2.1 NP types

2.2 Word order

In Mandarin, definiteness can be expressed through different types of NPs, but their distribution is subject to word order. The subject position typically necessitates a definite or generic NP, while the canonical object position remains neutral in terms of definiteness, allowing for the occurrence of both definite and indefinite NPs. This situation is different from Germanic and Romance languages, like English or Italian, in which definiteness does not give rise to the subject/object asymmetry (Cheng & Heycock & Zamparelli 2017). Apart from the subject position, the topic and post-ba positions also impose definiteness requirements on the occupying NPs.

As with the subject position, the topic position also exhibits clear prohibition against indefinite NPs, so that NPs that occur in this position must be either definite or generic. As a result, bare nouns, which are known to be unmarked for definiteness, are always interpreted as definite or generic when they are topics (Li & Thompson 1981: 86; Huang et al. 2009: 200).

The post-ba position seems to be less demanding on the occupying NPs regarding definiteness, as compared to the subject and topic positions. Ba is a special lexical category in Chinese languages, meaning “take” or “hold”. It is treated as a pure case-assigner (Huang et al. 2009: 174), an object marker (Li 2013: 119), or a functional element (Zhang 2019: 181). Structurally, the object NP is placed immediately after ba and before the verb, deriving the structure typical of ba sentence [NP1+ba+NP2+V+XP]. The ba construction carries a special meaning of “disposal” (Huang et al. 2009: 172), so that the subject of ba (NP₁) plays the role of causer while the post-ba NP (NP₂) is the target being disposed of. Consider the example in (4):

The post-ba NP is traditionally considered as definite or generic (Li & Thompson 1981: 465), but more recent research revised that specific-indefinite complex NPs (Num-Cl-NPs and bare Cl-NPs) are also acceptable in this position (Zhang 2019). What seems certain is that bare NPs following ba only exhibit definite or generic readings, while Num-Cl-NPs and bare Cl-NPs in the post-ba position are also acceptable only if they show specific-indefinite readings.

3.1 Development of definiteness distinctions in Mandarin-speaking children with TD

How Mandarin-speaking children distinguish between definite and indefinite reference so far has been examined mainly by means of elicited-production tasks.

Hickmann & Liang (1990) investigated the development of definiteness contrasts by collecting picture-based narratives produced by children of 4, 5, 6, 7, and 10 years old respectively. The stimuli for eliciting narratives are two sets of picture sequences, and each picture sequence describes an independent story. In the experiment, each child narrated stories to a blind-folded experimenter who cannot see the pictures, so as to eliminate the common ground between the speaker and hearer. The results showed that: (1) For the introduction of referent, it was not until 7-year-old did children make systematic use of numeral-classifier NPs (usually abbreviated to Num-Cl-NP, which is a specialized lexical form for referent introduction in Mandarin); while children under this age tended to indicate indefiniteness ambiguously by bare NPs. Adults used Num-Cl-NPs 86% of the time, while they used bare NPs and Dem-Cl-NPs 9% and 5% of the time. (2) For the subsequent mention of referent, the occurrence of bare NPs ranged between 65% and 82%, while Dem-Cl-NPs account for between 33% and 11%. As for controls, adults made use of bare NPs (52%) and Dem-Cl-NPs (44%) in similar proportions. Note that children at all ages rarely used Num-Cl-NPs for referents maintenance, which only account for 2% to 8%, suggesting that the target-like use of definite NPs was acquired earlier than that of indefinite NPs.

Hickmann et al. (1996) further replicated the same experiment in a larger cross-linguistic study, and the findings confirmed that Mandarin-speaking children still performed comparably to those in Hickmann & Liang (1990). Besides, the more recent research by Wu & Huang & Zhang (2015) also reported similar results in children.

According to the above findings, the target-like production of definite expressions arises earlier than that of indefinite ones in Mandarin-speaking children, and they have already shown mastery of definite expressions prior to the age 5.

3.2 Definiteness contrasts in children with DLD

The literature on definite-indefinite distinctions by children with DLD involves primarily children who acquire Romance and Germanic languages. The mastery of definiteness contrasts therefore can be examined by observing children’s performance on articles.

Earlier research mainly focused on the inclusion or omission of articles in children’s utterances (see the cross-linguistic review in Leonard (2016)), and the common phenomenon is that children with DLD omitted articles more frequently than both typically-developing children matched on age (TDA hereafter) and typically-developing younger children matched on language abilities (TDY). The inconsistent use of articles and tense morphology has been considered clinical markers of DLD (Rice & Wexler 1996).

Unlike earlier studies that focused on the presence or absence of articles in obligatory contexts, Polite & Leonard & Roberts (2011) investigated the definite-indefinite distinction in the production of articles by English-speaking children with DLD. The experiment was carried out verbally, and no visual stimuli were provided during the whole experiment process. The experimenter first introduced children with a short story (1–3 sentences in length) as background information, and then asked questions designed to elicit articles in the anaphoric-definite and nonspecific-indefinite environments respectively, which was in line with the Condition 2 and 5 in Schafer & de Villiers (2000). According to the results, in the definite condition, children with DLD showed significantly less accurate use of definite articles (19.79%) than children in the TDY (44.44%) and TDA (76.34%) groups; while in the indefinite condition, the three groups did not differ from each other in the use of indefinite articles (DLD: 86.97%; TDA: 89.14%; TDY: 84.32%).

Polite & Leonard & Roberts (2011) explained that the poor performance on definite articles by children with DLD may result from working memory limitations, as they found that the substitution of indefinite articles for definite ones (50 occurrences) occurred more frequently than the omission of definite articles (25 occurrences). They considered that the retention of a previously established referent in mind may impose demands on short-term memory; therefore, children with DLD failed to recall the referents established in previous discourse and then re-introduced them with an indefinite article in the subsequent mention.

Using the same approach, Blom & Vasić & Baker (2015) and Chondrogianni & Marinis (2015) investigated article distinctions by 7-year-old Dutch-speaking and English-speaking children with DLD respectively. Blom et al. (2015) examined indefinite articles in the non-specific condition, while Chondrogianni & Marinis (2015) the nonreferential-predicational, nonreferential-instrumental, and referential-specific conditions. As for definite articles, Blom et al. (2015) included the bridging condition, while Chondrogianni & Marinis involved both the bridging and anaphoric conditions.

Blom et al. (2015) found that article use in Dutch children with DLD is compromised in both definite and indefinite contexts. To be specific, the DLD group was accurate 61% of the time when using definite articles, greater than the data (19.79%) in Polite et al. (2011); whereas the accuracy for indefinite articles dropped to 55%, lower than that (86.97%) in Polite et al. As for error patterns, children with DLD made more omission errors (30%) than substitution ones (8%) in the definite condition, whereas they substituted (20%) and omitted (27%) articles at similar rate in the indefinite condition. Blom et al. (2015) believed that the children with DLD performed better on definite articles because the bridging condition is less demanding than the anaphoric condition, and besides, the participants are two years older in their research. The lower accuracy in indefinite articles, however, was due to the reason that Dutch articles are acquired later. The findings indicated that processing limitations may lead to less stable lexical knowledge of articles, which impedes the successful integration of lexical, syntactic, and pragmatic information necessary for appropriate use of articles.

Chondrogianni & Marinis (2015) found that the English-speaking children with DLD in their research had higher accuracy in the bridging environment (52.4%) than in the anaphoric one (24.3%). The mean score obtained in the bridging environment is close to that reported in Blom et al. (2015) (61%), while the data in the anaphoric condition is almost in line with that in Polite et al. (2011) (19.79%) (Note also that the participants are two years older than those in Polite et al. (2011)). The results suggest that the vulnerability of using anaphoric definite articles found in the pre-school children with DLD may continue into school age. As for indefinite contexts, children in the three groups did not differ from each other in the non-referential predicational (DLD: 84%; TDA: 82.8%; TDY: 81.7%), nonreferential-instrumental (DLD: 63.9%; TDA: 75.3%; TDY: 73.3%), and referential-specific conditions (DLD: 60.7%; TDA: 70.7%; TDY: 71.4%). With respect to error patterns, the children with DLD made an equal number of substitutions and omissions within each definite condition, and they made significantly more substitutions than the TDA controls in both the anaphoric and bridging conditions, whose error types were primarily omissions. In the indefinite contexts, omissions were the predominant error pattern in the three semantic conditions across the groups. Chondrogianni & Marinis argue that children with DLD had difficulties in integrating discourse information.

It can be summarized that children with DLD have difficulties primarily with definite forms, and they appear to have problems more with the anaphoric condition than the bridging one.

3.3 Definiteness distinctions in children with ASD

Definiteness distinctions in ASD are a topic that has not been widely investigated, as compared to the rich literature on children with TD and children with DLD.

Modyanova (2009: 95) examined the comprehension of determiners in a combined group of individuals with ASD. Participants were in the range between 6 to 18 years old and they were not further subdivided in terms of age. The experiment was carried out by children controlling a Velcro-backed toy to stick to an identical or different types of objects that were permanently attached to a page made of felt cloth, in accordance with the experimenter’s instructions that contain either indefinite determiners “a” and “another”, or definite determiners “that” and “the”.

The findings indicated that children with ASD differed from the control group merely on definite determiners, but not on indefinite ones. Variations were observed within the ASD group, because the children with ASD who have normal language (ALN hereafter) and normal IQs (the “Asperger group” in Modyanova (2009)) presented near-perfect knowledge of both definite and indefinite determiners, while the children with ALI and normal IQs (the “PDD-NOS group”) and the children with ALI and below-average IQs (the “Autism group”) performed poor on definite determiners.

Modyanova believes that children’s overall grammatical development is a significant predictor of their knowledge of definiteness, and that the acquisition of the definite determiner is influenced by the computational semantic system of language, rather than solely by pragmatics.

In Schaeffer (2018), article choice was compared between Dutch-speaking children with ALN and children with DLD. The experiment includes an elicited production task with pictures and video clips as stimuli, in which the indefinite article was examined in both specific and non-specific indefinite environments while the definite article was examined in the anaphoric environment. The age range of the individuals in each group was between 5 and 14 years. The mean ages for the ALN, DLD, and TDA groups were 10;00, 9;06, and 9;10 respectively, and no age groupings were made. Note that children in the ALN group exhibited normal intelligence.

The results showed that the three participant groups demonstrated target-like performance on both specific-indefinite (ALN: 96%; DLD: 98%; TDA: 99%) and nonspecific-indefinite conditions (ALN: 96%; DLD: 98%; TDA: 99%). As for the definite condition, both the ALN (80%) and DLD groups (81%) were significantly outperformed by the TDA controls (95%), and no significant differences were found between the two clinical groups. The error pattern in the definite condition found in the DLD and ALI groups is characterized by substitution errors, which aligns with the error pattern identified in Polite et al. (2011).

From the studies above, to conclude, children with ALN who have normal IQs are more likely to perform at a target-level on the comprehension of both definite and indefinite articles, as well as on the production of indefinite articles (but not definite ones). On the other hand, children with ALI, regardless of whether they have normal IQs or not, demonstrate clear problems with understanding definite articles, but not with indefinite ones.

3.4 Research questions for the current study

Based on previous studies and the characteristics of Mandarin, our research questions are formulated: (1) Do pre-school children with ALI and children with DLD show difficulties in the comprehension of definite expressions? (2) Are the three definite expressions equally interpreted within the DLD and ALI groups respectively? (3) If there is a significant difference in comprehending one or more definite expressions between the two clinical groups, what are the contributing factors?

4.1 Participants

The recruitment and screening of participants took place in mainland China. A total of 88 Mandarin-speaking children were recruited in the current study: 28 children with suspected DLD, 32 children with ALI, and 28 children with typical language. In addition, there were 8 children who had to be excluded for unwillingness to cooperate. The children in the DLD group were defined as having suspected DLD because they did not receive a medical diagnosis of DLD by speech-language pathologists. Parent/guardian consent forms have been obtained for all children. Table 1 provides detailed information of the participant groups.

Each child was administered a battery of standardized tests to assess their receptive vocabulary knowledge, language abilities, and nonverbal intelligence: (1) the receptive vocabulary was assessed with Peabody Picture Vocabulary Test–Revised Chinese Version (PPVT-R) (Sang & Miao 1990); (2) The rating scale for pre-school children with language disorder–Revised (RSPCLD-R) (Lin et al. 2008) or The rating scale for school children with language disorder–Revised (RSSCLD-R) (Lin et al. 2009) was administered to assess the language abilities of children below or above 6 years old respectively, and each scale consequently provided three indicators, namely the overall language, receptive language, and expressive language; (3) children’s IQs were measured by using Wechsler Preschool and Primary Scale of Intelligence, Fourth Edition (WPPSI-IV Chinese Version) (Li & Zhu 2014), and three resulting indicators — full-scale IQ, nonverbal IQ, and verbal comprehension index (VCI) — were included as the criteria for selecting children.

The children at risk for DLD⁵ were screened from kindergartens and rehabilitation centers. Specifically, 16 children came from 4 kindergartens in Jiaozuo (Henan Province), 7 children from 2 kindergartens in Guangzhou (Guangdong Province), 3 children from 1 rehabilitation center in Qingdao (Shandong Province), and 2 children from 1 rehabilitation center in Hefei (Anhui Province). Before the process of screening took place, the teachers or headmasters first selected a larger group of children, by using a screening inventory for the criteria described in Leonard (2014: 15). A child was included into the DLD group if they (i) scored 1.25 standard deviations or more below the norm for their age on at least 2 out of the 4 language indicators (the inclusion criteria for at least two indicators is based on Tomblin & Records & Zhang (1996)), namely PPVT-R, overall language, receptive language, and expressive language; (ii) had a full-scale IQ of at least 75 and a nonverbal IQ of at least 70 (Plante 1998); (iii) did not have a diagnosis of additional developmental disorders (such as autism).

The children in the ALI group were all recruited from special schools and rehabilitation centers. 14 children were recruited from the rehabilitation center in Qingdao, and the remaining 18 children from 7 distinct rehabilitation centers in Hefei. Each child received a medical diagnosis of autism by qualified hospitals, and they were admitted if they fulfilled criteria (i) and (ii) in defining DLD.

As for TDA controls, 16 children were recruited from the affiliated kindergarten of Guangdong University of Foreign Studies, 4 children from the kindergartens in Jiaozuo, and 8 children from the offspring of the staff in the rehabilitation centers in Hefei. None of the participants have symptoms of ASD, mental impairment, or physiological damage. They were included if they scored above the control mean for their age across the four language indicators obtained in the PPVT-R and RSPCLD-R/RSSCLD-R assessments.

Table 2 presents children’s scores on standardized assessments. The three groups were matched on age in months, and the DLD and ALI groups were further matched on VCI and non-verbal IQs. We used one-way ANOVA to compare non-verbal IQs, and non-parametric tools (Kruskal-Wallis tests) to compare ages⁶ and VCI due to the non-normally distributed data in the ALI group (p = .027 for ages and p = .011 for VCI, as ascertained by the Shapiro-Wilk test), and the lack of homogeneity of variance in VCI (p = .023, as confirmed by the Leven’s test).

The Kruskal-Wallis test revealed that there was no significant difference in age among the DLD, ALI, and TDA groups (H(2) = 1.802, p = .406). The one-way ANOVA showed main effect of non-verbal IQs among the three groups (F(2,85) = 12.879, p < .001). The post-hoc tests with Bonferroni corrections indicated that the mean value of non-verbal IQs was significantly different between the TDA and DLD groups (p < .001), between the TDA and ALI groups (p = .001), while no significant differences were found between the DLD and ALI groups (p = .459). As for VCI, the Kruskal-Wallis test showed that there was a significant difference among the three groups (H(2) = 43.983, p < .001). The post-hoc test with Bonferroni corrections indicated a trend for the TDA group to have a higher VCI than the DLD (p < .001) and ALI groups (p < .001), and there were no significant differences between the DLD and ALI groups (p = 1).

4.2 Experiment design

The experiment in this study was adapted, with some changes, from its English version created by van Hout & Harrigan & de Villiers (2010), in which definite NPs were tested in a short context with the presence of two pictures appearing in sequence. The process involves a child viewing a picture and the experimenter providing a description for one of the three referents in the picture. This action singled out the particular referent as the antecedent, while the other two referents were intentionally not mentioned throughout the process. The first picture then remained present and the second picture occurred, in which all the three referents changed their initial positions or actions. Subsequently, the child was prompted to respond to a question involving a definite NP. As for the question type, we included both yes-no⁷ and wh-questions.

4.3 The question types and syntactic positions feasible for the comprehension experiment

4.4 Procedures and materials

Children were tested individually with two adult experimenters: one of them (E2) operated a tablet computer to present pictures via Microsoft PowerPoint 2016, while the other (E1) sat across from the child and asked questions that are printed on a manual. During the experiment, E1 and the child were initially situated at opposite ends of a table so as to prevent E1 from seeing the pictures, and the tablet computer was placed vertically between them, with the screen facing towards children. Before the experiment began, E1 told the child that they needed to complete a collection of questions on the manual, but they were currently situated in a position that obstructs the visibility of the display. Children were then asked to look at the pictures on the screen and help E1 find the answers. As the first batch of questions that contain Dem-Cl-NPs was finished, E2 found an excuse to leave. E1 then sat next to the participant so that they can look at the screen together, and continued with the second batch of questions aimed for third-person pronouns and bare NPs. The different treatment to Dem-Cl-NPs and to the other two definite expressions is motivated by the following considerations: First, although Dem-Cl-NPs can be used either deictically or anaphorically, the deictic force is far more dominant when the referents can be mutually seen by the speaker and hearer. Therefore, during the experiment, the procedure for examining Dem-Cl-NPs should be controlled so that the speaker and hearer do not see the stimuli at the same time; otherwise, the deictic and anaphoric interpretations of Dem-Cl-NPs can appear simultaneously, leaving hearers unable to identify whether the referent in the immediate situation or in the previous discourse is intended. Second, the bare NP and third-person pronoun are prototypical anaphors, and such anaphors can only acquire occasional deictic use through extensive or symbolic pointing, albeit in a marginal way (Jiang 2016: 485). Therefore, we no longer blocked the visual common ground between the experimenter and the child when we examined bare NPs and third-person pronouns.¹⁰

Figure 1 presents the sample pictures for subject Dem-Cl-NPs in yes-no questions. There are three animate referents of the same type in each picture, and they were all designed to be equally prominent in size and style. The action that each referent is performing differs in the two pictures. As the first picture appeared on the screen, E1 said to children the introductory sentence in (6a), so that “a boy” can be singled out as the antecedent. At the end of each introductory sentence, we added an additional interrogative clause, like kàndào le ma? “Can you see (it)?” or yǒu-méi-yǒu? “Yes, or no?”, for confirming the attention of children. Having received children’s affirmative reply that the intended referent was in focus, the second picture appeared. E1 then asked the question in (6b), and waited for children to respond.

Figure 1

Sample pictures for Dem-Cl-NPs in yes-no questions.

Figure 2 shows the sample pictures for subject Dem-Cl-NPs in wh-questions. The question sentence contains the wh-expression nǎ-er “where” or gàn shénme “do what”, like the cases in (7b) and (8c):¹¹

Figure 2

Sample pictures for subject Dem-Cl-NPs in wh-questions.

Figure 3 displays the sample pictures for the increased memory condition. The design differs from Figure 2 by inserting an additional picture that introduces a distinct referent immediately following the first picture. This is done to prevent the referent referred to by the definite NP in the third picture from being adjacent to the antecedent clause, like the sentences in (8a) to (8c):

Figure 3

Sample pictures for Dem-Cl-NPs with increased memory.

Figure 4 shows the sample pictures for object Dem-Cl-NPs in yes-no questions, and each item consists of three inanimate objects of the same kind, along with one animate character. The question sentences are illustrated in (9):

Figure 4

Sample pictures for object Dem-Cl-NPs in yes-no questions.

Figure 5 presents the sample pictures for bare NPs. Similarly with Figure 4, each picture comprises three inanimate entities and one animate character. The difference lies in that, in the first picture an inanimate object was involved in the activity of the animate character, and was therefore singled out as the antecedent. In the second picture, the previously selected inanimate entity was placed in a different position by the animate character; as for the remaining two inanimate entities, one stays in situ, and one changes its position conspicuously. Since the candidate entities for the reference of the bare NP were all inanimate, the wh-question sentences only involve the wh-word nǎ-er “where”.

Figure 5

Sample pictures for bare NPs.

Figure 6 shows the sample pictures for third-person pronouns in yes-no and wh-questions. Each item comprises three different animate characters, and each referent is performing a distinctly different action. The wh-expressions in the wh-sentences involve nǎ-ér “where” and gàn shénme “do what”.

Figure 6

Sample pictures for third-person pronouns.

Sample pictures for testing post-ba, topic, and object Dem-Cl-NPs in wh-questions are illustrated in Figure 7. Each item consists of three inanimate entities of the same type as well as three different animate characters. An inanimate entity was singled out as the antecedent in the first picture, when the second picture appeared, children were required to identify the animate character who disposed the previously mentioned inanimate entity. Therefore, the wh-expressions involve only shuí “who”.

Figure 7

Sample pictures for post-ba, topic, and object Dem-Cl-NPs in wh-questions.

4.5 Coding and statistics analysis

Each testing condition has 5 question items of the same type, and the answers produced by children were coded either as 0 (incorrect) or 1 (correct), meaning that a participant can receive a maximum score of 5 for each condition. No specific time limit was established for waiting children’s responses. However, if children were clearly unresponsive due to distractions, or failed to follow the directions promptly, they were informed that the question item would be repeated once (starting from the first picture). Children were allowed to self-correct their answer only once for each item, if any, and the score was determined based on their second response.

The data were analyzed using SPSS version 24. Given the data’s non-conformity to the normal distribution, as confirmed by Shapiro-Wilk tests, and its lack of homogeneity of variances, as shown by Levene’s tests, the Kruskal-Wallis test was used for between-group comparisons, the Friedman test or the Wilcoxon signed-rank test for within-group comparisons, and the Spearman’s rank correlation test for assessing the data’s correlations.

4.6 Hypothesis and Prediction

If, as the literature suggests, children with DLD showed impaired performance on production of articles due to cognitive or pragmatic factors, we hypothesize that the DLD group should achieve significantly better performance on the comprehension of definite NPs, ideally on par with TDA children. If the DLD group also struggles with comprehension tasks, we then have reasons to believe that their impairment in using articles may stem from comorbid deficits in syntactic or semantic knowledge of definiteness, not solely by cognitive or pragmatic reasons. Given that the language profiles of DLD and ALI are reported to be substantially comparable, we predict that the DLD group may perform similarly to the ALI group in the majority of the conditions in the experiment.

5.1 Dem-Cl-NPs

Table 3 shows the raw scores of Dem-Cl-NPs among the three participant groups. The dependent variables consisted of positions (subject and object positions) and question types (yes-no and wh-questions), and the independent variables comprise the groups (DLD, ALI, and TDA). No random factors are included. The results revealed strong evidence of differences between the conditions (positions and question types) and the groups: H(2) = 52.074, p < .001 for subject Dem-Cl-NPs in yes-no questions; H(2) = 31.786, p < .001 for subject Dem-Cl-NPs in wh-questions; H(2) = 52.325, p < .001 for object Dem-Cl-NPs in yes-no questions; H(2) = 46.547, p < .001 for object Dem-Cl-NPs in wh-questions. Post-hoc pairwise comparisons with Bonferroni corrections indicated that the TDA group showed significantly better performance than the DLD (p < .001) and ALI groups (p < .001) in all conditions, while the DLD and ALI groups did not differ from each other across the conditions (p = 1 for all).

To examine within-group comparisons for each condition in the participant groups respectively, we then ran the Friedman test (nonparametric repeated measures ANOVA). The results showed that the accuracy rate of Dem-Cl-NPs was influenced by either positions or question types in the DLD group χ²(3) = 43.130, p < .001, and in the ALI group χ²(3) = 36.961, p < .001, but not in the TDA group χ²(3) = 3.370, p = .338. Post-hoc pairwise comparisons with Bonferroni corrections showed that, for the DLD group, there was a significant difference between the yes-no question and wh-question conditions in both preverbal (p < .001) and postverbal Dem-Cl-NPs (p = .023); on the other hand, there was no significant difference between preverbal and postverbal Dem-Cl-NPs in the yes-no question condition (p = 1), nor in the wh-question condition (p = .067). For children with ALI, differences were also significant for question types in both preverbal (p < .001) and postverbal Dem-Cl-NPs (p = .047), while no significant differences were found between preverbal and postverbal Dem-Cl-NPs in the yes-no question (p = 1) and wh-questions (p = .176) respectively.

Apparently, the condition of yes-no question was highly prone to error for children in the DLD and ALI groups, as they produced a large number of affirmative responses, like duì “yes”, shì “yes”, or the affirmative clauses as shown in (10a) and (10b) (E: experimenter; C: children):

Given that the second picture in each item of yes-no question conditions contains an interference referent that fulfills the description of the question sentence (and the answer should always be “no”), children who answered “yes” presumably solicited answers by simply looking at the second picture, neglecting the close connection between the Dem-Cl-NP and its antecedent, as well as the contextual connection between the two pictures. It can be summarized that in the yes-no question condition, the comprehension of Dem-Cl-NPs in the DLD and ALI groups is driven by the deictic treatment of Dem-Cl-NPs, whereas in the wh-question condition, the anaphoric interpretation of Dem-Cl-NPs can be elicited to some extent.

To summarize, there was no significant difference observed between the DLD and ALI groups across the four conditions of Dem-Cl-NPs. The comprehension of Dem-Cl-NPs in the two clinical groups did not vary with positions (though the mean accuracy of subject Dem-Cl-NPs was nearly twice that of object Dem-Cl-NPs in the wh-question condition), but with question types.

5.2 Third-person pronouns and bare NPs

Table 4 shows the raw scores of third-person pronouns and bare NPs in the three groups. A Kruskal-Wallis test showed that there was a significant difference for third-person pronouns in the yes-no question (H(2) = 51.293, p < .001) and wh-question conditions (H(2) = 44.254, p < .001). The post-hoc comparisons with Bonferroni corrections showed that, in the yes-no question condition, significant differences were found between the TDA and DLD groups (p < .001), and between the TDA and ALI groups (p < .001), while the DLD and ALI groups did not differ from each other (p = 1); as for the wh-question condition, the differences were significant between the TDA and DLD groups (p < .001), between the TDA and ALI groups (p < .001), and between the DLD and ALI groups (p = .021).

Upon scrutinizing the raw scores of third-person pronouns in the wh-question condition and the experiment pictures (see Data Accessibility), we found that the ALI group displayed considerable fluctuations in accuracy rates across the five items, while the DLD group demonstrated a noticeably higher level of accuracy and stability in their performance. This is shown in Table 5:

Similar to the situation in Dem-Cl-NPs, the comprehension of third-person pronouns in the two clinical groups was also affected by the question type, as the Wilcoxon signed-rank test revealed a significant difference in the DLD (Z = –4.312, p < .001) and ALI groups (Z = –3.990; p < .001), but not in the TDA group (Z = –1.000, p = .317).

With respect to bare NPs, between-group comparisons continued to be performed by Kruskal-Wallis tests, and results revealed that children’s performance on bare NPs differed significantly for groups, H(2) = 29.411, p < .001. Post-hoc pairwise comparisons with Bonferroni corrections indicated that the scores of bare NPs were significantly different between the TDA and ALI groups (p < .001), between the TDA and DLD groups (p = .001), but not between the DLD and ALI groups (p = .402).

So far, the subject position of wh-questions was shown to be able to elicit the best performance of definite expressions in the ALI and DLD groups, which may reflect their ideal level of knowledge about definiteness. Therefore, we used Friedman tests to compare the scores of Dem-Cl-NPs, third-person pronouns, and bare NPs in the subject position of wh-questions within each participant group, to investigate whether the three definite expressions are equally developed. The results showed that the ALI group exhibited uniform performance across all definite expressions (χ²(2) = 5.896, p = .052). The TDA group demonstrated a significant difference in overall comparisons (χ²(2) = 16, p < .001), but not in pairwise comparisons (p = .326 for Dem-Cl-NPs and third-person pronouns; p = .326 for Dem-Cl-NPs and bare NPs; p = 1 for bare NPs and third-person pronouns). The DLD group exhibited a significant difference in both overall (χ²(2) = 18.325, p < .001) and pairwise comparisons (p = .003 for Dem-Cl-NPs and bare NPs; p = .012 for Dem-Cl-NPs and third-person pronouns; p = 1 for bare NPs and third-person pronouns).

5.3 The results of Dem-Cl-NPs with increased memory, and of two indicators in WPPSI-IV

Table 6 presents children’s performance on Dem-Cl-NPs with increased memory, and the resulting scores of the working memory and processing speed indicators from the WPPSI-IV tests.

Kruskal-Wallis tests showed that there is a significant difference in Dem-Cl-NPs with increased memory, H(2) = 34.075, p < .001. Post-hoc comparisons with Bonferroni corrections indicated that there was a significant difference between the TDA and DLD groups (p < .001), between the TDA and ALI groups (p < .001), but not between the DLD and ALI groups (p = .705).

Moreover, there is a significant group effect in the working memory indicator (H(2) = 12.345, p = .002): the TDA group scored significantly higher than the DLD (p = .013) and ALI (p = .004) groups, while the DLD and the ALI groups did not differ from each other (p = 1). Finally, the processing speed index from the WPPSI-IV also demonstrates statistical significance (H(2) = 28.884, p < .001) across the participant groups. The TDA group performed significantly better than the DLD (p = .001) and ALI groups (p < .001), and no significant differences were found between the DLD and ALI groups (p = .406).

Additionally, a correlation analysis was made to ascertain the influence of nonverbal working memory and processing speed on the interpretation of definite expressions. This analysis examined the correlations between the index scores of working memory and processing speed obtained from the WPPSI-IV, and the scores of the Dem-Cl-NP, Dem-Cl-NP with increased memory, bare NP, and third-person pronoun in the subject position of wh-question condition, respectively. A Spearman’s rank-order correlation test did not indicate significant differences within each participant group, and this was evidenced by p-values greater than .05 for all pairwise associations.

5.4 Results of Dem-Cl-NPs in the post-ba and topic positions

The scores of Dem-Cl-NPs in the post-ba and topic positions are presented in Table 7. Kruskal-Wallis tests showed a significant difference in the post-ba (H(2) = 40.240, p < .001) and topic positions (H(2) = 43.526, p < .001). The post-hoc comparisons with Bonferroni corrections indicated that both the ALI and DLD groups were significantly lower than the TDA group in the post-ba (ALI-TDA: p < .001; DLD-TDA: p < .001;) and topic (ALI-TDA: p < .001; DLD-TDA: p < .001) positions, while they did not differ from each other (p = 1) in the two conditions.

Previously in Section 5.1, we found that children in the DLD and ALI groups had a discernible decline in the mean accuracy of object Dem-Cl-NPs in the wh-question condition, which contrasts sharply with the mean accuracy of subject Dem-Cl-NPs (see Table 3). This decline was caused by a sort of responses indirectly related to definiteness, or what we called “excessive reference errors”, due to which children answered wh-questions by describing all the referents on the second picture, even if the target referent was a singular one. This problem is also frequently seen in the topic and post-ba Dem-Cl-NPs. We put the description and discussion of the error in Supplementary File_2.

6.1 The asymmetric performance on yes-no and wh-questions by the ALI and DLD groups

We used both yes-no and wh-questions to examine Dem-Cl-NPs and third-person pronouns, and the question types turned out to have a significant impact on the ALI and DLD groups’ understanding of definite expressions. In the yes-no questions, both groups displayed a strong tendency to provide affirmative responses, even though the correct answer was always negative. The asymmetry is even more obvious in third-person pronouns, as the DLD group experienced a decrease in mean accuracy from 84.2% (in wh-questions) to 29.2% (in yes-no questions). Why was the discrepancy thus huge?

The lower performance observed in the yes-no questions may be attributed to children’s limited recognition of the connection between the pictures. The phenomenon of considering interconnected pictures as isolated ones has been observed in typically developing children. Emslie & Stevenson (1981) reported that, in their picture-based story-telling task, children of 2 years old used indefinite articles 20% of the time for maintaining a given referent. This occurred because the children treated previously mentioned referents as new ones when the referents re-appeared in the subsequent pictures. The same phenomenon was also found in children between the ages of 2; 6 and 5; 6 in De Cat (2011). This type of error has been called Incoherence Errors, and it occurs as children show difficulties with identifying the continuity between events/pictures (De Cat 2011: 858), which is considered a cognitive, but not a linguistic limitation (De Cat 2011: 850). Therefore, it follows that the children in the ALI and DLD groups likely responded to yes-no questions based solely on their observation of the second picture.

Another question naturally arises: now that children in the ALI and DLD groups had difficulties with the connection between pictures, why did they perform significantly better on wh-questions? When children were confronted with yes-no questions, their focus at that moment was to determine whether the event described by the question sentence exists or not. Therefore, as long as a referent in the picture fulfils the description of the question sentence, they answered yes, without considering the requirement to connect with prior discourse. On the other hand, when the children were dealing with wh-questions, their focus shifted from confirming the existence of the event to seeking specific information corresponding to wh-words such as who, what, and where. Given that each picture has three referents, it is not feasible to answer the questions solely based on the deictic interpretation of definite expressions. Consequently, the children were compelled to select an appropriate referent. To conclude, we argue that the difficulties with discourse integration and the deictic treatment of definite NPs (Karmiloff-Smith 1979: 72) contributed together to the lower performance on the yes-no questions in the ALI and DLD groups. As such type of errors ought to be categorized to the “performance-based error” (Chondrodrianni & Marinis 2015: 26) that did not accurately reflect children’s grammatical abilities, the subsequent analysis of children’s knowledge of definite expressions will focus solely on their performance in the wh-question condition.

6.2 The knowledge of definite expressions in the ALI and DLD groups

We first look at children’s morpho-syntactic knowledge regarding definite NPs. Despite neither the ALI nor DLD groups achieving target-like performance on any of the three definite expressions, the referent that a child located to answer the wh-questions always stemmed from the pictures provided, and most importantly, the reference of the definite expressions was made to a singular referent by children¹², though there were two interfering referents in each picture simultaneously. At this point, the Dem-Cl-NP, bare NP, and third-person pronoun in the children’s language most likely project as a DP, since D relates to the referentiality or deixis of a nominal expression (Willim 2000: 321), and it has a “singularizing”/ “individualizing” function (Cheng & Sybesma 1999). As the [+referential] feature is present, the DP layer is active and visible for interpretation.

According to the DP hypothesis (Abney 1987), the D head is the location that hosts definite determiners. In Mandarin, pronouns base-generate in D since they are definite determiners followed by an elided NP complement (Postal 1970; Elbourne 2001; Huang et al. 2009: 297); bare NPs appear in D to receive definite readings by head movement (Huang et al. 2009: 301); and demonstratives directly base-generate in D (Huang et al. 2009: 296). If the three NP types all project into a DP, the ALI and DLD groups should, in theory, attain a high rate of accuracy evenly across the three definite expressions (as already confirmed by the TDA children in our study). However, the results did not occur in the children with ALI and children with DLD. In the ALI group, the comprehension of bare NPs (70.6%), third-person pronouns (61.8%), and Dem-Cl-NPs (59.4%) did not differ from each other significantly, but none of them achieved target-like (if we use the 90 percent criteria threshold in Brown (1973)). In the DLD group, children’s performance on third-person pronouns (84.2%) and bare NPs (83.6%) was approaching target-like, but they scored significantly lower on Dem-Cl-NPs (62.8%), presenting uneven performance on definite expressions. Therefore, the problem seems to lie more in children’s semantic knowledge of definite expressions, than in the morpho-syntactic knowledge of the DP structure.

van Hout & de Ree & de Ree (2008) and van Hout & Harrigan & de Villiers (2010) provide evidence that typically developing children (Mean = 4) interpret definite NPs differently than adults. In their experiments, children were shown a picture with a set of identical referents. One referent was explicitly mentioned in the verbal discourse, while the others were not. Children tended to consider all the referents present in the picture as definite, whereas adults accepted the referent that had been previously mentioned as definite. From this perspective, definiteness derived from the immediate situation (the visual context) and that from the verbal discourse was unranked in children’s language. Although the same phenomenon was not observed in the TDA group in our study (probably due to the older mean age of our participants), the analysis can still be used to explain the below-average performance of the DLD and ALI groups. When a child (in the two clinical groups) was processing a definite NP, two sources of semantic interpretations of definiteness emerge simultaneously, despite the antecedent already established in the previous discourse: one is the interpretation motivated by uniqueness, which resists having an antecedent (Geist 2021: 22); the other is prompted by familiarity, which necessitates an antecedent (Roberts 2003: 294). The ALI and DLD groups demonstrated a preference for anaphoric interpretations of the three definite expressions (at least 59% of the time), but there were still quite a few occurrences where they relied on the visual context (the immediate situation).

It should be noted that the condition of uniqueness is not met when the children wrongly chose the referent based on the immediate situation. This is because, apart from the referent that had been individualized in the verbal discourse, there were two other referents left unmentioned and therefore remained in the immediate situation. One anonymous reviewer referred us to Wexler (2011: 39, footnote 6),¹³ who posited that children have the same lexical entry as adults but have problems computing the uniqueness requirement specified in the presupposition of definite articles, and that uniqueness is a weak/variable property for them. Therefore, children seem to employ a more flexible semantic strategy towards the uniqueness condition than adults, and the definite article can still be used even when there are multiple referents in the relevant situation, not necessarily a unique one. This phenomenon can be frequently seen in the DLD and ALI groups in our study, but not in the TDA group. So far, can we conclude that the non-target-like performance of the DLD and ALI groups on the comprehension of definite expressions is solely due to their insufficient semantic knowledge of definite expressions? That is, they cannot fully identify which particular situations (the immediate situation or previous discourse) definiteness stems from. It may be still inadequate to draw a conclusion if we re-consider the following literature in which no visual context is provided (therefore collapsing the immediate situation).

Children with DLD (5;9–9;8) in Blom & Vasic & Baker (2015) and Chondrogianni & Marinis (2015), and TD children (3;6–5;5) in Schafer & de Villiers (2000) produced definite articles more accurately in bridging conditions (that simply involve uniqueness) than in anaphoric conditions (that necessitate familiarity). Schafer & de Villiers hold a structural point of view, suggesting that the apparent DP in children aged 3;6 to 5;5 is in fact a functional projection theP that merely expresses [±unique] feature, and the real DP which involves [±hearer] feature is projected only when children can take listener’s point of view into account. This idea exactly aligns with the more recent views on the dichotomy of two levels of definiteness within a DP. Cheng & Heycock & Zamparalli (2017) map the unique and anaphoric distinction onto two syntactic levels, a structurally lower w(eak)DP (predicative/argumental) and a structurally higher s(trong)DP with a pronominal index (argumental). Based on Cheng et al. (2017), Geist (2021) further defines the wDP and sDP into the u(niqueness-based)DP and a(naphoric)DP. Jenks & Konate (2022) propose that in the DP that expresses anaphoric interpretations, an individual-denoting category D or indexP occupies the [Spec, DP] position, whereas in the plain unique DP there is no such an index in the specifier.

Taking all the evidence into accounts, the growth of DP structure in children seems to indicate a bottom-up developmental process from uDP¹⁴ (immediate situation/bridging) to aDP (anaphoric). The DP construction is most likely in place in the DLD and ALI groups, but its inner layers stay in a transitional stage where the semantic components of uniqueness and familiarity intertwine to some extent. Children in the two clinical groups probably already knew the semantic concepts of the two types of definiteness, but from time to time, they cannot block the semantic interference from the structurally lower layer (uDP) when they were processing the semantic property of the structurally higher one (aDP or indexP). The comprehension of definite expressions in the two clinical groups is likely to be impacted by the unstable knowledge of the syntax-semantics interface within the DP.

Since definiteness in Mandarin generates subject-object asymmetry, the syntactic and semantic knowledge regarding definite NPs in the DLD and ALI groups we discussed above is only applicable to the subject position. In the object, post-ba, and topic positions, however, the comprehension of definite NPs seems to be interfered by factors indirectly related to definiteness, which has been discussed in Supplementary File_2.

In summary, our findings support the theory proposed by Modyanova (2009: 118), who followed the spirit of Wexler (2011), that the development of determiners in children is predicted by their overall grammatical abilities, rather than solely by pragmatic factors.

6.3 The third-person pronoun as a potential marker to the distinction between ALI and DLD

The difference between the two clinical groups lies in the comprehension of third-person pronouns. Previous studies found that the production of pronouns has always been problematic for individuals with ASD, which is characterized by the avoidance of using pronouns or ambiguous use of pronouns, even in children with ALN (see the literature review in the section “Pronouns and ASD” in Nagano & Zane & Grossman (2021: 1565)). However, the comprehension experiment conducted by Nagano & Zane & Grossman (2021) yields divergent results, showing that children and adolescents with ASD (10 to 17 years old) were capable of comprehending third-person pronouns like the controls, though “they may struggle to apply this knowledge during production” (Nagano et al. 2021: 1574). Note that the individuals with ASD in their study show normal language and IQs, thus having ALN. Besides, the eye-tracking experiment in the same research revealed that the ASD and control groups showed different looking patterns, suggesting that the same interpretation of third-person pronouns in the two groups may be achieved through different processing paths. Nagano et al. (2021: 1575) noted that the processing difference observed in individuals with ASD may lead to their non-target performance in production tasks.

In our study, the children with ALI showed clear difficulties with the comprehension of third-person pronouns. As we have discussed in Section 6.2, the below-average performance on definite expressions in the ALI and DLD groups is due to the fact that the knowledge of syntax-semantics interface inside the DP is not well-developed. On Dem-Cl-NPs and bare NPs, the DLD and ALI groups performed similarly. This can be attributed to the assumption that processing Dem-Cl-NPs is more difficult due to its complex forms and dual functions (the anaphoric and deictic usages), while processing bare NPs is less demanding due to its simpler form. However, the performance by the ALI group no longer conforms to this assumption when it comes to the third-person pronoun. The DLD group continued to present near-target-like performance because third-person pronouns also share a simple form, but the ALI group did not perform any better on third-person pronouns than they did on Dem-Cl-NPs, and the results of Dem-Cl-NPs and third-person pronouns in the ALI group overlap greatly on the following statistical indicators: the interquartile range (4–2) and median (3) of third-person pronouns were completely identical to those of Dem-Cl-NPs, while the mean scores were approximately the same (Dem-Cl-NP: 2.97 (59.4%) vs. third-person pronoun: 3.09 (61.8%)). We argue that there must be an additional factor at work that constrains a better interpretation of third-person pronouns in the ALI group.

According to Nagano et al., the discrepancy in the processing paths between the ASD and the control groups arises from the shortcoming in working memory and attentional allocation abilities within the ASD group. In addition, they found that children with ASD demonstrated significantly facilitated processing in the condition where the “full NPs” (the combination of definite article and common noun) were used to refer to a previously introduced referent, in comparison to the condition where third-person pronouns were used. They noticed that the preference for processing “full NPs” observed in the ASD group coincided with the behavior of the patients with Alzheimer disease who processed narratives faster when “full NPs”, instead of pronouns, were used, which is attributed to the working memory defect¹⁵ in the Alzheimer group (Almor 1999).

The analysis in Nagano et al. (2021) provides valuable insights into how the ALI group in our study treated bare NPs and third-person pronouns differently. First, according to Cheng & Sybesma (1999) and Huang et al. (2009: 283), definite bare NPs in Mandarin are equivalent to [the + NP] combination in English. Therefore, bare NPs are the counterpart of “full NPs” described in Nagano et al. (2021) As the processing of “full NPs” was more facilitated than the processing of third-person pronouns in the ALI group, the performance on bare NPs therefore derived no significant differences as compared to that of the DLD group. Second, as for third-person pronouns, despite the simplicity of their form, the ALI group scored significantly lower than the DLD group due to the combined factors of (i) the difficulties with the knowledge of syntax-semantics interface inside the DP structure, and (ii) the processing difference (that impacts the interpretation of third-person pronouns); on the other hand, the DLD group was affected simply by the first factor. Therefore, the combined impact of linguistic deficits and processing differences could potentially result in reduced interpretation of third-person pronouns in the ALI group. What is the underlying factor responsible for the processing difference in the ALI group — the deficiencies in working memory, or in attentional allocation abilities, or in both?

First, we believe that working memory cannot be the leading factor. We examined children’s non-verbal working memory in the subtest of WPSSI-IV, and the condition of Dem-Cl-NPs with increased memory. The results indicated that there were no significant differences between the two clinical groups in the two tasks. Therefore, it would be less reasonable to conclude that non-verbal working memory only comes into play for children with ALI.

We then considered the potential deficiency in attentional allocation abilities in the ALI group. As presented in Table 5, the exceptionally high level of accuracy observed in Items 1 and 3 may be ascribed to the saliency of the target referent, which happens to be the most active or conspicuous referent in the picture. In the Items 2 and 5, the target referent and the interfering referents display equal levels of prominence, therefore leading to decreased accuracy rates. The reason for Item 4 receiving the lowest accuracy rate is due to the interfering referent being the most prominent one. In comparison, the DLD group exhibited noticeably higher levels of accuracy and stability in their performance across the items. It follows then that although the referent introduced by the indefinite NP has been made salient in the discourse, children in the ALI group still tended to reallocate their attention to the referents that are most appealing to them, rather than the one that corresponds to the antecedent. The occurrence of this phenomenon in the third-person pronoun, as opposed to the bare NP, can be attributed to the lack of lexical content in the third-person pronoun, because it is a pure anaphor and its interpretation in verbal discourse is entirely dependent on the antecedent. Therefore, the lack of lexical content in the third-person pronoun renders the ALI group more vulnerable to being distracted by extralinguistic distractions.

In Mandarin, the equivalents of “he”, “she”, and “it” all share an identical phonological form, and it remains unclear whether this overlap would render the third-person pronoun more difficult to comprehend for Mandarin-speaking children. In this case, it would be interesting to look at children who acquire languages in which the third-person pronoun makes a distinction between gender and animacy. As the attentional allocation ability may be the point where the DLD and ALI groups diverged in their comprehension of third-person pronouns, future research may consider the role of referents’ conspicuousness as variables, and an eye-tracking experiment is strongly recommended.