see Table 5).

Results of this section illustrated that the DLD group produced significantly fewer target sentences in passives than the TD group. Furthermore, unlike the TD group which performed well on both actives and passives, the DLD group produced more target responses in actives than in passives.

Types of non-target responses

The participants’ production of actives was divided into three types: the SVO structure (target structure), ba-construction, and other responses (including saying ‘I don’t know’ and giving no response). Their production of passives was divided into five types: the SVO structure, ba-construction, short passives, and long passives, and other responses. Distributions of the production types were shown in Figure 5.

Figure 5. Proportion of each production type.

Mann-Whitney tests did not demonstrate significant differences between the two groups in the two non-target responses for actives: ba-construction: U = 146, p = .182, r = .342, other responses: U = 184, p = .766, r = .176. For passives, the DLD group produced significantly more other responses than the TD group (U = 103.5, p = .011, r = 0.567), while no significant differences were observed between the two groups in other three non-target responses for passives: short passives: U = 158, p = .315, r = 0.242, SVO structure: U = 135, p = .101, r = 0.342, ba-construction: U = 184, p = .766, r = 0.176.

To sum up, results of this section demonstrated that the DLD group did not show significant differences from the TD group in these non-target responses except that they made significantly more other responses than the TD group.

Participants’ scores on NVWM

Participants’ NVWM

The mean, SD, and range of scores of NVWM are displayed in Table 6.

Table 6. Participants’ scores of NVWM.

Group	Mean	SD	Range
DLD	92.9	9.3	72–107
TD	99.4	8.5	87–118

Because the data of NVWM were normally distributed (DLD: W = 0.201, p = .065; TD: W = 0.129, p = .200), independent t-tests were conducted to examine whether there were differences between the two groups in their NVWM. Results of independent-sample t-tests illustrated that the DLD group was significantly lower than the TD group in NVWM scores, t(38) = 2.316, p = .026, g = 0.741.

Individually, NVWM scores of three children in the DLD group were below 90, five were equal to or above 100, and the rest were between 90 and 99; one child in the TD group whose NVWM score was below 90, nine were between 90 and 99, three were above 109, and the rest were between 100 and 109.

Results illustrated that the DLD group was significantly lower in NVWM scores than the TD group, but the majority of children in the DLD group might preserve NVWM since their scores were within the average range.

Correlation analyses

The correlations between participants’ performance in the comprehension task (accuracies) and the production task (proportions of target responses) and their NVWM were measured with the Kendall correlation analysis.

As shown in Table 7, NVWM was significantly correlated with both the comprehension of actives and passives in the DLD group, while it was significantly correlated with passives in the TD group. However, there was only a significant correlation between NVWM of children with DLD and their performance on the production of passives.

Table 7. Results of correlations analyses (r).

		Actives		Passives
		DLD	TD	DLD	TD
NVWM	comprehension	.465*	.142	.416*	.419*
	production	.120	−.281	.485*	.158

Notes: NVWM= nonverbal working memory; *refers to p < .05.

Correlation analyses showed that the performance of children with DLD on the comprehension of both actives and passives was significantly linked to their NVWM. However, only the performance of children with DLD on the production of passives was significantly linked to their NVWM.⁷

Discussion

Lower performance on passives in DLD compared to TD

Using a sentence-picture matching task, we found that children with DLD performed worse than their TD peers on the comprehension of passives, and they demonstrated worse performance on the comprehension of passives than actives. This finding is consistent with studies on children with DLD who speak other languages (Bishop et al., 2000; Marinis & Saddy, 2013; van der Lely, 1996). Furthermore, similar to children with DLD in these studies, those in this study were also more likely than their TD peers to choose the picture in which the thematic roles of agent and patient were reversed in passive condition. In the elicited production task, children with DLD produced fewer target passives than TD children when they were primed with passives. The result was reminiscent of children with DLD in Leonard et al. (2006), in which both English-speaking and Cantonese-speaking children with DLD demonstrated worse performance than age-matched TD children on the production of passives. In addition, like children with DLD in Leonard et al. (2006), those in this study produced actives (e.g. default SVO structure and ba-construction) even when they were primed with passives. The performance of children with DLD tending to analyse passives as actives in both the comprehension and production tasks may indicate that they indeed assigned the agent and patient roles in a canonical manner, thus suggesting that they avoided applying a movement analysis of passives.

Mandarin passives involve both complementation and movement. The operations of complementation and movement increase the complexity of processing these structures, as predicted by the DCM (Jakubowicz, 2005, 2011). As simpler structures are privileged over more complex ones during language development due to economy considerations (Chomsky, 1995), it is expected that TD children acquire passives later than actives (Deen, 2011; Deng et al., 2018; Marchman et al., 1991; Pinker et al., 1987), and passives remain affected in children with DLD (Bishop et al., 2000; Leonard et al., 2006; Marinis & Saddy, 2013; Norbury et al., 2002; van der Lely, 1996). The complexity of syntactic structure is arguably linked to the computational load (Jakubowicz, 2011; Stanford et al., 2019; Torrens & Yagüe, 2018), and more complex structures require more computational resources, such as working memory. The weak working memory of children with DLD is linked with their processing of complex syntactic structures (Delage & Frauenfelder, 2020; Durrleman & Delage, 2016; Marinis & Saddy, 2013). As a result, worse performance of children with DLD than TD children on the process of passives than actives would be expected to relate to the complexity of passives and the weak working memory of children with DLD. However, previous studies on the links between syntactic processing of children with DLD and their working memory mainly focused on verbal working memory (VWM), thus it was uncertain whether working memory deficits in children with DLD also appeared in the nonverbal domain.

NVWM in children with DLD

Working memory involves the temporary storage and manipulation of information (Baddeley, 2003). Children with DLD show deficits in their working memory (Archibald & Gathercole, 2007; Baddeley, 2003; Briscoe & Rankin, 2009; Coady & Evans, 2008; Larson & Ellis Weismer, 2022; Montgomery et al., 2010; Talli & Stavrakaki, 2020; Vugs et al., 2017), and their low working memory scores correlate with their low scores on syntactic processing (Delage & Frauenfelder, 2020; Durrleman & Delage, 2016; Ellis Weismer et al., 2017; Fortunato-Tavares et al., 2015; Marinis & Saddy, 2013). Unlike the vast majority of studies that have explored the links between working memory and syntactic processing in children with DLD focused on VWM, this study investigated NVWM of children with DLD and its correlations with the comprehension and production of Mandarin passives.

NVWM, which to recall deals with visual-spatial information, is reported to correlate with the language abilities of children with DLD (Montgomery & Evans, 2009; Vugs et al., 2016) and even predict their language abilities (Baddeley, 2003; Ellis Weismer et al., 2017; Karasinski & Ellis Weismer, 2010). However, controversial findings have been reported in previous studies on whether children with DLD display deficits in NVWM because of differences in task demands and the nature of the nonverbal stimuli (Ellis Weismer et al., 2017). To be specific, some studies have reported that children with DLD show lower NVWM than TD children (Bavin et al., 2005; Henry et al., 2012; Larson & Ellis Weismer, 2022; Marton, 2008), and thus report they may be impaired for NVMW (Henry et al., 2012; Vugs et al., 2013), whereas other studies found that children with DLD demonstrate similar performance on NVWM tasks to their TD peers (Baird et al., 2010; Blom & Boerma, 2020; Ellis Weismer et al., 2017), and thus it has been argued that they are preserved for NVWM (Archibald & Gathercole, 2007).

Using two sub-tests for working memory index (WMI) in the WPPSI-IV, this study revealed that children in the DLD group obtained significantly lower NVWM scores than their age-matched TD peers, which is in line with other studies that show weaker performance of children with DLD in NVWM than TD children (Bavin et al., 2005; Henry et al., 2012; Larson & Ellis Weismer, 2022; Marton, 2008). Nonetheless, upon close inspection, only three among the 17 children in the DLD group in this study were below 90 and the rest of them were within the average score range (90–109) in which the majority of the population aged two- to seven-year-old is situated, which concurs with Archibald and Gathercole (2007). In Archibald and Gathercole (2007), composite scores of measures for NVWM fell in the average range for their DLD group overall, with a minority of children in this group scoring below the average. Therefore, children with DLD might indeed largely (although not systematically) show preserved NVWM, as claimed by Archibald and Gathercole (2007).

Links between passive performance and NVWM in children with DLD

In this study, we examined the links between NVWM of children with DLD and their performance on the comprehension and production of Mandarin long passives. The results of this study showed that NVWM of children with DLD was significantly correlated with their performance on the comprehension and production of passives. Furthermore, the performance of TD children on the comprehension of passives was also correlated with their NVWM. The findings of this study were consistent with those of Ellis Weismer et al. (2017), in which NVWM predicted the detection of morphosyntactic errors for children with DLD as well as TD children in a grammatical judgement task. Therefore, the links between NVWM and passives’ processing in this study are in line with the claim that NVWM is a predictor of language abilities (Baddeley, 2003; Ellis Weismer et al., 2017; Karasinski & Ellis Weismer, 2010).

However, the significant correlation between NVWM measure with the comprehension and production tasks cannot entirely account for performance on passives. Indeed, despite this correlation, children with DLD are relatively spared for NVWM while they show impairments on passives. We speculate that this link may then be due to the fact that NVWM simply enhances certain instances of task performance, and in the case of our study it would play a role in the sentence-picture matching task and the elicited production task. More specifically, the participants with stronger NVWM would be better equipped to process information in the pictures of these tasks than those with weaker NVWM. Recall that in both the sentence-picture matching task and the elicited production task, the participants needed to recognise and obtain information from the pictures presented to them first and then to make a choice or describe the pictures presented to them. Children with DLD who are able to parse visual information more efficiently may have an edge when subsequently having to comprehend or produce syntactic structures associated with these pictures, but this edge does not suffice to override their syntactic impairment entirely, thus their weakness with passives despite an absence of weakness in NVWM.

In sum, deficits in NVWM may influence task performance, including on passives, but difficulties in the comprehension and production of passives cannot be reduced to difficulties in NVWM. This is similar to what has already been claimed for NVI and language deficits in this population, namely that despite correlations previously reported between NVI and syntactic task performance⁸ (Fortunato-Tavares et al., 2015; Marinis & Saddy, 2013), children with DLD are often of normal nonverbal intelligence (Leonard, 2014) while they are impaired in language abilities, which indicates that nonverbal challenges may impact task performance without being underlyingly responsible for language impairments in this population.

The finding of this study may have clinical implications for the treatment of children with DLD. Namely, training on NVWM would not be expected to be as effective as training on VWM for enhancing mastery of passives in DLD since NVWM is largely preserved in children with DLD, in contrast to VWM.

Limitations

First, VWM was not tested in this study. Previous studies have revealed that children with DLD show deficits in their VWM and their difficulties in VWM are related to their processing of complex syntactic structures, thus it might have been interesting to compare VWM and NVWM in terms of links to passives, which we leave for future work.

Second, short passives were not investigated in this study. In Mandarin, the formation of both long and short passives involves complementation and movement. Although some have argued that short passives may be parsed while circumventing a movement analysis (Borer & Wexler, 1987), explaining earlier mastery in TD children (Armon-Lotem et al., 2015), it could nevertheless have been interesting to compare performance with these structures on that of long passives, which we also leave for future work.

Third, there were only 17 children with DLD investigated and 10 test items (5 items per condition) in the comprehension or production task in this study, which caused all the data retrieved from the two tasks to be non-normally distributed (confirmed with the Shapiro-Wilk tests), and nonparametric analyses were consequently conducted. Future work should seek to include more participants so that parametric tests could be conducted.

Fourth, all active sentences were tested before the passive sentence in both the comprehension and production tasks, which may have yielded an order effect. Therefore, the order of presenting active and passive sentences should be balanced between participants in future studies.

Conclusion

This study investigated the comprehension and production of passives in preschool Mandarin-speaking children with DLD using a sentence-picture matching task and an elicited production task, as compared with age-matched TD children. Participants’ NVWM was also tested with two sub-tests in the WPPSI-IV. Our results demonstrated that children with DLD performed worse than TD children on both the comprehension and production of passives. In the comprehension task, children with DLD performed worse on passives than on actives; they were less accurate and more likely to choose pictures with reversed thematic roles than their TD peers for passives. In the production task, children with DLD produced fewer target responses than TD children for passives. For NVWM, although that of the DLD group was lower than that of TD children, only three children in the DLD group were lower than average NVWM scores. The performance of children with DLD in both the comprehension and production tasks was correlated with their NVWM, despite most of them not showing deficits in NVWM. One might speculate that these global cognitive skills are useful for successful task performance, despite not being the underlying reason for the grammatical challenges associated with children with DLD.

Acknowledgments

We are grateful to the children, parents, teachers and therapists who participated in the study and to our research team in Guangdong University of Foreign Studies, especially Ms. Yunting Wang, whose collaboration made the study possible.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The Swiss National Science Foundation (grant no. PR00P1_193104/1), the National Social Science Foundation of China [grant no. 17AYY08], Guangdong Planning Office of Philosophy and Social Science [GD19YYY05].

Notes

¹ Target sentences were modelled by one of the experimenters (E1) before the participant gave his/her answer in Leonard et al. (2006), as exampled in the following.

E1: The bird wants to throw something.

Bird: Should I throw the airplane or the baseball?

Child: the baseball (Bird then throws the baseball) The bear wants to hug someone.

Bear: Should I hug Ernie or Snow White?

Child: Snow White (Bear then hugs Snow White)

Freddy: I wasn’t paying attention. What just happened?

E1: Let’s tell Freddy what happened to the ball and what happened to Snow White. The baseball got thrown by the bird and …….

² The passives in Mandarin can also be expressed through other morphemes such as jiao, gei, and rang.

³ The ba-construction is another non-canonical structure of the default active SVO structure. In the ba-construction, the object of an SVO sentence surfaces as the object of ba in the ba-construction to receive an emphatic reading associated with a resultative subevent.

⁴ In the RSPCLD-R or RSSCLD-R, the comprehension and production of global language abilities are tested, including semantics, syntax, pragmatics and vocabulary.

⁵ The standard score ranges of WPPSI-IV are divided in to seven levels: below 70 is extremely low, 70–79 is borderline, 80–89 is low average, 90–109 is average, 110–119 is high average, 120–129 is superior, and above 130 is very superior (The Test Tour, 2016).

⁶ In this study, Hedges’ g value was used to measure the effect size of the parametric tests, r value was used to evaluate the effect size of the non-parametric tests.

⁷ It is controversial whether syntactic priming reflects shorter-term activation of syntactic knowledge (Pickering & Branigan, 1998) or implicit learning (Bock & Griffin, 2000). Since NVWM of TD children was not significantly correlated with their performance on passives, it is not immediately obvious that syntactic priming sufficed to provide short-term activation of syntactic knowledge.

⁸ Significant correlations between NVI of children with DLD and their comprehension and production of passives were found in this study, comprehension: r = 0.489, p = .012, production: r = 0.554, p = .004.