Is Referent Reintroduction More Vulnerable to Crosslinguistic Influence? An Analysis of Referential Choice among Japanese–English Bilingual Children

Abstract

This study aims to examine whether a crosslinguistic influence (CLI) is exerted on the referring expressions of the spoken narratives of Japanese–English bilingual children in different discourse contexts. Thirteen early bilingual (school-age) children separately presented Japanese and English narratives for a wordless picture book and a speechless video clip. Further, seven Japanese and nine English monolingual children participated as controls. The linguistic devices that the children adopted to introduce, reintroduce, and maintain the topic were compared with those of their monolingual controls to detect any CLI. As predicted, CLI for English on Japanese was observed but not vice versa. In Japanese, bilinguals utilize significantly more noun phrases (NPs) compared with their monolingual counterparts. More crucially, this was observed only in the referent reintroduction context, indicating that only discourse contexts that require the integration of much pragmatic information may be vulnerable to English influence. Null forms are barely utilized in English narratives; thus, no influence from Japanese was observed. We present the referential choice patterns in the elicited spoken narratives of bilingual school-age children acquiring an under-researched language pair. By controlling for the discourse context, we demonstrate that CLI is more likely to manifest in the reintroduction context. These findings offer additional evidence for the interface and structural overlap hypothesis, further highlighting the criticality of considering information structure as an influencing condition.

1. Background

1.1. Referring Expressions in Narratives and Information Structure

The relationship between information structure and the appropriate referential choice (referring expression) in narrative discourse has been well documented in the discourse analysis literature (Chafe 1976, 1994; Du Bois 1987; Gundel et al. 1993; Nakamura 1993). Regarding the selection of an appropriate referential form and the creation of referential cohesion in a narrative, thereby enabling effective communication of an event being described, the major persisting argument is that discourse participants manage shared and unshared information among themselves and that the speaker is constantly assessing the listener’s knowledge or the accessibility of the referent (the extent to which the referent is activated in the listener’s mind). Extant investigations within the psycholinguistics framework further support that referential forms are determined by the activation level of a referent, proving that the more activated (in terms of memory and discourse conditions), less salient forms are utilized (Ariel 1990; Arnold 2010; Torregrossa et al. 2019).

The information structures of narratives have been analyzed by assuming the following three-way division of the information status of the referents, and this seems to best capture the different levels of information statuses in a stretch of discourse: new (inactive; a newly introduced referent to the discourse), accessible (semi-active; a previously introduced referent, replaced by a new one but remaining in the speaker’s/listener’s consciousness and being half-active accordingly), and given (active; a referent introduced and maintained in the discourse) (Chafe 1994). Many studies on different languages indicated the existence of a universal tendency in which new information was typically expressed via salient forms (lexical forms), whereas given information was expressed via reduced forms (pronominal or null forms) (Chafe 1994). Researchers have also identified the language-specific rules that speakers must follow, selecting pronominal or null/ellipted forms to express given information depending on the language syntax (e.g., Allen 2000; Baker and Greenfield 1988; Clancy 1997). Generally, extant studies indicated that the speakers of a language must assess the degree of activeness of a referent in the listener’s mind or consider the other’s viewpoint and select the appropriate language-specific referential form to effectively communicate.

The extant literature on the narrative development of young monolingual children reveals that the acquisition of the ability to use referring expressions in an adult-like manner requires a long and gradual process and that the acquisition period varies among the different referring contexts. The existing crosslinguistic investigations acknowledge that the ability to mark given referents (maintenance) is the earliest to develop (it can develop before school age (3–4 years old), whereas the ability to mark a new (introduction) or accessible (reintroduction) referent is not normally acquired until later development stages, usually during school years (9–10 years old). For example, Karmiloff-Smith (1985), Nakamura (1993), and Hickmann et al. (1996) reported that referent maintenance develops early, during childhood. Nakamura’s (1993) study revealed that Japanese-speaking children utilize reduced forms (null forms in Japanese) to maintain a referent reliably in an adult-like manner. Hickmann et al. (1996) also reported that English-, French-, German-, and Mandarin Chinese-speaking children maintained the referent via language-specific linguistic devices (definite noun phrases (NPs) or pronominal forms) from around the age of four. Studies also show that children do not mark new referents in a reliable manner until halfway through their school years—young children typically use definite NPs instead of indefinite ones in English (Karmiloff-Smith 1985) or employ NPs without the subject marker ga (bare NP) or ellipsis in the subject position in Japanese (Nakamura 1993)—indicating that assuming knowledge in the listener (or the activation level) can represent a challenging task for young children when introducing a referent. Additionally, an adult-like usage of referent reintroduction emerges at a later stage of childhood (Bamberg 1987; Karmiloff-Smith 1985; Orsolini et al. 1996) because the reintroduction context features introduction and maintenance: the referent is given information that does not appear immediately before the one in focus (Orsolini et al. 1996) or is accessible in Chafe’s (1994) classification. Therefore, selecting an appropriate form requires the careful consideration of multiple factors, such as recency (Arnold 2010), the necessity of referent disambiguation (Arnold 2010), and pragmatic predictability (Bamberg 1987). This process thus develops slower compared with referential choices in other contexts.

Briefly, investigations into the development of referring expressions among monolingual children have revealed that the choice of referential form requires the integration of different levels of linguistic knowledge and cognitive abilities that are involved in language use and processing. Put differently, it is a highly sophisticated skill that develops correspondingly with cognition.

1.2. Crosslinguistic Influence (CLI) in Bilingual Children’s Referring Expressions

A vast majority of the research in the field of simultaneous bilingual acquisition supports the view that the two target languages develop separately in language-specific ways at the earliest stages, which has been referred to as the Separate Development Hypothesis (SDH) (e.g., De Houwer 1990; Deuchar and Quay 2001; Genesee 1989; Genesee et al. 1995; Meisel 2001; Itani-Adams 2007; Mishina-Mori 2002; Paradis and Genesee 1996; see De Houwer 2005; Paradis et al. 2021 for summaries). Although some recent studies challenge the above view and argue that certain features are “shared” at the level of syntactic representation, accounting for the observed crosslinguistic priming effects (Unsworth 2023), the SDH has gained much support in the past decades, drawing evidence from studies showing the comparable development of bilinguals and monolinguals (e.g., De Houwer 1990), as well as from pragmatic differentiation studies (e.g., Genesee et al. 1995). Regardless of the difference in assumptions, an accumulation of studies supports the view that cross-language interactions occur under certain linguistic conditions (Hulk and Muller 2000; Müller and Hulk 2001). One of the major hypotheses that is repeatedly tested among bilingual children with different language pairs is that structures at the syntax–pragmatics interface and/or with a partial overlap in the two languages are susceptible to CLI and that the language that exhibits multiple options for expressing a grammatical concept (thus exhibiting ambiguity in the input) tends to be affected by the language with a single option for expressing the same concept, causing excessive usage of the shared structure (Hulk and Muller 2000; Müller and Hulk 2001).

The first part of this hypothesis bases its idea on the concept of an “interface”, which stands on the assumption that language is a sum of different modules, such as syntax, semantics, phonology, and morphology, and the interface is where these different modules interact with each other (Ramchand and Reiss 2007; Rothman 2009). Interface structures thus require the knowledge of multiple modules of language, which includes both a language “internal” interface, such as a syntax and semantics interface, and a language “external” interface, involving, for example, a syntax and discourse context (Sorace 2011). Studies have suggested that the external interface, most notably being referential choice, is more likely to cause CLI than the internal interface (Sorace 2011; Tsimpli and Sorace 2006), giving stronger support for the claim that the integration of information from different cognitive domains such as a discourse context can be a substantial trigger for interlingual interaction. The idea that syntactic structures modulated by cognitive domains such as discourse pragmatics are more likely to reflect bilingual–monolingual differences than the structures that do not has also developed in investigations of advanced L2 learners (near-native) and the incipient stages of L1 attrition, proposed as the Interface Hypothesis (IH) (Sorace and Filiaci 2006; Sorace and Serratrice 2009; Sorace 2011). Thus, Müller and Hulk’s (2001) proposal that the interface structure is likely to trigger interlingual interactions in simultaneous young bilinguals has become relevant for a larger scope of bilinguals. The second portion of the hypothesis, on the other hand, has been termed a structural overlap/ambiguity hypothesis. Studies on both interface structures such as the realization of subjects or objects (e.g., Paradis and Navarro 2003), as well as non-interface structures, such as verb–object word order (e.g., Döpke 1998) and deverbal compounds (e.g., Nicoladis 2003), report that partial overlaps and ambiguities can explain CLI and its directionality.

Although it has been reported that a structural overlap can solely induce CLI, the combination of the syntax–pragmatics interface and the partial sharedness of structures have attracted considerable interest among CLI researchers, and some studies have particularly focused on the selection of the argument forms (overt or null) among preschool-age bilinguals acquiring null and non-null argument language pairs as a typical case of the syntax–pragmatics interface with partially overlapping structures to test the hypothesis (e.g., Hacohen and Schaeffer 2007; Haznedar 2010; Jachimek et al. 2022; Mishina-Mori et al. 2015; Mishina-Mori 2020; Nakano 2019; Paradis and Navarro 2003; Serratrice et al. 2004; Yip and Matthews 2007). In most cases, these studies report supportive evidence for the IH, together with the finding that the influence is unidirectional, from unambiguous to ambiguous structures, thereby supporting a structural overlap/ambiguity hypothesis.

As the studies above suggest, CLI was initially assumed to be a typical feature of the initial stages of bilingual grammar. However, more recent studies on school-age children revealed that interlingual influence is a common feature among bilinguals of any age range, as detected in production and comprehension studies, in which the syntactic features were assumed to be fully acquired (e.g., Argyri and Sorace 2007; Chen and Lei 2012; Montrul 2010; Sorace et al. 2009; Torregrossa and Bongartz 2018). For example, Chen and Lei’s (2012) study on Chinese–English bilingual school-age children pointed to the differences between bilingual and monolingual referring expressions in oral narratives, which can be interpreted as the result of transfer from English to Chinese. They reported the prominent usage of overt subjects/objects in the bilinguals’ Chinese (null argument language) compared with their monolingual peers when presenting elicited narratives in both languages, particularly in the reintroduction context, and this indicated a possible structural vulnerability-induced transfer. The observation in which the reintroduction context stood out as the locus of interaction deserves attention because its vulnerability to influence can be explained by the involvement of different conditions regarding its activation levels, the ambiguity of referents, and its semantic predictability. Extant studies that utilized acceptability judgments further supported the view that CLI is exerted on bilinguals’ linguistic intuitions (Argyri and Sorace 2007; Sopata et al. 2021). For example, Argyri and Sorace (2007) reported differences in the acceptability of pragmatically unnecessary overt subjects, which were utilized in the null argument language among English–Greek bilingual children, in contrast to their monolingual peers, thus indicating that bilingual and monolingual children exhibited different discourse–pragmatics knowledge. Thus, it can be argued that if interdependence occurs among older individuals who may have completed the initial stages of language development, CLI may be a feature of bilinguals’ language use at all ages, not limited to children developing speech who have reduced input in both languages.

The idea that structural conditions such as interface and ambiguity may specify the major loci of CLI, however, has been challenged. There is a growing body of studies proposing alternative explanations for CLI, i.e., extralinguistic factors such as the nature of bilingual processing and language dominance, highlighting evidence of CLI witnessed in conditions not involving interface/ambiguity and/or counterevidence for the overlap/ambiguity hypothesis. First, the nature of processing has received much attention. For example, Nicoladis (2002), Nicoladis (2006), and Nicoladis and Gavrila (2015) proposed that “co-activation” of the two languages triggers CLI based on their observation of bidirectional CLI in the use of compound nouns in English and French and in the use of adjective–noun order in English and French and in English and Welsh. Other studies also appealed for the co-activation hypothesis since the ambiguity condition did not result in the proposed directionality of influence (Nicoladis 2006; Foroodi-Nejad and Paradis 2009; Engemann 2022). Engemann (2022) investigated the use of path verbs in French and English, in which she found a bidirectional influence between the two languages regardless of the partial overlap identified in the path verb characteristics in the two languages. Studies finding strong evidence for the role of language dominance in CLI have also received attention (Faitaki and Murphy 2023; Yip and Matthews 2000). For example, Yip and Matthews (2000) claimed that language dominance best explained the transfer in their Cantonese–English bilingual toddlers. They presented evidence of the influence of Cantonese on English in the corpus, including pre-noun relative clauses and in situ wh-questions.

Studies on interface structures (in particular, referential choice) have also shifted to pursuing the possibility that processing plays a major role in bilingual–monolingual differences, which were once considered as CLI (e.g., Sorace 2011). Sorace et al. (2009), based on the evidence that overspecification occurred in bilinguals acquiring two null-subject languages, i.e., regardless of the acquired languages, argue that processing two languages, rather than one, can result in the overuse of overt forms in bilinguals as compared to monolinguals. More specifically, Sorace (2011, 2016) argues that when faced with a choice from the two possible forms (e.g., explicit or implicit forms), bilinguals tend to choose explicit forms as the “default” choice: i.e., they pay less attention to the listener’s perspective to alleviate the processing burden and prioritize avoidance of miscommunicating the referent. In other words, this phenomenon may be accounted for not only by the influence of the other language but also in terms of the unique nature of bilingual processing.

An important approach that is missing, however, is a close examination of the discourse–pragmatics conditions in the syntax–pragmatics interface: only a few studies have considered the effect of discourse conditions when analyzing CLI, i.e., whether the discourse context requiring the integration of different information levels may be more vulnerable to influence. Although the hypothesis concerns the involvement of a discourse context, no studies have conducted a careful analysis comparing the patterns of referential choice in contexts that differ in the complexity of involved factors. Furthermore, the foregoing literature review revealed that most investigations on school-age bilinguals were limited to bilinguals acquiring non-null-subject languages (e.g., English and French) and null-subject languages with rich inflectional morphologies, such as Greek (Torregrossa et al. 2021) or Italian (e.g., Serratrice 2007; Torregrossa and Bongartz 2018). However, only very few studies have considered bilinguals acquiring null-subject languages without rich inflection (e.g., Chinese, Korean, and Japanese: Chen and Lei (2012) on Chinese–English bilinguals, as discussed above; Blais et al. (2010) on Japanese–French bilinguals). Liu et al. (2020) confirmed the overproduction of subjects in Chinese–English bilinguals but the participants were late bilinguals. Kang (2013) investigated subject realization in Korean–English bilingual children but their strong dominance in Korean made the study less comparable with others in the current analysis.

In this study, we tested the interface and overlap hypotheses by analyzing the referring expressions in different discourse contexts in the narratives of school-age bilingual children acquiring Japanese and English. The referring expression in the bilingual narratives of school-age children represents an ideal testing ground for CLI since it can be assumed that children have established the basic syntaxes of both languages, as well as adult-like referential strategies; moreover, a complex interplay exists between discourse conditions and the selection of referential forms, which is the most fundamental feature of interface structures. Furthermore, the Japanese–English pair is a rarely studied language pair in this field, except for by Mishina-Mori et al. (2015), Mishina-Mori (2020), and Nakano (2019), all of whom focused on preschool-age children. Since Japanese is a null argument language, i.e., it allows the omission of referents (both subjects and objects) with few syntactic clues, such as agreement markers, a typical interface structure with a superficial overlap is created.

1.3. Referring Expressions in Japanese and English

Japanese is a subject–object–verb (SOV) language exhibiting a postpositional structure, where grammatical relations are indicated by postpositional markers; subjects, objects, and topics are marked by ga, wo, and wa, respectively, thus generating great flexibility in the word order. Subjects and objects are unexpressed when the referent can be understood from the discourse, which has also been referred to as the ellipsis (Hinds 1984). No subject–verb agreement markers exist, as observed in subject-drop languages, such as Spanish or Italian, which enable the listener to understand the referent of the null subject, although the honorific or passive forms can indicate the agent of the action (Yoshimura and MacWhinney 2010). In principle, pronouns or the syntactic replacement of NPs do not exist in the language (Iwasaki 2013; Shibatani 1990), although there are different strategies for replacing NPs, e.g., by utilizing deictic pronouns before nouns (e.g., sono hito “that person”) or the English translation of third-person pronouns (kare “he” and kanojo “she”). Although first- and second-person pronouns do exist, they are close to lexical items, and the choice is determined by sociolinguistic conditions (Shibatani 1990). Thus, in principle, reduced forms in Japanese are null forms, and pronouns are infrequent (Tsuchiya 2015).

When introducing a referent in narrative discourse (or marking new information), the NP in the subject position is in principle followed by the subject marker, ga (Minami 2011). In the reintroduction context in which the referent reappears after being replaced by another (marking accessible information), the NP is followed by ga/wa, or pronouns or null forms can be utilized. Null forms are the norm when maintaining a referent (marking given information) (Minami 2011). Referents in the object position accompany the object marker, wo, regardless of the context of the discourse.

Conversely, English is an SVO language that exhibits a prepositional structure, and word order is key to indicating grammatical relationships. Reduced forms in the English language are largely pronouns, and null arguments are ungrammatical in principle; they are restricted to several specific conditions (see Orfetelli and Hyams 2012).

In most cases, referent introduction is marked by the indefinite article, “a”, before the NP. Further, referent reintroduction, following replacement, is typically indicated by the definite article, “the”, preceding the NP or by the utilization of pronouns; moreover, referent maintenance is marked by the utilization of pronouns (Chafe 1994; Du Bois 1987; Minami 2011).

Table 1. Japanese and English referring expressions in different discourse contexts.

Discourse Context	Introduction	Reintroduction	Maintenance
Information status	New (inactive)	Accessible (semi-active)	Given (active)
Japanese	S: NP–ga O: NP–wo	S: NP–ga/wa O: NP–wo (Pronoun)/Null	Null
English	An NP	The NP Pronoun	Pronoun

Note: Adapted from “Referent introduction and maintenance in the English narratives of monolingual and bilingual children”. By Mishina-Mori et al. (2018, p. 10). Copyright 2018 by the Graduate School of Intercultural Communication. Adapted with permission.

presents a summary of the different discourse contexts of referring expressions in narratives involving both languages.

2. Research Questions (RQs)

The following RQs will be addressed in this study:

• How do school-age Japanese–English bilinguals and their monolingual (Japanese or English) counterparts utilize referential forms to introduce, reintroduce, and maintain the topic?
• Is there any indication of CLI (a prominent usage of overt forms or NPs in the Japanese narratives and a prominent usage of null arguments in the English narratives) in the bilingual children’s usage of referring expressions in different discourse contexts?

We expected to observe a unidirectional influence (from English to Japanese), as predicted by the structural overlap hypothesis. Namely, we expected significantly more usage of overt forms (NPs) in the bilinguals’ Japanese narratives than in those of their monolingual counterparts. Conversely, we did not expect the usage of null arguments in the English narratives of both groups of children (bilinguals and monolinguals), indicating the absence of any influence from Japanese. Further, we predicted that the influence would mainly manifest in the reintroduction context of a narrative owing to the assumed complexity of information structure.

It is important to note that, echoing Shin et al. (2023), the goal of the bilingual–monolingual comparison in this study is to understand the mechanism of interlingual influences between the coexisting grammars, which is considered to play a crucial role in constructing bilinguals’ unique linguistic system. Analyzing the similarities and differences between bilingual grammar and those of monolinguals is indispensable when investigating the nature of crosslinguistic influence on bilingual grammar.

3. Method

3.1. Participants

Twenty-nine children who were between the ages of 8 and 14 years participated in this study. Among them were 13 simultaneous/early bilinguals (mean age, 11.7), as well as seven Japanese monolinguals (mean age, 10.6) and nine English monolinguals (mean age, 10.7) whose economic statuses were comparable with those of the bilingual participants. The bilinguals and Japanese monolinguals resided in the suburban areas of Tokyo and were recruited through personal contact by the authors; the English monolinguals were recruited in California (U.S.A.) through personal contact by one of the authors, as well as through a nonprofit local Japanese weekend language school in the same area. The children could be considered as least bilingual or monolingual English speakers as their use of Japanese was limited to the weekly lessons at this school.

The linguistic profiles of the bilingual children were obtained through parental interviews or a language background questionnaire. Nine of the bilinguals were simultaneous bilinguals who had been exposed to both languages from birth, mostly through a one-parent–one-language policy in bilingual homes. Four of the bilinguals could be considered early successive bilinguals who acquired Japanese as their first language before they were exposed to English shortly before or at the start of elementary school (four to six years of age) since they attended international schools in Japan or local/international schools in English-speaking countries. The children employed both languages regularly in their homes and/or schools. The demographic information of the bilingual participants is listed in

Table 2. Demographic variables of the bilingual participants.

Child	Age	M/F	Simultaneous/Successive (Age of L2 Onset)	Language in the Home	Language of Instruction in School
B-1	13	M	Successive (5)	Japanese	English
B-2	12	M	Simultaneous	English/Japanese	Japanese
B-3	9	M	Simultaneous	English/Japanese	Japanese
B-4	9	M	Simultaneous	English/Japanese	Japanese
B-5	13	M	Simultaneous	English/Japanese	Japanese
B-6	9	M	Simultaneous	English/Japanese	Japanese
B-7	13	F	Simultaneous	English/Japanese	English
B-8	13	F	Successive (4)	Japanese	English
B-9	13	F	Successive (6)	Japanese	English
B-10	13	F	Successive (6)	Japanese	English
B-11	11	F	Simultaneous	English/Japanese	Japanese
B-12	9	F	Simultaneous	English/Japanese	Japanese
B-13	14	M	Simultaneous	English/Japanese	English

.

Since language dominance can play a significant role in the level of crosslinguistic influence (e.g., Yip and Matthews 2007; Torregrossa and Bongartz 2018), we evaluated the bilingual children’s overall dominance following the methods of Montrul (2016), who defined the term as “[the] relative weight and relationship of the two languages of a bilingual in terms of language use and degree of proficiency” (p. 16). More specifically, we considered the following factors: (1) a linguistic proficiency component, evaluated by linguistic measures calculated from their narrative production;1 (2) an external component, judged based on the language of the community within which the children were embedded; and (3) a functional component, based on the language use patterns and distribution of the two languages in the different domains of the children’s lives.

First, linguistic proficiency was measured with the Subordination Index (SI, the number of clauses per T-unit)2 (Miller et al. 2019). The SI presents the syntactic complexity of a learner’s language abilities and has been found to distinguish L2 learners at different proficiency levels (De Clercq and Housen 2017; Iwashita 2006). The index was calculated for each language of the bilinguals and compared with their monolingual counterparts. We avoided interlingual comparisons since comparing syntactic complexity across languages is generally considered less reliable, especially if the two languages have typologically distinct structures (De Clercq and Housen 2017). As is indicated in

Table 3. Subordination Index of bilinguals and monolinguals for Japanese and English language abilities.

	Japanese		English
	Bilingual	Monolingual	Bilingual	Monolingual
N	13	7	13	9
Mean	1.32	1.36	1.12	1.27
SD	0.16	0.12	0.12	0.20
Range	1.15–1.73	1.25–1.58	1.0–1.36	1.03–1.53
t		00.57		−2.19
df		18		20
p		0.577		0.04

, a t-test analysis revealed that there was a significant difference in the SI of English (p = 0.04) but not Japanese (p = 0.577) between bilinguals and monolinguals, which shows that the bilinguals had similar proficiency to their monolingual counterparts in Japanese but somewhat lower proficiency than their monolingual counterparts in English.

Second, we considered the community language or environmental language (Qi and Di Biase 2020) of the children. Although both Japanese and English were used in the daily lives of each child, the children may have been somewhat more dominant in their use of Japanese since it was the language of their environment (Qi and Di Biase 2020). Although some of the participants went to schools in which the majority of their classes were taught in English, Japanese was prevalent in their community, the media, and other domains of their lives. Previous studies suggested that bilingual children educated in foreign language immersion schools are typically more dominant in their community language if they are embedded in a monolingual society (e.g., Andreou et al. 2020; Bostwick 2001).

The language use patterns and distributions of the two languages—the amount of exposure and use of the two languages on a regular basis—were also taken into consideration. The children used both languages on a daily basis (home/school), but the distribution of the two languages in different domains of the children’s lives was mixed among the children. Some children used Japanese with family and friends and English at school for subject learning/academics, and some went to local schools in Japan and interacted with one of their parents in English at home. However, what is common among all participants is that they were most likely to use Japanese with their peers: through personal communication participants mentioned that Japanese was their language for peer interaction. As Serratrice (2007) also argues, the language used among peers can be the most influential language for school-age children, and thus, that language may be more dominant in those children.

From the above analysis, it can be concluded that the current bilingual participants are somewhat more dominant in Japanese, although there may be some individual variations depending on different domains.

3.2. Data Collection

The children were required to narrate two stories in each language. One of the stories was based on a wordless picture book, Frog, Where Are You? (Mayer 1969), consisting of 24 scenes involving eight characters. It is a story about the adventure of a young boy and a dog who go searching for their frog that escaped from their home. They encounter several different animals in the woods during their search mission. The other story was narrated based on a speechless video clip, The Museum Guard (Adam 2012), from a Chaplin animation series. The clip was approximately 5 min long and involved Chaplin and seven other characters. Therein, the animated Chaplin serves as a museum guard under the supervision of the museum owner and interacts with many visitors. Different elicitation methods were adopted to ensure the generalizability of the observed tendency across different elicitation types and story contents. The frog story was selected to gather data comparable with previous findings as it has been adopted in numerous narrative elicitation tasks, including those involving Japanese monolingual children (Nakamura 1993; Suzuki et al. 2014; Miyata and Inaba 2014) as well as Japanese–English bilingual children (Minami 2011). The Chaplin story (animation), on the other hand, was chosen since the approximate length of the video/story as well as the number of characters that appear in the story were comparable with those of the frog story, meaning that the number of characters introduced in the story was more or less comparable between the two (eight characters in the frog story: the frog, the boy, a dog, a mole, bees, an owl, a deer, and the family of the frog; eight characters in the Chaplin story: Chaplin, the manager/boss, three different visitors to the museum, two children, and their mother). Using two materials also enabled us to collect ample referential choice data from a relatively small number of participants.

In both tasks, the children were allowed to first preview the contents by looking through the picture book or watching the video on an iPad. Thereafter, they were required to narrate the story to the researcher while looking at/watching the book/video. We opted for a retelling of the stories as they looked through the book/video rather than telling the story without visual information so that the contents of both stories was not affected by the children’s memories. By adopting this method, the two elicitation tasks differed in terms of the time pressure that participants faced when retelling the story: in the wordless picture book task, the participants maintained control of the pacing for their retelling of the story (they turned the pages themselves), whereas when using the speechless video clip, participants experienced a time pressure when retelling the story as the story unfolded in front of their eyes. This may have more or less affected the referential choice pattern, since the limited processing capacity caused by the time pressure may have triggered more uses of reduced forms as they required less factors for language retrieval (Jescheniak et al. 2001; Torregrossa et al. 2019). However, we expected that this would not cause a substantial difference given that the children were provided with sufficient time to preview the story beforehand.

The children held the book or iPad (10.9 inch) in front of them so that the visual information was only available to them, and this created a naïve listener condition, thus controlling for shared knowledge between the speaker and listener, which was assumed to affect the referencing behavior (Lucero et al. 2021; Serratrice 2007). In both elicitation tasks, the researcher listened to the children’s stories and engaged in backchanneling naturally while refraining from asking questions. The children first narrated the frog story to a bilingual researcher, followed by the Chaplin story. The order of language use during elicitation was determined by the children’s preferences to minimize tension, although it turned out that all the children started with English.3 All the elicitation events were video- recorded using a Sony HDR-XR520V (Sony corporation, Tokyo, Japan) and audio-recorded using a Sony IC recorder ICO-UX533F (Sony corporation, Tokyo, Japan). The monolingual children were subjected to the same procedure, except that they only narrated the story in one language.

3.3. Transcription, Coding, and Analysis

The data were transcribed and coded by bilingual researchers employing the CHAT (MacWhinney 2000) and JCHAT (Oshima-Takane et al. 1998) conventions into English and Japanese data, respectively. The researchers are proficient bilinguals who use the two languages on a regular basis. Following studies analyzing naturalistic speech among bilingual children (De Houwer 1990; Mishina-Mori 2005; Nicoladis 1994), we utilized utterance as a unit of analysis when transcribing the data, which is defined as “a word or a group of words with a single intonation contour” (Lanza 1992, p. 638). Further, we performed quantitative analyses employing CLAN (MacWhinney 2000).

The children’s utterances with referents were coded regarding their (1) grammatical roles—subject or object; (2) discourse context of the occurrence—introduction, reintroduction, and maintenance; and (3) referent form—NP, pronoun, and null form. Thereafter, the occurrences of each referent form in the different discourse contexts were counted, and the proportion of each form among all the referential form usages was calculated.

Example 1 below presents an excerpt from our bilingual English data, revealing examples of the coding for referential functions. The boy and the dog were introduced in the first line, and the frog in the second line, after which the boy was reintroduced in the third line. Subsequently, the boy was referenced again with a pronoun, which was coded as maintenance. Example 2 lists the examples of each type of referent form.

Example 1: Referential functions (bilingual, 10 years old)

One day a little boy and a dog was looking in a jar.	[introduction]
What’s in a jar was a frog.	[introduction]
After that the boy was tired.	[reintroduction]
So he went to sleep in his bed.	[maintenance]

Example 2: Referential forms

NP	The boy   looked for the frog
	otokonoko wa kaeru wo sagashimashita.
	boy   TOP frog OBJ search-PAST4
Pronoun	He   looked for the frog.
	kare wa kaeru wo sagashimashita.
	He TOP frog OBJ search-PAST
Null	Ø   looked for the frog.
	Ø   kaeru wo sagashimashita
	frog OBJ search-PAST

First, we calculated the percentage occurrence of each referential form compared with all the subject/object references utilized in the introduction, reintroduction, and maintenance contexts. Thereafter, we compared the narratives of the bilinguals and monolinguals.

4. Results

4.1. Japanese

Table 4. Mean percentages of Japanese referring expressions used for referent introduction, reintroduction, and maintenance in bilinguals and monolinguals (number of occurrences)—frog story.

	Introduction		Reintroduction		Maintenance
	Bilingual	Monolingual	Bilingual	Monolingual	Bilingual	Monolingual
Null	0% (0)	4.7% (2)	23.5% (63)	46.1% (53)	57.5% (104)	65.1% (56)
Pronoun	5.6% (5)	0% (0)	1.1% (3)	0% (0)	3.3% (6)	2.3% (2)
NP	94.4% (85)	95.3% (41)	75.4% (202)	53.9% (62)	39.2% (71)	32.6% (28)
Total	100% (90)	100% (43)	100% (268)	100% (115)	100% (181)	100% (86)

and

Table 5. Mean percentages of Japanese referring expressions used for referent introduction, reintroduction, and maintenance in bilinguals and monolinguals (number of occurrences)—Chaplin story.

	Introduction		Reintroduction		Maintenance
	Bilingual	Monolingual	Bilingual	Monolingual	Bilingual	Monolingual
Null	1.5% (1)	2.6% (1)	44.3% (121)	63.6% (75)	87.6% (352)	92.3% (229)
Pronoun	1.5% (1)	0% (0)	0.4% (1)	0% (0)	2.2% (9)	0.8% (2)
NP	97.0% (64)	97.4% (37)	55.3% (151)	36.4% (43)	10.2% (41)	6.9% (17)
Total	100% (66)	100% (38)	100% (273)	100% (118)	100% (402)	100% (248)

and Figure 1 and Figure 2 show the proportions of different forms of referring expressions in the introduction, reintroduction, and maintenance contexts as utilized by the bilinguals and their monolingual peers to narrate the frog story and the Chaplin video, respectively. The analysis of preferred forms for each context was reported first, followed by the between-group comparison of different referential forms. When the observed frequencies of the pronouns and/or null forms were <5, the data were collapsed with each other for statistical analysis. The statistical analysis of the preferred forms in each group was performed using a Chi-square analysis following the method of Chen and Lei (2012). For the bilingual–monolingual comparison, a Mann–Whitney U test was utilized instead of a t-test as the data set was small and a normal distribution was not expected.

In the introduction context, most of the referring expressions were NP forms (94.4% for bilinguals and 95.3% for monolinguals in the frog story; 95.4% for bilinguals and 97.4% for monolinguals in the Chaplin video); there were slight usages of the reduced forms in both groups. The results of the Chi-square analysis confirmed a significant difference between the usage of NPs and the reduced forms for the bilinguals (χ² (1, n = 90) = 71.111, p < 0.01) and monolinguals (χ² (1, n = 43) = 35.372, p < 0.01) in the frog story data and for the bilinguals (χ² (1, n = 65) = 53.554, p < 0.01) and monolinguals (χ² (1, n = 38) = 35.103, p < 0.01) in the Chaplin video data.

In the reintroduction context, the bilinguals and monolinguals exhibited different tendencies. For example, in the frog story data, although the bilinguals and monolinguals utilized more NPs than other forms, the bilinguals utilized significantly more NP forms (75.4%) than the reduced forms (24.6%), χ² (1, n = 268) = 69.015, p < 0.01, whereas there was no significant difference between the usages of NPs (53.9%) and reduced forms (46.1%) among the monolinguals, χ² (1, n = 115) = 0.704, p < 0.40133. Conversely, clear bilingual–monolingual differences were observed in the Chaplin video data. The bilinguals preferred NPs (55.1%) to the reduced forms (44.9%), with a statistical difference between the two, χ² (1, n = 274) = 2.861, p < 0.10. However, the monolinguals exhibited a strong preference for the null forms (63.6%) compared with NPs (36.4%), and the difference was statistically significant, χ² (1, n = 118) = 8.673, p < 0.01.

In the maintenance context, the null form was most prominent among all the forms (57.5% for bilinguals and 65.1% for monolinguals in the frog story data; 87.3% for bilinguals and 92.3% for monolinguals in the Chaplin video data), followed by NPs (39.2% for bilinguals and 32.6% for monolinguals in the frog story data; 10.2% for bilinguals and 6.9% for monolinguals in the Chaplin video data), as well as a few occurrences of pronouns. Null forms were significantly more frequently utilized among the bilinguals compared with the usage of NPs, followed by pronouns, χ² (2, n = 181) = 82.405, p < 0.01, in the frog story data and χ² (2, n = 403) = 553.097, p < 0.01, in the Chaplin video data. The monolinguals utilized more null forms than NPs, χ² (1, n = 86) = 10.465, p < 0.01, in the frog story data and χ² (1, n = 248) = 184.661, p < 0.01, in the Chaplin video data.

Thus, the data revealed that the bilinguals and monolinguals exhibited different tendencies in their referential choice only in the reintroduction context.

Next, a bilingual–monolingual comparison was conducted in each context, employing the Mann–Whitney U test. The analysis of the frog story data revealed that the bilinguals’ usage of NPs and null forms was significantly different from that of their monolingual peers only in the reintroduction context (75.4% vs. 53.9%, U = 21, p < 0.05; 23.5% vs. 46.1%, U = 20.5, p < 0.05, respectively). Although a statistically significant difference was not observed in the Chaplin video data (55.1% vs. 36.4%, U = 23, p > 0.05; 44.2% vs. 63.6%, U = 23, p > 0.05, respectively), the bilinguals utilized more NPs and less null forms compared with their monolingual peers.

In the maintenance and introduction contexts, the data did not yield significant differences between the usages of NPs and null forms by the bilinguals and monolinguals. Thus, we confirmed our prediction that the bilinguals tended to utilize more NPs compared with their monolingual peers when reintroducing a referent, and this can be interpreted as an influence from their English.

Two further analyses for the Japanese data were conducted to add validity to (1) the bilingual data set, which involved both simultaneous and early successive bilinguals who may have differed in their developmental trends for the two languages and (2) the small sample size of the whole data. First, we examined if there were any differences between the simultaneous bilinguals and the successive bilinguals in terms of the level of influence from English, as the successive bilinguals may have been less vulnerable to influences from English due to their later exposure to English as compared with the simultaneous group and thus may have exhibited a weaker level of influence. We calculated the ratio of each referential form for simultaneous and successive bilinguals in the reintroduction and maintenance contexts (where an influence was most likely to be observed) and compared the two using a Mann–Whitney U test. The results for both the frog story data and the Chaplin video data revealed that although the percentages of NPs in simultaneous bilinguals constantly exceeded those of successive bilinguals, there were no statistical differences between the two groups. In the frog story data, the percentages of the NP forms in the simultaneous and successive groups were 76.5% and 72.7%, respectively (U = 14, 0.1 > p), in reintroduction, and 42.7% and 32.3%, respectively (U = 14, 0.1 > p), in the maintenance context. The Chaplin video data also showed similar results: 63.3% and 41.2% (U = 12.5, 0.1 > p) for the reintroduction context and 12.1% and 6.5% for the maintenance context (U = 12.5, 0.1 > p), respectively. The fact that simultaneous and successive bilinguals exhibited roughly similar tendencies corroborates the view that in language features that are acquired later in life, which involve language-external features (such as discourse–pragmatics), simultaneous and successive bilinguals tend to show similar developmental trends, as compared with features acquired early, which are typically narrow syntax, in which simultaneous and successive bilinguals show distinct patterns of development (Sopata 2019).

Second, we looked at individual data about the participants and found that despite the seemingly large individual variation, the data still support a bilingual–monolingual difference. In the bilingual Japanese frog story data, the percentages of NPs ranged from 52% to 95% in the reintroduction context, with constantly more NPs used than null forms. In the monolingual data, on the other hand, the percentages of NPs ranged from 18% to 71%, with three participants exhibiting a reverse pattern (null > np) compared with the rest of the children (np > null). Thus, a tendency for the bilinguals to use more NPs than their monolingual peers was confirmed. In the Chaplin narrative, on the other hand, the percentages of NPs ranged from 27% to 74%, whereas the monolingual data ranged from 20% to 54%. Although a bilingual–monolingual difference was not as clear in the frog story data, the bilinguals showed a mild tendency to use more NPs compared with their monolingual peers: four bilingual participants used more NPs than the null form, whereas only one monolingual child preferred NPs. Thus, although the number of participants was very small, the data suggest that bilinguals tend to use more NPs than monolingual children.

In summary, the Japanese data of bilinguals and monolinguals exhibited very similar patterns regarding usages in the introduction and maintenance contexts, although a clear difference was observed in the reintroduction context: the bilinguals preferred NPs to the null forms, and they utilized significantly more NPs than their monolingual peers in the frog story.

4.2. English

Table 6. Mean percentages of English referring expressions used for referent introduction, reintroduction, and maintenance in bilinguals and monolinguals (number of occurrences)—frog story.

	Introduction		Reintroduction		Maintenance
	Bilingual	Monolingual	Bilingual	Monolingual	Bilingual	Monolingual
Null	0% (0)	0% (0)	0% (0)	1.3% (2)	0% (0)	0% (0)
Pronoun	1.4% (1)	4.2% (2)	23.4% (58)	33.3% (52)	59.0% (108)	74.5% (123)
NP	98.6% (73)	95.8% (46)	76.6% (190)	65.4% (102)	41.0% (75)	25.5% (42)
Total	100% (74)	100% (48)	100% (248)	100% (156)	100% (183)	100% (165)

and

Table 7. Mean percentages of English referring expressions used for referent introduction, reintroduction, and maintenance in bilinguals and monolinguals (number of occurrences)—Chaplin story.

	Introduction		Reintroduction		Maintenance
	Bilingual	Monolingual	Bilingual	Monolingual	Bilingual	Monolingual
Null	0% (0)	0% (0)	0% (0)	0.5% (2)	1% (3)	0.8% (3)
Pronoun	5.1% (3)	11.8% (6)	49.2% (129)	61.8% (218)	87.5% (253)	91.6% (339)
NP	94.9% (56)	88.2% (45)	50.8% (133)	37.7% (133)	11.5% (33)	7.6% (28)
Total	100% (59)	100% (51)	100% (262)	100% (353)	100% (289)	100% (370)

as well as Figure 3 and Figure 4 summarize the referential choices in the three contexts regarding the bilingual and monolingual English narratives of the frog and Chaplin stories, respectively. The very few occurrences of null forms (0–3) were collapsed with those of pronouns during the statistical analysis.

When introducing referents, the bilinguals and monolinguals strongly preferred NPs (98.6% for bilinguals and 95.8% for monolinguals in the frog story; 94.9% for bilinguals and 90.4% for monolinguals in the Chaplin video data), and the reduced forms were infrequent. The results of the Chi-square analysis confirmed the statistical difference between the usages for the bilinguals, χ² (1, n = 74) = 70.054, p < 0.01, and monolinguals, χ² (1, n = 48) = 40.3333, p < 0.01, in the frog story, as well as for the bilinguals, χ² (1, n = 59) = 47.61, p < 0.01, and monolinguals, χ² (1, n = 52) = 33.923, p < 0.01, in the Chaplin video data.

The referential choice patterns of both groups regarding the reintroduction context were similar in the frog story data. The bilinguals and monolinguals mostly utilized NPs (76.6% and 65.4%, respectively), followed by pronouns (23.4% and 33.3%, respectively). The statistical analysis confirmed that a significant difference existed between the usages of NPs and the reduced forms among the bilinguals, χ² (1, n = 248) = 70.258, p < 0.01, and monolinguals, χ² (1, n = 156) = 14.769, p < 0.01. Conversely, in the Chaplin video data, we observed a different tendency. In the bilinguals’ data, the usage of NPs and pronouns were uniformly distributed (50.8%, 49.2%), χ² (1, n = 262) = 0.061, p < 0.80481, whereas the monolinguals utilized pronouns (62.1%) significantly more often than NPs (37.4%), χ² (1, n = 356) = 22.753, p < 0.01.

When maintaining the referents, pronouns were used most frequently among the bilinguals and monolinguals in both stories (59% and 74.5% in the frog story data and 85.7% and 91.6% in the Chaplin video data, respectively), followed by NPs (41.4% and 25.5% in the frog story data and 11.5% and 7.6% in the Chaplin video data, respectively), and the null forms were not observed. However, the Chi-square analysis revealed that the difference was not significant in the bilingual frog story data, χ² (1, n = 183) = 5.951, p < 0.01471, whereas the difference was significant in the monolingual data, χ² (1, n = 165) = 39.764, p < 0.01. The Chaplin video data exhibited a statistical difference between the usage of NPs and pronouns among the bilinguals, χ² (1, n = 289) = 172.073, p < 0.01, and monolinguals, χ² (1, n = 368) = 264.522, p < 0.01.

Thus, in the English frog story data, the referential choice patterns of the bilinguals and monolinguals were different in the maintenance context: NPs and pronouns were uniformly distributed among the bilinguals, whereas the monolinguals strongly preferred pronouns.

A further statistical analysis was performed via a Mann–Whitney U test to examine if there were differences between the bilinguals’ and monolinguals’ usage of each referential form in the three discourse contexts. In the frog story data, we observed a significant difference in the usage of NPs and pronouns only in the maintenance context (41.0% vs. 25.5%, U = 26.5, p < 0.05; 59.0% vs. 74.5%, U = 28.5, p < 0.05, respectively). In the reintroduction context, the bilingual–monolingual difference for each referential form was not significant, although the bilinguals utilized more NPs and less pronouns than their monolingual peers (76.6% vs. 65.4%, U = 33.5, p > 0.05; 23.4% vs. 33.4%, U = 34.5, p > 0.05, respectively). In the Chaplin video data, we did not observe any difference between both groups in all three contexts. Additionally, in the reintroduction context, the bilinguals utilized more NPs and less pronouns compared with their monolingual peers, but no statistical difference was confirmed (50.8% vs. 37.4%, U = 35.5, p > 0.05; 49.2% vs. 62.1%, U = 38.5, p > 0.05, respectively).

In summary, the referential choice patterns of the bilinguals were largely comparable with those of their monolingual peers except for the maintenance context in the frog narrative: the monolinguals strongly preferred pronouns, although such preferences were not observed in the bilingual data. Both sets of data also confirmed that the null forms were rarely utilized by the bilinguals and monolinguals; thus, no trace of crosslinguistic transfer from Japanese to English was observed.

5. Discussion

5.1. Referential Choice Patterns in the Two Languages (RQ1)

The obtained data indicated that the bilinguals and monolinguals generally exhibited similar referential choice patterns in Japanese and English, with some notable differences. In the Japanese narratives, the bilinguals and monolinguals exhibited similar tendencies except in the reintroduction context. As expected, both groups strongly preferred utilizing NPs to introduce referents; they exhibited a mild to strong preference for utilizing null forms in the maintenance context. However, when reintroducing referents, the bilinguals exhibited a mild to strong preference for NPs, whereas the monolinguals did not exhibit such a tendency, as also confirmed by Chen and Lei (2012). It can be inferred that the difference manifested the complexity of information structure in this particular context (e.g., Orsolini et al. 1996), triggering an influence from the English structure, as will be discussed in the next section.

In the English narratives, other than in the introduction context where NPs were predominantly utilized by both groups, the bilingual–monolingual difference was prominent: although the bilinguals rarely utilized the null forms, they tended to utilize NPs when pronouns would have been a more reasonable option in the maintenance and the reintroduction contexts. This indicates that the bilinguals preferred explicit forms even when the referent was already established in the discourse and could be easily inferred from the context, i.e., when it was highly activated. This was unexpected since maintenance was the context with the least complexity, and children’s usage approximates those of adults from the earliest stages of development in different languages, including English (Hickmann et al. 1996). The possible factors contributing to this unexpected referential choice are discussed in the next section.

5.2. CLI and the Language-Internal and Language-External Factors (RQ2)

As predicted, we observed significantly higher usages of NPs and fewer of the null forms among the bilingual children compared with their monolingual peers in the Japanese narratives. This over-informativeness, as observed in the excessive use of NPs in the bilingual children’s Japanese, can be interpreted as a manifestation of the influence of English: the bilinguals were inclined to adopt the shared structure (overt form) in Japanese, which was, in principle, the sole option in English. Conversely, in the English narratives, the null forms were rarely utilized, and the bilingual data were comparable with the monolingual ones, indicating the absence of CLI from the Japanese language. Thus, the data manifested an influence of English on Japanese but not vice versa. This finding is consistent with previous ones in which the language with the more unambiguous structure affects the one with the more ambiguous structure (e.g., Haznedar 2010; Paradis and Navarro 2003; Serratrice et al. 2004): the shared structure between both languages (overt form) triggers more frequent usage of those forms in Japanese, but English is not affected by the presence of Japanese. Thus, these findings support Hulk and Muller’s (2000) hypothesis that the syntax–pragmatics interface with a partial structural overlap between both languages is susceptible to CLI. This evidence is worth considering because the directionality of the influence was from English and not vice versa even though the included bilinguals in this study were mildly more dominant in Japanese, as predicted by the hypothesis. This finding indicates that language dominance does not play a major role. The data also add to the accumulated evidence to support the view that CLI is also observed in older children (e.g., Chen and Lei 2012; Serratrice 2007; Torregrossa and Bongartz 2018; and Sopata et al. 2021 on referential choice; Bosch and Unsworth 2021 on word order; and Engemann 2022 on the use of motion verbs), which is also in line with the current view that CLI is a “hallmark” (Torregrossa and Bongartz 2018) or a “permanent feature” (Austin 2021) of bilingualism. More importantly, our findings revealed the influence of a language with a pronominal system on one without it, indicating the usage of NPs as a form of overspecification.

Furthermore, as expected, CLI from English to Japanese was observed only in the reintroduction context, not affecting referential choice in other contexts, and this correlates with Chen and Lei’s (2012) results. This result indicates that Hulk and Muller’s (2000) hypothesis alone cannot explain why CLI only manifests in the reintroduction context and not in the others. We conclude that a contribution of processing complexity, in which the reintroduction context is involved, exists; different discourse–pragmatic factors involved in the selection of the appropriate referential form in each context, such as the distance from previous mention and the number of competing referents, among others (e.g., Arnold 2010; Bamberg 1987; Orsolini et al. 1996). Put differently, our data suggest that this particular context is the locus of interlingual influence. This finding supports the view regarding the collaboration between linguistic and processing factors: CLI, which is triggered by language-internal factors, is more likely to occur when the discourse context requires the integration of information from multiple domains (Sorace 2011; Torregrossa and Bongartz 2018).

The possibility of the general effects of bilingualism per se, however, cannot be ignored if we take into account the fact that the bilinguals preferred NPs to pronouns in English in the reintroduction and maintenance contexts, which was not predicted from the overlap/ambiguity hypothesis. The current study hypothesized that the overlapping NPs affect the referential choice in Japanese, in which the referential options are, in principle, limited to null or NPs; however, no predictions were available for the use of pronouns in bilingual English. The relatively high percentages of NPs as compared with pronouns in bilingual English could be a sign of the general effects of bilingualism, supporting the idea that NPs/overt forms are the default options when there are limited resources due to processing two languages simultaneously (Sorace 2011). If so, then, we cannot exclude the possibility that such an effect also takes place in bilingual Japanese to a certain extent. Thus, the data suggest that both CLI and processing effects may take place.

One could also argue that the less frequent use of pronouns and the more frequent use of NPs in bilingual English is due to a lack of experience with using pronouns in the language. Our results appear to be comparable with those of Torregrossa and Bongartz (2018), who observed a prominent usage of NPs in German–Italian bilinguals’ Italian narratives, where pronouns were typically utilized by their monolingual peers. They speculated that being dominant in German, the children lacked experience in utilizing pronouns in Italian and thus resorted to adopting NPs to avoid null forms and to ensure accuracy. This analysis may also apply to our data since the bilingual participants were generally less dominant in English. However, there seems to be no firm evidence supporting the lack of experience/input hypothesis; in fact, a longitudinal analysis of some of the current bilingual participants reveals that the referential choice pattern remains constant after an interval of two years. Although still a preliminary analysis, this could weaken the claim that the children’s use of pronouns was immature, which would in turn lend support to the bilingual processing account.

5.3. Similarities and Differences between the Two Elicitation Methods

Although the overall tendencies of referential choice among the three different forms (noun, pronoun, and null) remained constant in both sets of data, the Chaplin video elicited more implicit forms in both languages compared with the frog story: null forms in Japanese and pronominal forms in English. We conjecture that this is most likely due to the differences in the nature of the story; that is, the frog story required more overt mention of the referent compared with the Chaplin video. The frog story proceeded with two main characters (the boy and the dog) who often engaged in different activities simultaneously in one scene; therefore, the participants were more often required to differentiate the boy from the dog using overt forms. On the other hand, Chaplin was the major protagonist in the video, which made it more natural to refer to him in inexplicit forms, and the other characters appeared much less frequently. Furthermore, there was less need to differentiate between the multiple characters since they had different roles in the story, which were evident (e.g., the boss of the museum vs. the guard working under him, or the guard and the visitors), and the agent of each action was relatively less ambiguous.

The time restriction of the video elicitation (the speaker did not have control over the progress of the story) may have resulted in the more frequent use of pronouns (Jescheniak et al. 2001; Torregrossa et al. 2019). Fatigue could have also played a role: since the two stories were not counterbalanced, with the frog story always preceding the Chaplin video, it could be that participant fatigue may have led to less capacity to control the referential forms, although this effect may have been minimal given that the children were given ample time for preparation before their retellings of the stories. Thus, we conjecture that the difference in the story lines was the major source of differences between the two narrative elicitations.

6. Conclusions

Through the analysis of bilingual and monolingual Japanese narratives, we demonstrated that CLI can be explained as an interplay between structural features and the nature of discourse context. Further, we observed that the overspecification regarding the utilization of NPs was also evident in the bilinguals’ English narratives, which could not be accounted for by CLI. This calls for explanations from multiple perspectives, and indicates that investigations on bilinguals acquiring languages with distinct syntactic structures may also play a major role in further clarifying the unique features of referential behavior among bilingual children.

Although the current data are limited in terms of the number of participants involved, our findings contribute to the field by stressing the importance of taking both the structural features and the discourse context into consideration when analyzing the mechanism of CLI in the use of referring expressions. A more detailed analysis of the reintroduction context, including the number of competing characters and predictability of the referents, may clarify the potential differences in the referential choices of bilinguals and monolinguals.