PREVIOUS PRODUCTION STUDIES ON MANDARIN RCS

As we said, results on the production of Mandarin RCs are mixed. Cheng (Reference Cheng1995) examined 27 children aged from 3 years, 6 months (3;6) to 6;3, using an elicitation task, and observed a subject preference (i.e., subject RCs vs. object RCs, 42% vs. 19%). In contrast, Su (Reference Su2004) tested 40 children aged from 5;0 to 6;5 and 31 adults, using the elicitation methodology devised by Hamburger and Crain (Reference Hamburger, Crain and Kuczaj1982) and Crain, McKee, and Emiliani (Reference Crain, McKee, Emiliani, Frazier and de Villiers1990), and did not find any preference (e.g., subject vs. object RCs, 84% vs. 88%, for children aged 5;0–5;6). These results should be considered with some caution because they were obtained with two trials for each sentence type. It has been recommended that in elicited production task at least four tokens of each sentence type should be used (Thornton, Reference Thornton, McDaniel, McKee and Cairns1996).

Hsu et al. (Reference Hsu, Hermon and Zukowski2009) tested 23 children aged from 4;0 to 6;5 and 10 adults, using an elicited production task with the help of picture presentations modeled after Hamburger and Crain (Reference Hamburger, Crain and Kuczaj1982) and developed by Zukowski (Reference Zukowski2009). They found that both children and adults produced more subject RCs than object RCs (children, 83.5% vs. 38.9%; adults, 95% vs. 79.5%). However, some of their materials were arguably problematic due to the properties of the verbs: in the subject RC elicitation task, some verbs could be used intransitively, and this could have inflated the subject advantage. For instance, (2a) was used to depict two girls in the experimental picture: one girl was singing and another girl was drawing. Then, a mouse appeared in front of one of the two girls and the experimenter asked, “Which girl is the mouse watching?” to elicit the subject RC in (2b). Here, changge and huahua could either be interpreted transitively to describe a generic activity, like “sing a song” or “draw a painting,” or intransitively as in “sing” and “draw.” The intransitive verb only requires one argument and assigns the agent theta role to its argument. In contrast, the transitive verb requires two arguments and assigns the agent and patient theta roles to its arguments. The theta-role assignment of intransitive verbs is possibly easier than that of transitive verbs. Therefore, the use of optionally intransitive verbs may enhance the production of subject RCs. By contrast, the elicitation of object RCs always involved transitive verbs, and was thus more demanding.

1. (2)

   

Chen and Shirai (Reference Chen and Shirai2014) reported an object RC advantage in early child speech, by analyzing the spontaneous production of four monolingual Mandarin-speaking children (aged 0;11–3;5). In total, they observed 34 subject RCs and 78 object RCs. Some of their examples of subject RCs are given in (3a) and of object RCs in (3b). That object RCs emerge early and are used more frequently during the period investigated in the Mandarin-learning children's speech is in contrast with the findings concerning the early spontaneous production of English-speaking children (a subject RC preference found in Diessel & Tomasello, Reference Diessel and Tomasello2000) and of Japanese-speaking children (no asymmetry reported in Ozeki & Shirai, Reference Ozeki, Shirai, Matsumoto, Oshima, Robinson and Sells2007). Although Chen and Shirai's (2014) study uncovered an early stage of Chinese children's RC production, it has some limitations. The object advantage established was merely on a numerical basis, and the complexity of the clauses was not examined. When we look at their examples, we observe lots of simplified structures, such as (3c), which is a headless RC with an intransitive verb.

1. (3)

   

To sum up, the aforementioned studies on the production of Chinese RCs are mixed and present some limitations in one way or another, thus making it hard to evaluate which preference holds in Mandarin and which is the source of difficulty in the production of RCs.

A relevant feature of children's RCs across languages is the use of a resumptive strategy, mostly involving resumptive pronouns and less frequently resumptive NPs (noun phrases; e.g., Catalan: Gavarró et al., Reference Gavarró, Cunill, Muntané and Reguant2012; English: Pérez-Leroux, Reference Pérez-Leroux1995; French: Labelle, Reference Labelle1990, Reference Labelle1996; Italian: Guasti & Cardinaletti, Reference Guasti and Cardinaletti2003). A resumptive pronoun occurs inside the RC and is related to the relative head, in that it encodes its grammatical function inside the RC, as in (4a) (example from Cheng, Reference Cheng1995); a resumptive NP is a double of the relative head and occupies the extraction site, as in (4b) (example from Hsu et al., Reference Hsu, Hermon and Zukowski2009).

1. (4)

   

Both resumptive pronouns and resumptive NPs are attested in Chinese children's production data. Cheng (Reference Cheng1995) found that children produced resumptive pronouns in subject RCs (3.2%, 4 out of 122) and in object RCs (7%, 4 out of 57); they only produced resumptive NPs in object RCs (15.8%, 9 out of 57). Su (Reference Su2004) did not differentiate resumptive pronouns and resumptive NPs and observed resumptive elements in subject RCs (3.2%, 2 out of 63) and in object RCs (8.8%, 6 out of 68). Hsu et al. (Reference Hsu, Hermon and Zukowski2009) only found RCs with resumptive NPs, with fewer instances in subject RCs (3.4%, 10 out of 296) than in object RCs (18.3%, 38 out of 208) in the child production; adults only used resumptive NPs in object RCs (1.9%, 3 out of 160). To summarize, Chinese children produce resumptive elements less often in the subject extraction site (4a) than in the object extraction site (4b). Adults also produced some resumptive NPs.

Resumptive pronouns have been discussed in the literature. Gu (Reference Gu2001) stated that resumptive pronouns are obligatory in oblique object RCs and are optional in other positions (see Ning, Reference Ning2008). Similarly, in several Romance languages, resumptive pronouns in RCs are optional/obligatory in substandard varieties (see Guasti & Cardinaletti, Reference Guasti and Cardinaletti2003, and references cited therein). Although resumptive NPs are produced by Chinese children and adults, there is thus far no systematic investigation of their status in Mandarin, and it is taken for granted that RCs with resumptive NPs are prohibited. In the light of this, it is important to clarify this issue in Mandarin.

Therefore, two issues are left open to inquiry: first, which preference is displayed by Mandarin-speaking children in the production of subject/object RCs in comparison to children speaking head-initial RCs; and second, what is the status of resumptive NPs for Mandarin-speaking children and adults, when compared to the optionality of resumptive pronouns in Mandarin.

THEORETICAL APPROACHES TO RC ACQUISITION

In the sentence-processing literature, the subject/object asymmetry is captured in terms of principles according to which a chain between a moved element and its trace tends to be established as soon as possible (e.g., the active filler hypothesis, Frazier & Flores D’Arcais, Reference Frazier and Flores d’Arcais1989; the minimal chain principle, De Vincenzi, Reference De Vincenzi1991). According to this view, a subject chain is favored because a subject trace can be hypothesized almost immediately having seen the relative head. By contrast, in the case of object extraction, it is assumed that the parser attempts to posit a subject trace, but cannot because an overt subject is present. Thus, the object chain is more difficult to build because the embedded subject blocks the parser's attempt to posit the trace; in a sense, it acts as an intervener. The notion of intervention in a dependency between a moved element and its trace figures prominently in the linguistic literature, and this has led Friedmann et al. (Reference Friedmann, Belletti and Rizzi2009) to attempt a unification between ideas stemming from the linguistic literature and those stemming from the processing literature. In particular, these authors have proposed that relativized minimality is implicated in the explanation of the subject/object asymmetry in child language (see also Adani, van der Lely, Forgiarini, & Guasti, Reference Adani, van der Lely, Forgiarini and Guasti2010; Belletti, Friedmann, Brunato, & Rizzi, Reference Belletti, Friedmann, Brunato and Rizzi2012; Belletti & Rizzi, Reference Belletti, Rizzi, Piattelli-Palmarini and Berwick2013; Contemori & Belletti, Reference Contemori and Belletti2013; Guasti, Stavrakaki, & Arosio, Reference Guasti, Stavrakaki and Arosio2012; for criticism, see Goodluck, Reference Goodluck2010).Footnote ²

Relativized minimality (RM; Rizzi, Reference Rizzi1990, Reference Rizzi and Belletti2004) is a locality constraint on dependencies within a sentence. In configuration (5), a local structural relation cannot hold between X and Y if Z is similar to X, that is, if there is a qualified intervener in the dependency between X and Y.

1. (5)

   1. a. X . . . Z . . . Y

   2. b. Z intervenes between X and Y if and only if Z c-commands Y and Z does not c-command X (Rizzi, Reference Rizzi and Belletti2004, p. 225).

Although RM was formulated as a linguistic principle, its insights are very similar to principles formulated within the sentence-processing literature to account for the subject/object asymmetry as we sketched earlier. In this respect, Friedmann et al.'s (2009) proposal is that a dependency is harder or ungrammatical if there is a qualified intervener between the antecedent and its gap. More specifically, these authors argued that the difficulty in RC acquisition is modulated by the nature of the relative head and of the NP in the embedded clause. Under a raising analysis, the relative head is attracted by a complex attractor [+R, +NP], where R and NP express the relative feature and the lexical restriction feature, respectively. See Example (6), with the schematic featural representation given below each clause.

1. (6)

   1. a. the child_i that t_i draws the teacher

      [+R, +NP][+NP]

   2. b. the child_i that the teacher draws t_i

      [+R, +NP][+NP]

In the subject RCs in (6a), the lexically restricted relative head does not cross over the lexically restricted element in the clause, that is, the embedded object, whereas, in object RCs (6b), the lexically restricted relative head has to cross over another lexically restricted element, that is, the embedded subject. In object RCs, the relative head bears [+R, +NP] features and the embedded subject bears a [+NP] feature, with the latter having a subset of the features of the former. To consider them distinct as required by RM, one has to compute the subset–superset relation, but by hypothesis, limited computational resources sometimes prevent younger children from making this computation. Accordingly, a RM violation arises, and children have greater trouble in understanding and producing object RCs than subject RCs. Note that children are not violating RM; they are just unable to compute the subset–superset relation.

The RM approach makes precise predictions for the acquisition of Mandarin RCs, namely, it anticipates that object RCs should be also harder than subject RCs as in the head-initial RCs. Following Huang (Reference Huang1982) and Aoun and Li (Reference Aoun and Li2003), let us assume the hierarchical structure of Mandarin RCs in (7).Footnote ³ Intervention is determined on the basis of c-command, so that in a subject RC, the relative head (xiaopengyou “child”) c-commands its trace without the intervention by the embedded object (laoshi “teacher”) as shown in (7a). In contrast, as shown in (7b), in an object RC the embedded subject (laoshi “teacher”) intervenes between the relative head (xiaopengyou “child”) and its trace. To reiterate, under the raising analysis, the embedded subject has a [+NP] feature and thus it has a subset of the features of the relative head. According to the RM approach sketched above, children have problems in computing the subset relation, and thus, Mandarin object RCs are expected to be more difficult to acquire than subject RCs, as happens in other languages.

1. (7)

   1. a. Mandarin subject RCs

      

   2. b. Mandarin object RCs

      

It could be argued that structural intervention does not matter and what counts and influences children's acquisition is linear intervention. Contrary to structural intervention, which is established in terms of c-command, linear intervention is based on precedence.

Linear and structural factors cannot be separated in head-initial RCs, for example, English RCs, but can be teased apart in Mandarin RCs. Consider the English RCs in (6) again. In subject RCs, the embedded object the teacher does not intervene between the relative head the child and its trace both linearly and hierarchically. In object RCs, the subject the teacher intervenes between them both linearly and hierarchically. Both linear and structural factors predict that, in languages such as English, subject RCs should be easier than object RCs, and this is what has been reported.

Now consider the Mandarin RCs in (1) and their hierarchical structure in (7) again. In the case of Mandarin subject RCs, as exemplified in (1a), the embedded object (laoshi “teacher”) linearly intervenes between the relative head (xiaopengyou “child”) and its trace; however, as shown in (7a), the object does not structurally intervene between the relative head and its trace. By contrast, in the case of Mandarin object RCs, as exemplified in (1b), the embedded subject (laoshi “teacher”) does not linearly intervenes between the relative head (xiaopengyou “child”) and its trace, but, as shown in (7b), it structurally intervenes between them. If only structural factors matter, we would expect object RCs to be problematic for children; but if only linear factors matter, we would expect the opposite.

Franck, Lassi, Frauenfelder, and Rizzi (Reference Franck, Lassi, Frauenfelder and Rizzi2006), based on the results of a series of production experiments on subject–verb agreement, pointed out that both hierarchical and linear intervention create interference, but they do so to different degrees. Structural intervention has stronger effects than linear intervention. If this is the case, we anticipate not only a subject advantage in Mandarin, as we said earlier, but also some mild difficulties with subject RCs in Mandarin, due to linear intervention.

In summary, the RM approach predicts that object RCs should be harder than subject RCs because of structural intervention. If linear intervention also matters, we should find some penalty in the production of subject RCs.

In this study we examine the production of RCs by children and adults with four goals in mind: (a) to investigate which acquisition pattern holds in the production of Mandarin RCs; (b) to examine how the acquisition pattern develops across ages; (c) to characterize children's difficulties, if any, by looking at the pattern of nontarget structures and errors; and (d) to show how structural and linear factors influence the production of Mandarin RCs. To these purposes, we ran an elicitation experiment with children and adults (Experiment 1) and a grammaticality judgment task with adults (Experiment 2). The latter experiment is the first grammaticality study focusing on Mandarin RCs with resumptive NPs and attempting to tease apart prescriptive grammar from the grammar of spoken Chinese in the domain of RCs.

EXPERIMENT 1

Results

The experiment yielded a total of 2,500 responses from children and 400 responses from adults. In both cases, half resulted from the elicitation of subject RCs and half from the elicitation of object RCs. The next subsection lays out the analysis of target RC responses, and the following subsection briefly describes nontarget RC responses and then reports the analysis of RCs with resumptive NPs, followed by a comparison with the results in previous studies. The last subsection briefly reports other responses.

The analysis of target RC responses

Of the 1,480 target responses from children, 985 were subject RCs and 495 were object RCs. Of the 320 target responses from adults, 200 were subject RCs and 120 were object RCs. Table 1 shows the percentages, the raw scores, the means, and the standard deviations of target responses in both conditions for all groups. Subject RCs were more frequently produced than object RCs by both children and adults, and the proportion of target responses increased with age, at least from 3 to 6 years of age.

Table 1.

Percentages, raw scores (N), means, and standard deviations of correct response in each age group

These observations are confirmed by the statistical analysis. We used R (R Development Core Team, 2013) and lme4 (Bates, Maechler, & Bolker, Reference Bates, Maechler and Bolker2012) to perform a series of analyses. The p values were obtained by likelihood ratio tests of a full model without the effects against a model with the effects. We fitted responses to a mixed-effects model with sentence type (i.e., subject RCs vs. object RCs) and age (i.e., 3-, 4-, 5-, 6-, 7- and 8-year-olds and adults) as fixed factors and subjects and items as random factors. The reference categories were object RCs for the sentence type factor and 3-year-olds for the age factor. The sentence type factor predicted the ability to produce RCs, that is, object RCs were significantly less likely to be produced than subject RCs, χ² (1) = 50.87, p < .001; Wald Z = 14.62, p < .001.Footnote ⁴

As expected, age yielded a significant effect, χ² (6) = 97.55, p < .001. By changing the reference categories, we compared each age group with other age groups. Table 2 reports a summary of the results of the statistical analysis. We observed a robust increase in the production of target responses at age 4 and at age 6. The performance of 4-year-olds significantly differed from that of 3-year-olds, but not from that of 5-year-olds. The performance of 5-year-olds significantly differed from that of adults, while the performance of 6-year-olds did not show a significant difference from that of adults. A crucial point is that, given that the eldest child tested in previous studies was 6;5 (e.g., Hsu et al., Reference Hsu, Hermon and Zukowski2009; Su, Reference Su2004) and that 6-year-olds were not distinguished from 5-year-olds, our study is the first to report that the production of target responses does not differ between adults and 6-year-olds.

Table 2.

Summary of the age factor in the mixed-effects model for producing target RCs

Note: N = 2,900, SD of subjects = 2.41, SD of items = 2.21, log likelihood = –960.2.

The analysis of nontarget RC responses

Of the 196 nontarget RC responses from children, 62 were passive RCs, 41 RCs with resumptive pronouns, and 93 RCs with resumptive NPs. Of the 80 nontarget RC responses from adults, 35 were passive RCs and 45 RCs with resumptive NPs. Note that all nontarget RC responses from adults were obtained in the object RC elicitation task.

Passive RCs were only observed in the object RC elicitation: 4 passive RCs were produced by 5-year-old children and 58 were produced by 8-year-old children. This is consistent with the cross-linguistic finding that children tend to be comfortable with the use of passives at a relatively late age (e.g., Hirsch & Wexler, Reference Hirsch, Wexler, Deen, Nomura, Schulz and Schwartz2006). Adults produced 35 passive RCs. Given that the amount of passive sentences was not abundant, we only conducted a chi-square test by comparing the observed number of passive RCs with the expected number. The observed + expected data were summarized in one contingency table with one degree of freedom. Although there were descriptively more passives produced by 8-year-olds than by adults, the difference was not significant, χ² (1) = 3.15, p > .05.

Similarly, RCs with resumptive pronouns were only observed in the object RC elicitation. The only resumptive pronoun that children produced was wo “me.” Fifteen such sentences were produced by 4-year-olds, 21 sentences by 5-year-olds, and 5 sentences by 8-year-olds. This is in line with what has been found across languages (see Contemori & Belletti, Reference Contemori and Belletti2013; Gavarró et al., Reference Gavarró, Cunill, Muntané and Reguant2012; Guasti & Cardinaletti, Reference Guasti and Cardinaletti2003; Labelle, Reference Labelle1990, Reference Labelle1996; Pérez-Leroux, Reference Pérez-Leroux1995). The children in our study did not produce subject RCs with resumptive pronouns, which has been reported in previous studies (e.g., Cheng, Reference Cheng1995), but note that the number of resumptives in subject RCs was low even in Cheng (Reference Cheng1995).

RCs with resumptive NPs were produced by children both in the subject RC (1.3%, 13 out of 998)Footnote ⁵ and in the object RC elicitation (11.8%, 80 out of 678), but only in the object RC elicitation by adults (22.5%, 45 out of 200). They were observed in every age group, but the majority were produced by younger children: 31 by 4-year-olds and 37 by 5-year-olds. Relatives with resumptive elements decreased from age 6 on (2 by 6-year-olds, 15 by 7-year-olds, and 7 by 8-year-olds).

To see which factors influence the production of RCs with resumptive NPs, the results were fit to a mixed-effects model with extraction site (i.e., subject vs. object) and group (i.e., children vs. adults) as fixed factors and subjects and items as random factors. The reference categories were object RCs with resumptive NPs for the extraction site factor and children for the group factor. We found a significant effect of extraction site: object RCs with resumptive NPs were significantly more common than subject RCs with resumptive NPs, χ² (1) = 53.63, p < .001; Wald Z = –9.38, p < .001.Footnote ⁶ No significant effect of age was found, χ² (1) = 1.51, p = .22.

Our data are in agreement with previous studies as far as children's production is concerned; for example, in Hsu et al. (Reference Hsu, Hermon and Zukowski2009), 4- to 6-year-old children produced 3.4% resumptive NPs in the subject condition (10 out of 296 utterances) and 18.3% in the object condition (38 out of 208). However, they differ with respect to the adults’ production. In our study, adults produced 22.5% object RCs with resumptive NPs (45 out of their 200 utterances in the object RC elicitation), while in Hsu et al.'s study, this production was 1.9% (3 out of 169 utterances). We argue that this discrepancy is partly due to the different tasks employedFootnote ⁷ and the smaller number of participants in Hsu et al.'s study (10 adults vs. 20 adults in ours).

We also examined the individual performance to see if every participant consistently produced resumptive NPs. We counted the number of participants who produced subject and object RCs with resumptive NPs at least once. In total, there were 8 (out of 125) children who produced resumptive NPs in subject RCs, 28 (out of 125) children who produced resumptive NPs in object RCs, and 9 (out of 20) adults who produced resumptive NPs in object RCs. In addition, we counted the number of participants who produced them at least five times (out of 10). Regarding subject RCs with resumptive NPs, 1 (out of 20) 7-year-olds produced them more than five times. Regarding object RCs with resumptive NPs, 3 (out of 24) 4-year-olds, 4 (out of 21) 5-year-olds, and 4 (out of 20) adults produced them at least five times. This observation indicates that some individuals allow resumptive NPs freely.

The analysis of other responses

To investigate what children do when they fail to produce RCs, we examined the different types of responses (including errors) in our study. Table 3 and Table 4 provide the percentages and number of other responses in each age group. Adults did not produce any sentence in this category.

Table 3.

Percentages and total number of different errors in each age group in the subject RC elicitation

Note: V-NP, NP-V-NP, and Reduced RCs, a declarative sentence with a null subject, a declarative sentence, and subject RCs with omission of de, respectively.

Table 4.

Percentages and total number of different errors in each age group in the object RC elicitation

Note: NP-V, a declarative sentence with a null object.

First, younger children typically answered with an NP,Footnote ⁸ an NP de NP, and a declarative sentence with a null subject or a null object (V-NP and NP-V, respectively). As we can see in the two tables, the NP and NP de NP responses were more common in the object RC elicitation than in the subject RC elicitation. In contrast, the NP-V responses in the object RC elicitation were less frequent than the V-NP responses in the subject RC elicitation.

We fitted responses to a mixed-effects model with condition (i.e., subject vs. object RC elicitation) and age (i.e., 3-, 4- and 5-year-olds) as fixed factors and subjects and items as random factors. As for the NP responses, we found a significant main effect of condition, χ² (1) = 19.09, p < .001; Wald Z = –4.59, p < .001,Footnote ⁹ and age. There was a significant decrease from age 3 to age 4, χ² (2) = 38.15, p < .001; Wald Z = –4.16, p < .001,Footnote ¹⁰ but not from age 4 to age 5 (Wald Z = –0.63, p = .53). As for the NP de NP responses, a significant effect of condition was found, χ² (1) = 13.29, p < .001; Wald Z = –3.97, p < .001,Footnote ¹¹ but not of age, χ² (2) = 0.86, p = .65. As for the V-NP and NP-V responses, a significant effect of condition, χ² (1) = 35.21, p < .001; Wald Z = 7.12, p < .001,Footnote ¹² and age were found. There was a significant decrease from age 3 to age 4, χ² (2) = 30.57, p < .001; Wald Z = –3.90, p < .001,Footnote ¹³ but not from age 4 to age 5 (Wald Z = –1.07, p = .29).

Second, declarative sentences (NP-V-NP) were very common, especially when object RCs were elicited, and especially in children from 3 to 7 years of age. The statistical analysis supported this observation. We fitted responses to a mixed-effects model with condition (i.e., subject vs. object RC elicitation) and age (i.e., 3-, 4-, 5-, 6-, 7- and 8-year-olds) as fixed factors and subjects and items as random factors. Children were significantly more likely to use declarative sentences when they failed to produce object RCs than subject RCs, χ² (1) = 55.91, p < .001; Wald Z = –13.58, p < .001.Footnote ¹⁴ As for the age factor, the only significant decrease was observed from age 7 to age 8, χ² (5) = 29.68, p < .001; Wald Z = –3.09, p < .01.Footnote ¹⁵ No significant decrease was observed across other age groups.Footnote ¹⁶

Third, as shown in Table 3, subject RCs with omission de were very rare, with only eight sentences being observed in the subject RC elicitation and only being produced by children at 5 and 6 years of age.

Fourth, as shown in Table 4, reversal RCs were observed in every age group, but they were very rare in contrast to what are reported in research on other languages (e.g., Catalan: Gavarró et al., Reference Gavarró, Cunill, Muntané and Reguant2012; Italian: Guasti & Cardinaletti, Reference Guasti and Cardinaletti2003).

In sum, the results suggest that subject RCs are targetlike sooner than object RCs, and the response pattern of the 6- and 7-year-olds differed from that of the 8-year-olds (e.g., the declarative response was still common in the former age group, but not in the latter age group), although their production of target RCs did not differ.

EXPERIMENT 2

Our results in Experiment 1 showed that adults produced RCs with resumptive NPs. Although this possibility is banned in prescriptive grammar (Cheng, Reference Cheng1995; Hsu et al., Reference Hsu, Hermon and Zukowski2009), our data suggest that they may be licit in spoken Chinese. In Experiment 2, we tested this possibility using a grammaticality judgment task. In particular, we attempted to determine whether native speakers of Mandarin Chinese accept RCs with resumptive NPs.

Results

Table 5 reports the distribution of responses for each sentence type. As expected, the acceptance rate of RCs with gaps was higher than that of RCs with resumptive NPs. However, adults accepted RCs with resumptive NPs to some extent.

Table 5.

The distribution (%) of responses for each sentence type

^aWith resumptive NPs.

We fitted responses to the model with resumption (i.e., RCs with gaps vs. RCs with resumptive NPs) and extraction site (i.e., subject vs. object) as fixed factors and subjects and items as random factors. The reference categories were RCs with gaps for the resumption factor and object for the extraction site factor. We found a significant effect of resumption, χ² (1) = 72.21, p < .001,Footnote ¹⁷ which indicates that RCs with resumptive NPs were less likely to be accepted as compared to RCs with gaps. As for the extraction site, we did not find a significant difference, χ² (1) = 1.64, p = .20. Therefore, when we considered the individual performance, we did not distinguish between subject RCs and object RCs.

We examined the individual performance for judging RCs with resumptive NPs. There were 9 (out of 80, i.e., 11.3 %) participants who accepted them all the time (including “completely acceptable” and “acceptable” responses), and 8 (out of 80, i.e., 10%) participants who rejected them all the time (including “completely unacceptable” and “unacceptable” responses). In addition, there were 51 (out of 80, i.e., 63.8%) participants who accepted this structure from one to five times.