Abstract
This article investigates the binding properties of the Korean reflexive caki. Korean caki allows a local antecedent, a long-distance sentence-internal antecedent, and (unusually) an extrasentential antecedent. Two experiments were conducted with Korean-speaking child participants (mean age = 5;8; age range = 5;1–6;4) and adult controls. The first tested local binding versus long-distance intrasentential binding, and the second tested extrasentential binding. The results show, first, that the children allowed both a local and a long-distance antecedent for caki, with a preference for the long-distance reading (counter to many claims that children show a robust preference for local antecedents crosslinguistically). Second, the children allowed caki to refer to an antecedent in a preceding sentence. These results indicate that, at an early stage, Korean children have acquired adult-like knowledge of the complex properties of caki, including what might be considered the computationally taxing option of extrasentential binding.
Introduction
Typologically, reflexives subject to self-reference come in at least two kinds: those whose antecedent must be local (e.g., within the same clause), and those whose antecedent may occur in a higher clause (but typically still within the same sentence). The former, referred to as local reflexives, are typically composed of two morphemes (polymorphemic), as in the English him/herself, while the latter, referred to as long-distance (henceforth, LD) reflexives, are typically composed of a single morpheme (monomorphemic; Pica’s generalization [Pica, 1987]).
In Korean, there are two different types of reflexives: caki-casin is polymorphemic, and, in line with Pica’s generalization, is a strictly local reflexive; the monomorphemic reflexive, caki, is a long-distance reflexive that allows both LD and local antecedents, with a preference for LD antecedents. Between the two reflexives, caki is particularly noteworthy due to an unusual feature: It allows not only intrasentential binding, but also extrasentential binding, in which the antecedent is outside the sentence in which caki occurs (Cho, 1996; Han & Storoshenko, 2012; Kang, 1988, among many others). Consider the choice of antecedents for caki and for caki-casin at the sentence level, as exemplified in (1) and (2): (1) Yang-i
i
[So-ka
j
caki-lul
i>j
kuly-ess-tako] sheep- malhay-ss-eyo. say- ‘Sheep
i
said that Cow
j
drew self
i>j
.’
In (1), caki may refer to either the local c-commanding antecedent So ‘Cow’, or the LD antecedent Yang ‘Sheep’, though it has been reported that adults prefer the LD interpretation (Choi & Kim, 2007; Kang, 1998, among many others). In contrast, caki-casin can only refer to the clause-mate So ‘Cow’, not the matrix subject Yang ‘Sheep’, as shown in (2): (2) Yang-i
i
[So-ka
j
caki-casin-lul*i/j kuly-ess-tako] sheep- malhay-ss-eyo. say- ‘Sheep
i
said that Cow
j
drew self*i/j.’
Second, the property of allowing reference to an antecedent that occurs outside the sentence (i.e., extrasentential binding) makes caki unusual among reflexives. With regard to this property see the context sentence and the target sentence of the two reflexives in (3) and (4): (3) Context: Yang
i
-i kkalkkaltaymye wus-ess-eyo. sheep- ‘Sheep
i
laughed joyfully.’ Target Sentence: So
j
-ka cakii>j-lul wuskikyey kuly-ess-ki cow- ttaymwun-i-lay-yo. because-be- ‘(It is) because Cow
j
drew self
i>j
comically.’
In (3), caki allows a reading in which the clause-mate nominal, So ‘Cow’, in the target sentence serves as the antecedent to caki. Importantly, it also allows the extrasentential antecedent (henceforth, ES antecedent), Yang ‘Sheep’, as well, which is the unusual feature of caki. However, in (4), the Korean reflexive caki-casin limits the binding to the local antecedent. Thus, the local subject So ‘Cow’ is the only eligible antecedent for the reflexive in the target sentence: (4) Context: Yang
i
-i kkalkkaltaymye wus-ess-eyo. sheep- ‘Sheep
i
laughed joyfully.’ Target Sentence: So
j
-ka caki-casin*i/j-lul wuskikyey kuly-ess-ki cow- ttaymwun-i-lay-yo. because-be- ‘(It is) because Cow
j
drew self*i/j comically.’
In this article, we investigate the L1 acquisition of the Korean reflexives, focusing on caki. The other Korean reflexive, caki-casin, serves as a control to provide a benchmark against which to compare the participants’ preferences for caki. Specifically, we investigate whether children allow intrasentential binding (Experiment 1) and extrasentential discourse binding (Experiment 2) for caki. It should be noted that the possibility of extrasentential binding in the adult grammar is the most unusual property of caki, and it has never (to our knowledge) been investigated in child language before. Therefore, this article makes a unique contribution to our understanding of child language. We also examine whether children show the same preferences for caki that adults do.
Acquisition of reflexives
The acquisition of polymorphemic reflexives has been investigated for decades, and the consensus is that their referential properties are acquired by about age three (Chien & Wexler, 1991; Clackson, Felser, & Clahsen, 2011; Jakubowicz, 1984; Otsu, 1981, among many others). While not as much is known about the acquisition of monomorphemic reflexives, several important points stand out. One is that there are conflicting reports on whether children acquiring monomorphemic reflexives initially go through a stage in which they favor a local reading of such reflexives (contradicting Pica’s generalization; Chien & Wexler, 1991), although adults typically prefer the LD interpretation.
Chien, Wexler, and Chang (1993) tested 80 Mandarin-acquiring children aged three to eight using a yes-no judgment task on the polymorphemic reflexive taziji and the monomorphemic reflexive ziji. The former allows only a local reading, while the latter allows both local and LD readings. They found that both adults and children preferred the local reading of ziji (the local properties of taziji were acquired early). In the case of Danish, Jakubowicz and Olsen (1988) found that both adults and older children (age: 9;0–9;6) strongly preferred the LD reading of the Danish monomorphemic sig, but young children (age: 3;0–3;5) strongly preferred the local reading, selecting it 93% of the time.
In another study of children’s interpretation of monomorphemic reflexives, Okabe (2008) conducted a Truth Value Judgment Task (henceforth, TVJT; Crain & Thornton, 1998) with the Japanese LD reflexive zibun ‘self’, which is known to allow both LD and local antecedents but to prefer the LD antecedent. The results showed that Japanese children (age: 4;6–6;11) had a slight bias toward the local antecedent over the LD antecedent. However, it is important to note that the adult controls in the study also preferred the local antecedent over the LD antecedent; these findings for adults raise doubts about the experiment’s validity.
Turning to Korean, H. Lee and Wexler (1987) conducted an experiment using an act-out task (i.e., a ‘Simon-says’ game) with biclausal hearsay-construction test sentences, as shown in (5a) and (5b): (5) a. Koyangi-ka malha-ki-lul [Mary-ka caki-lul cat- manchesstta-ko kuraess-ttae]. touch- ‘According to the cat, Mary
i
said that she/he/it
j
touched self
i/j
.’ b. Holangk-ka malha-ki-lul [Mary-ka caki-lul tiger- manchesstta-ko kuraess-ttae]. touch- ‘According to the tiger, Mary
i
said that she/he/it
j
touched self
i/j
.’ (H. Lee & Wexler, 1987, p. 5, (4&5))
In the experiment, two puppets, a female cat and a male tiger, were presented to the participants, and the participant’s name was used as the embedded subject (i.e., ‘Mary’ was replaced by the name of the participant). Thus, there were two conditions: Same gender (e.g., either a female cat and a female participant, or a male tiger and a male participant) and different gender (e.g., either a male tiger and a female participant, or a female cat and a male participant). They found that adults selected the LD interpretation of caki 60% of the time, showing the now-familiar preference for the LD interpretation of monomorphemic reflexives. In contrast, children (age range = 3;7–6;6) showed the opposite preference, preferring the local reading 71.7% of the time in the same gender condition and 60% of the time in the different gender condition at the age of 3;7. The preference for the local antecedent increased to almost 100% for the children at the age of 6;6. The findings thus seem to provide evidence for children’s preference for the local reading of caki.
In a later study, however, H. Lee (1990) conducted a TVJT using test sentences like the one in (6): (6) Grandpapa
i
-ka Pinocchio
j
-ka [cakii>j -lul Grandfather- kariki-ess-ta]-ko saengakha-ess-ta. point to- ‘Grandfather
i
thought that Pinocchio
j
pointed to self
i>j
.’ (H. Lee, 1990, p. 139, (6a))
With this type of test sentence, Lee found that monolingual Korean children between the ages of 6;6 and 8;0 actually preferred the LD interpretation of caki. This finding added a new wrinkle to research on children’s interpretation of monomorphemic reflexives, which had otherwise reported a ubiquitous preference for the local interpretation. In light of these findings – the preference for the local antecedent reported by H. Lee and Wexler (1987) and the preference for the LD antecedent observed by H. Lee (1990) – we cannot yet draw conclusions about Korean children’s preference for the antecedent for reflexive caki.
The wrinkle introduced by H. Lee’s (1990) results also shows up in Icelandic: Adults show a reliable preference for the LD interpretation (over 90%), and young children also prefer the LD interpretation. Hyams and Sigurjónsdóttir (1990) tested 105 children (age: 2;6–6;0) and found that even at the youngest age, the children preferred the LD interpretation. Sigurjónsdóttir (1992) later found that this preference was partially due to the verb choice in their experimental task. That is, for example, when the reflexive sig occurs in a subjunctive or infinitive clause with a verb such as raka ‘shave’, it takes either an LD antecedent or a local antecedent. In contrast, when verbs such as gefa ‘give’ are used with sig in a subjunctive or infinitive clause, sig favors the LD antecedent more strongly than it does with raka-type verbs (e.g., greiða ‘comb’ and klóra ‘scratch’). Nonetheless, even with this verb-based difference, it does not seem to be the case that children uniformly prefer the local reading of Icelandic sig, in contrast to previous findings from Mandarin, Danish, Japanese, and (except for Lee’s TVJT results) Korean. Table 1 summarizes prior studies on monomorphemic reflexives, showing whether they found a local or LD preference.
Results of studies on monomorphemic reflexives in child and adult language.
Note: LD = long-distance.
This discrepancy in the literature is one of the two foci of the current article. The other focus involves one of the more unusual features of caki. Han and Storoshenko (2012) discussed several properties that set Korean caki apart from LD reflexives in Germanic and in other East Asian languages (see also Cho, 1996; Gill, 1999; Kang, 1988, among many others). While each of these properties warrants attention, in this article we focus on the claim that caki may take antecedents that occur not just outside of the binding clause (i.e., LD antecedents), but outside the sentence entirely.
We therefore ask two research questions. The first involves sentence-level binding: Do children acquiring Korean favor a local or LD reading of caki? To answer the question, we postulate the following hypothesis: If Korean children are like adults in the interpretation of the Korean reflexive caki, they will accept intrasentential binding when selecting an antecedent for caki. Our second research question addresses the issue of discourse binding: Do Korean children allow caki to refer to an antecedent outside the sentence? We hypothesize that if Korean children are adult-like in their interpretation of the Korean reflexive caki, it will be evident in the children’s choice of an antecedent in extrasentential binding for caki. This possibility has never, to our knowledge, been investigated before.
These research questions are based on the assumption that caki is a rule-governed LD reflexive. However, there is another view: caki may be a logophoric pronoun that is subject to discourse-pragmatic conditions. Specifically, for example, when caki is the source of information – the logophoric role of SOURCE (Sells, 1987) – it can be discourse-bound (E. Lee, 2017). While this distinction in the properties of caki is an important issue, we will leave it unresolved and focus on whether Korean children interpret caki in adult-like ways.
Experiment 1: Local and LD binding of caki and caki-casin
Participants
Twenty-nine Korean-speaking children (mean age = 5;8; age range = 5;1–6;4) and 28 adult controls participated in the experiment. Data from four children were excluded from the analyses because their accuracy rates on filler items were within the range of chance. The remaining 25 children and all 28 adult controls answered at least 10 of the 12 filler questions correctly. All children were attending a kindergarten in Seoul, Korea at the time of testing.
Materials and procedure
In an attempt to avoid biases in participants’ referential preferences due to verb types, a norming study, an acceptability judgment task, was conducted with a 6-point Likert scale. The rationale behind the norming study is as follows. A verb like stab typically involves one person stabbing another, not one person stabbing him/herself, although the latter is possible. A reflexive object in a sentence with stab (e.g., ‘John said that Bill stabbed self’) would favor the LD reading, because the local reading is pragmatically odd (although not impossible). Likewise, a verb like shave favors a local reading, as it is pragmatically unusual (although not impossible) for someone to shave someone else. The norming study included 31 adult native speakers of Korean (all college students in Korea) and tested 10 verbs that we thought formed a range in terms of whether they semantically favor the LD or the local reading. 1 From the results of this norming study, we selected four verbs that tested neutral in terms of their antecedent preference: kulita ‘draw’, kalikhita ‘point out’, anta ‘hug’, and kkocipta ‘pinch’.
By varying the four verbs, the two context types, and the two reflexive types, we created 12 experimental stories in four conditions. In the experiment, each verb was unevenly distributed, being used three times per condition in order to reduce the burden on the young participants. For instance, in List 1, the verb kalikhita ‘point out’ was used with caki (LD context), caki (local context), and caki-casin (LD context). In List 2, the same verb was used with caki (local context), caki-casin (LD context), and caki-casin (local context). In List 3, the same verb was used with caki-casin (LD context), caki-casin (local context), and caki (LD context), and in List 4, it was used with caki-casin (local context), caki (LD context), and caki (local context). In all, there were 26 stories (2 warm-ups, 12 experimental stories, and 12 fillers). The fillers used different sentence structures than the critical items, and they did not include reflexives. The critical and filler items were presented in random order.
To examine children’s grammatical knowledge on Korean reflexives, we used a picture-based TVJT (Crain & Thornton, 1998). 2 The participants saw a series of drawings depicting a story, and then heard a sentence describing those drawings. The children were asked to judge whether the sentence truthfully described the story by responding either ‘true’ or ‘false’. The advantage of this task type is that the children’s binary reactions are based on an event generated within the context of the story, which might help keep the children more at ease than a more test-like task would. In addition, the story might keep the children more interested in the task.
The experimental procedure was as follows. First, the children were asked whether they knew the meaning of caki and it was found that they did. Next, in a warm-up exercise, two animated characters were introduced to the children: a narrator, who would say the test sentences describing the stories, and a well-known Korean character called Rupy, who was blindfolded. In each trial, after the narrator read the story to the children, the blindfolded Rupy appeared and said what happened in the story, using the test sentences. The children’s job was to help Rupy by saying if Rupy’s sentence truly described what happened in the story. On the other hand, in order to provide the children an explanation for why a particular interpretation that is almost true ends up being false, elements allowing plausible dissent were added to the stories. Figure 1 shows an example LD antecedent context. For this context, if a child interpreted the reflexive as taking the LD antecedent, s/he was expected to accept Rupy’s sentence; if a child interpreted the reflexive as taking the local antecedent, s/he was expected to reject Rupy’s sentence.
Sample pictures (with texts) of LD antecedent context with caki/caki-casin for intrasentential discourse binding.
In this example, the test sentence is Duck said that Bee pointed to self, and the picture shows Bee pointing to Duck (the subject of the matrix clause). On a local reading, this sentence is false, while on an LD reading, this sentence is true. The task varied the type of reflexive pronoun (caki vs. caki-casin) and so when the sentence contained caki-casin, we expected both adults and children to reject this test sentence. However, when the sentence contained caki, a close-to-categorical result was expected from the adult data, given Grice’s (1975) cooperative principle that people try to avoid ambiguity in pursuit of effective communication. More specifically, given that caki-casin is only felicitous locally, we expected adults to assume that if the local meaning had been meant, caki-casin would have been used. The preference for the LD antecedent for caki facilitates the avoidance of ambiguity, although both options are ostensibly grammatical in the adult language.
As for the children, we expected them to behave like the adults if they had an LD preference for caki. However, if children have a local preference for caki, they might reject the test sentence. Children with a local preference might initially access the local reading, and those that are able to revise their initial interpretation will then access the LD reading, and accept the sentence. Those that are unable to revise that initial reading will reject the sentence.
Results
Experiment 1 explores whether children have a local or long-distance preference for caki with special reference to the antecedent interpretation of caki-casin in the LD antecedent contexts and the local antecedent contexts. Figure 2 shows the antecedent interpretation of caki and caki-casin depending on the contexts by both adult controls and children.
Antecedent interpretation of caki and caki-casin depending on context for intrasentential binding by adults and children.
Looking at the adults’ results, the fact that (i) reflexive type and (ii) context type are significant predictors comes from the very low acceptance rates in the condition with caki-casin in the LD antecedent context. Overall, the acceptance rate for the conditions with caki is generally higher than it is for the conditions with caki-casin; and the acceptance rate for the conditions with the local antecedent context is higher than it is for the conditions with the LD antecedent context. The other factor, (iii) interaction between context type and reflexive type, is also a significant predictor because the interpretations of adults for caki and caki-casin are different in LD versus local contexts.
Turning to the children, their results are broadly similar to those of the adults: The effects of reflexive type and context type are due to the low acceptance rate of the condition with caki-casin in the LD antecedent context. The interaction effect comes from the fact that in the LD antecedent context, the child participants allowed caki but generally disallowed caki-casin.
More specifically, the adult controls accepted sentences with caki at a high rate in LD-context trials, while accepting sentences with caki-casin at a very low rate in LD-context trials. This is as expected: Adult native speakers prefer caki to refer to an LD antecedent, but do not allow caki-casin to refer to an LD antecedent. In the local antecedent context, the adults strongly accepted caki-casin, and they also judged caki to be acceptable, but less so. This too is an expected result. In the case of the children, they showed a high rate of acceptance of sentences with caki in LD-context trials, while their acceptance of sentences with caki-casin was low in LD-context trials. This shows that the children disfavored an LD antecedent with caki-casin, but favored an LD antecedent with caki. In the local antecedent context, the children found caki-casin acceptable, and they also found caki acceptable, but not to the same degree as caki-casin. In sum, both children and adults exhibited a strong preference for the local antecedent with caki-casin, and they both also accepted caki in both local and LD contexts, although both populations favored the LD antecedent context for caki.
The statistical analyses on the performance of the two participant groups, children and adults, employed an ordinary logit model based on logistic regression using SPSS 18.0. In Experiment 1, the independent variables are reflexive type and context type, while the dependent variable is the type of antecedent interpretation (as indicated by participants’ true/false responses). For the categorical dependent variable, a ‘false’ response was coded 0 and a ‘true’ response was coded 1. Table 2 shows a summary of the ordinary logit model for the data from adults and children.
Summary of the ordinary logit model for adults and children: intrasentential binding.
Taken as a whole, Table 2 shows that (i) reflexive type, (ii) context type, and (iii) interaction between context type and reflexive type are significant predictors (p < .05 for adults and children) in this regression model. 3
Experiment 1 establishes a baseline for intrasentential binding, and shows that children seem to readily allow caki to be bound by an LD antecedent. This is in contrast to some of the previous literature on monomorphemic reflexives, as noted earlier. However, it is consistent with the findings from H. Lee’s (1990) Korean TVJT study, as well as findings on Icelandic (Hyams & Sigurjónsdóttir, 1990; Sigurjónsdóttir, 1992). We turn now to the second contribution of this article: an experiment investigating the acquisition of the extrasentential binding properties of caki.
Experiment 2: Extrasentential (ES) binding of caki and caki-casin
Experiment 2 investigates whether native Korean-speaking adults and children allow extrasentential discourse binding for caki and caki-casin.
Participants
Experiment 2’s participants were 29 children who had not participated in Experiment 1. However, three of them performed poorly on the fillers, and their data were excluded. As a result, the analysis is based on the data of 26 native Korean-speaking children (mean age = 5;7; age range: 4;9–6;3) and 31 adult controls.
Materials and procedure
The experimental design and procedure used in Experiment 2 were identical to those of Experiment 1 (i.e., 12 critical items with four verbs, half involving caki and half involving caki-casin). The verbs and the characters used in Experiment 1 were used in Experiment 2 as well. In Experiment 2, the context allowed for an ES antecedent and a clause-mate (local) antecedent. A sample set of stories involving the embedded verb kulita ‘draw’ is presented in Figure 3.
Sample pictures (with texts) of local antecedent context with caki/caki-casin for extrasentential discourse binding.
Figure 3 shows a local antecedent context. For example, the first picture introduces the two primary characters, along with some contextual information (they are drawing a picture in art class). The second picture shows both Penguin and Goat, and depicts Penguin as good at drawing. However, pictures 3 and 4 describe a reversed situation, in which Goat’s picture is better than that of Penguin. At the end of the story, the blindfolded Rupy appears and says two sentences. The first sentence (Goat laughed joyfully.) establishes the ES antecedent (Goat), and the next sentence (It was because Penguin drew self comically.) contains a reflexive (either caki or caki-casin).
Note that the conditions for plausible dissent are satisfied in this experiment as follows. In Figure 3, first, Rupy is blindfolded, and so is unable to be sure who painted what. Second, both the local and the ES referent can be considered as potential antecedents at different points in the story. The ES antecedent is possible in frame 2, when the two characters are considering drawing each other, while the local antecedent is shown to be possible in frame 4. Thus, both readings are plausibly true at some point in the story, giving Rupy a plausible reason to produce either the matching test sentence or the mismatching test sentence.
Results
The methods of data analysis were the same as those used in Experiment 1. The antecedent interpretation of the two Korean reflexives by adults and children in Experiment 2 is summarized in Figure 4.
Antecedent interpretation of caki and caki-casin depending on context for extrasentential discourse binding by adults and children.
For the adult data, the finding that (i) reflexive type and (ii) context type are significant predictors comes from the lower acceptance rate in the condition with caki-casin and the ES context. As for (iii), the interaction between context type and reflexive type is also a significant predictor because the adults interpreted caki and caki-casin differently in ES and local contexts. Turning now to the children, (i) reflexive type and (ii) interaction effect between context type and reflexive type are the significant predictors for the same reasons as for adults. Looking at the results of Experiment 2 specifically, the adults accepted caki both in the ES antecedent context and in the local context. In contrast, the adults rarely accepted caki-casin in the ES antecedent context, but usually accepted it in the local context, showing that the adults strongly dispreferred ES binding for caki-casin. The children accepted caki both in the ES condition and in the local condition. 4 In the case of caki-casin, they did not accept it in the ES condition, but they did accept it in the local condition. Comparing the results of the children with those of the adults, it is noteworthy that the adults showed an LD antecedent preference for caki in both intrasentential binding and extrasentential binding, while the children showed an LD antecedent preference for caki only in the case of intrasentential binding. In addition, the children did not allow caki-casin in the LD context.
For the statistical analyses, an ordinary logit model based on logistic regression was again employed. The independent variables are reflexive type and context type, while the dependent variable is the type of response (true vs. false). A summary of the ordinary logit model for all the data from the adults and children is given in Table 3.
Summary of the ordinary logit model for adults and children: extrasentential binding.
Table 3 shows that in the case of the adults, (i) reflexive type, (ii) context type, and (iii) interaction between context type and reflexive type are the significant predictors (p < .05) for this regression model. As for the children, however, (i) reflexive type and (ii) interaction between context type and reflexive type are the significant predictors (p < .05) for this regression model. 5 Note that the context type is not a significant predictor for the children’s data.
Discussion and conclusion
The goal of the present study was to assess children’s knowledge of intrasentential binding and extrasentential binding as shown in their interpretation of the Korean reflexive caki. The key questions we addressed were whether children favor a local or LD reading of caki at the sentence level, and whether children allow caki to refer to an antecedent outside the sentence at the discourse level. In order to answer the questions, we conducted experiments testing the interpretations of caki in both intrasentential binding and extrasentential binding contexts.
This study provides two notable contributions to our understanding of the acquisition of caki. First, the results of Experiment 1 revealed that children acquiring Korean do not have a local preference for caki, which appears to support the study’s first hypothesis. Furthermore, the children’s preference for the LD antecedent of caki (i.e., 77% for the LD antecedent vs. 59% for the local antecedent) provides positive evidence that children, like adults, may understand Grice’s (1975) maxim that effective communication entails avoiding ambiguity when possible. More specifically, Korean children appear to know that caki is felicitous with both local binding and LD binding but to prefer the LD antecedent, which avoids ambiguity and thus allows more effective communication. Of special note regarding the results of Experiment 1 is that unlike children acquiring Danish and Chinese reflexives, the Korean-speaking children’s preference for a nonlocal interpretation is consistent with the findings reported by Hyams and Sigurjónsdóttir (1990) for Icelandic. Second, and more interestingly, the Korean reflexive caki is unusual crosslinguistically in that it allows binding across a sentence boundary (Han & Storoshenko, 2012). The results of Experiment 2 showed that Korean children appear to have acquired the discourse-binding properties of caki, accepting an antecedent outside the sentence at very similar rates to the adult controls. These results seem to support the study’s second hypothesis, suggesting that extrasentential binding is a possibility that Korean children acquire fairly early and easily.
Although Experiment 1 found that the Korean children preferred the LD antecedent of caki in intrasentential binding, it does not resolve the discrepancy in the literature. A few studies have found an LD preference (e.g., in Korean and Icelandic), while others have shown a local preference (e.g., in Mandarin, Danish, and Japanese); while the situation calls for research on more languages, the research thus far suggests that it might not be possible to reach a single conclusion. The question about the source of the conflicting results on the acquisition of reflexives remains. Input may be among the factors. However, at least, in the case of Korean, there may be little relevant input. The CHILDES database (MacWhinney, 2000) includes 177 files of Korean adult and child data, within which there are very few examples of caki in the speech of caregivers to children (0.03%, 107 instances in 388,526 utterances; Ryu, Horie, & Shirai, 2015). This suggests that the property of the Korean reflexive caki as an LD reflexive must be learned on the basis of very limited data.
As for extrasentential discourse binding, Korean is considered a discourse-oriented language. In such languages, discourse information plays an important role, for example, by licensing topic drop (Huang, 1984). Sentence processing studies support this generalization, showing that in the choice of antecedent, discourse cues are stronger than sentence-internal morphosyntactic cues for speakers of discourse-oriented languages like Korean, but not for speakers of sentence-oriented languages like English (Kwon & Sturt, 2013).
The findings on extrasentential binding also raise a question as to how Korean children acquire discourse binding so early. One possible account is this: It is known that Danish sig acts like a pronominal anaphor with the features [+anaphoric, +pronominal] rather than a pure anaphor (Bickerton, 1987; Chomsky, 1982, among many others). Likewise, it has been argued that Korean caki has the features [+anaphoric, +pronominal], which allow both LD binding and discourse binding (H. S. Lee, 1987). If caki has these particular features, it is plausible to assume that Korean children know the reflexive can accept the discourse-bound option as well as the locally bound option by the age of 4;9–6;3 years. This suggestion is in line with the findings of other studies that have examined Korean children’s sensitivity to discourse. For example, Kim, O’Grady, and Cho (1995) reported that Korean 4-year-olds show better comprehension of scrambled OSV sentences when there is a felicitous context, replicating Otsu’s (1994) study with Japanese children. In addition, Song and Fisher (2005), who investigated the effect of discourse prominence on how Korean children find pronoun antecedents, found that children as young as 3 years of age can interpret a pronoun as co-referential with the subject of the preceding sentence.
On the other hand, it is possible to think of the results of Experiments 1 and 2 in terms of the subset principle (Berwick, 1985; Wexler & Manzini, 1987), which is commonly invoked to account for the acquisition of binding phenomena in various languages. For example, in the case of the acquisition of LD reflexives, the subset principle ‘predicts that children learning Chinese or Korean should take the local antecedent for the reflexive ziji or caki as unmarked’ (Chien et al., 1993, p. 230), and that therefore ‘there would be a stage in which the Korean child only chooses the local value for the antecedent of the reflexive’ (H. Lee & Wexler, 1987, p. 4). In other words, the subset principle predicts a developmental pattern in which children initially only seek the reflexive’s antecedent in a local domain. However, if a language permits reflexive antecedents in the LD domain, children will later come to accept the LD interpretation, after their knowledge of the binding domain is changed through input.
However, the subset principle struggles to account for the data presented here. That is, the principle states that children first acquire the most restrictive, smallest grammar compatible with the input (the subset) and then broaden that grammar to a superset when faced with positive evidence in the input that the target grammar is different. Thus, in order for the subset principle to apply, there should be relevant instances of the target grammar in the input. However, the Korean-speaking children in Experiment 1 showed a nonlocal preference on par with that of the adults, although, as mentioned, relevant input is very rare (0.03% in the CHILDES database Korean corpus). Furthermore, the subset principle is subject to grammatical constraints rather than discourse preferences. It is thus not clear what the subset principle would predict in this case. In sum, the results of the present study clearly demonstrate that at least Korean children do not start with the local interpretation for the LD reflexive caki in accord with a subset grammar.
Besides the main findings we have discussed, an additional finding of the two experiments is worth noting. The adults showed a long-distance antecedent preference for caki both intrasententially and extrasententially; the children differed from the adults in preferring long-distance interpretations within the sentence but not across sentence boundaries. That is, the children allowed caki to have ES and local antecedents at almost equal rates (i.e., 69% vs. 65%) in the extrasentential binding experiment. Hence, a question arises: If the children have an LD antecedent preference for caki as attested in the sentence-binding experiment, why did they not show that preference in the extrasentential binding experiment?
Considering this evidence of different preferences for LD antecedents in intrasentential binding and extrasentential binding, we speculate that intrasentential binding is preferable to extrasentential binding in terms of processing, given that the distance between the reflexive and the LD antecedent is shorter in intrasentential binding than in extrasentential binding. Independent evidence in support of this proposal comes from research that has demonstrated children’s processing limitations (Trueswell, Sekerina, Hill, & Logrip, 1999) in tracking discourse (Reinhart, 2006; Reuland, 2011; Sekerina, Stromswold, & Hestvik, 2004). Such processing limitations could in turn contribute to a greater preference for intrasentential antecedents for caki on the part of children, even if that means selecting a local antecedent, contrary to the usual tendency. However, importantly, to the extent that a number of factors such as world knowledge, the parser, and grammatical knowledge may have contributed to the children’s behavior in these experiments (Conroy, Takahashi, Lidz, & Phillips, 2009), this remains a speculative note.
Last but not least, the present study has some limitations, and questions remain regarding the acquisition of caki. Namely, although the reflexive caki-casin was used as a control to confirm that children could reject an LD interpretation, a follow-up study should include a control to confirm children’s ability to reject the local interpretation. Also, a study that asked the children follow-up questions in order to find out why they reject local binding would be illuminating.
In sum, the children’s interpretation of caki allowed either the LD antecedent (main clause subject in a biclausal sentence; extrasentential discourse topic) or the local antecedent, whereas their interpretation of caki-casin allowed only the local antecedent both in sentence-level binding and in discourse-level binding. Hence, this study has demonstrated that the children involved in all the experiments knew that the two reflexives were distinct from one another in terms of locality.
