Abstract
Aims and objectives/purpose/research questions:
We investigate how Alor Malay–Abui child bilinguals, a non-WEIRD and underrepresented population in the field, comprehend and produce reflexive and non-reflexive possessive constructions in Abui. Their profiles of language development are explored through functionalist accounts that highlight the close connection between language and domain-general learning capacities, with special emphasis on three models: Competition Model, usage-based approach, and processing determinism.
Design/methodology/approach:
A total of 17 children (experimental group) and 17 adults (comparison group) complete a forced-choice sentence comprehension task (where they listen to two sentences following a short video prompt and verbally select the sentence that best describes the clip) and an elicited production task (where they watch a series of video clips and describe what is happening in each clip in a semi-spontaneous way).
Data and analysis:
This study employs mixed-effects modelling for data analysis. The global model per experiment includes Condition (reflexive; non-reflexive) and Group (child; adult) as fixed effects and Participant and Item as random effects, with the maximal random-effects structure allowed by the model.
Findings/conclusions:
Results show asymmetric performance by bilingual children on the two construction types across both experiments. Our findings provide insight into the developing processor in bilingual children, influenced by various factors such as usage frequency, cognitive skills, and task effects, within the functionalist framework.
Originality and significance/implications:
The findings of this study align with functionalist accounts for language development, illuminating the intricate deployment of linguistic knowledge in the minds of bilingual children interfacing with language-usage experience and domain-general factors. By combining the three models of functionalism, our findings are more properly understood, highlighting their complementary nature in addressing developmental trajectories of linguistic knowledge for child bilinguals. Importantly, our study shifts our attention to a unique bilingual context that has not been explored – lesser-studied languages and learning environments, contributing to offsetting the sampling bias persistent in the field.
Introduction
Most studies on (bilingual) child language development have focussed on languages spoken in WEIRD (Western, Education, Industrialised, Rich, and Democratic) communities (Kidd & Garcia, 2022; Nielsen et al., 2017). Few have explored languages spoken in minority communities where the mode of acquisition differs notably (cf. Adamou, 2021). The present study addresses this gap by examining child bilinguals from a small-scale speech community in Indonesia who speak both an Austronesian (Alor Malay) and a Papuan (Abui) language. These languages are genealogically unrelated and have typological profiles that differ in the linguistic feature under investigation.
Alor Malay–Abui bilinguals learn two typologically distinct languages which have only come into contact with one another in the last 60 years. Abui, a vernacular Papuan language from the Timor-Alor-Pantar family, is spoken by approximately 41,000 people on the island of Alor in eastern Indonesia (Klamer, 2017; Kratochvíl, 2007). Abui is under threat from Alor Malay, the regional lingua franca and a variety of Eastern Indonesian Vehicular Malay, which is the target of language shift for many local languages of Eastern Indonesia (Klamer, 2017; Kratochvíl, 2007; Saad, 2020). Since the 1980s, Abui children are raised primarily in Alor Malay while they gain passive knowledge about Abui through overhead speech; they learn Abui later in adolescence or early adulthood, mostly through input from family and older peers (Kratochvíl, 2007; Saad, 2020; Saad et al., 2019). This pattern of acquisition, recently termed late vernacular production, is underreported in both the bilingualism and language shift literature (Saad et al., in press) and further strengthens the case for studying bilingual processing in non-WEIRD populations (Adamou, 2021).
We specifically investigate how bilingual children with this unique pair of languages cope with (non-)reflexive possessive constructions in Abui. These constructions provide an intriguing ground for testing the interactive nature of usage frequency with non-linguistic factors in the course of language development. Several studies have explored the L2/Ln acquisition of reflexive and non-reflexive possessives in Scandinavian languages by speakers of other Indo-European languages. For example, given that Norwegian distinguishes between reflexive and non-reflexive possessive pronouns, research on adult German- and French-speaking learners of Norwegian (Fabricius-Hansen et al., 2017; Helland, 2017) has shown that these learners struggle with using possessive pronouns correctly. They often overgeneralise, applying pronouns that resemble those in their native languages (e.g., German sein and Norwegian sin) to both reflexive and non-reflexive contexts, due to the phonological similarities between their native language and Norwegian. These studies highlight the challenges associated with the late acquisition of reflexive possessives.
Three models of functionalist accounts for language development
Bilinguals’ linguistic knowledge is deeply intertwined with usage experiences in both dominant and non-dominant languages and cognitive-psychological factors (Bialystok et al., 2008; Hurtado et al., 2014; Paradis et al., 2017; Unsworth et al., 2019). Given this nature, their language profiles are often explored through functionalist accounts that highlight the close connection between linguistic knowledge and domain-general learning capacities. According to these accounts, language acquisition involves learning conventionalised form–function pairings, which frequently exhibit many-to-many associations; learners must navigate various factors to acquire and apply these pairings (Bates & MacWhinney, 1989; Goldberg, 2019; Hilpert & Diessel, 2017; Langacker, 1987; O’Grady, 2015). Amongst the diverse landscape of functionalist accounts, we consider three influential models: Competition Model, usage-based approach, and processing determinism. While these models differ regarding the nature and role of language input, all of them emphasise the significant connection between linguistic properties – particularly the mapping relations between form and meaning/function – and non-linguistic, domain-general factors in shaping the trajectories of linguistic knowledge, complementing each other within the functionalist framework.
The Competition Model (MacWhinney, 1987) elucidates how the nature of linguistic cues involving form–function pairings shapes language development. It posits a direct mapping between surface forms and meanings/functions of utterances, suggesting that changes in cue strength drive the language acquisition processes (Bates & MacWhinney, 1989). Cue validity, the core construct for cue strength, includes cue availability (whether a cue is present when needed), cue reliability (whether a cue points to the correct outcome), and conflict validity (the reliability of a cue in competition with other cues; Bates & MacWhinney, 1989). Language acquisition progresses from cue availability to cue reliability and conflict validity as learners refine form–function associations amid cue competition (Bates & MacWhinney, 1989).
The usage-based approach emphasises the crucial role of input in shaping linguistic knowledge (Ambridge et al., 2015; Goldberg, 2019; Tomasello, 2003). Learners develop linguistic constructions through exposure to a vast array of linguistic items stored in memory, forming clusters of form–function pairings (Goldberg, 1995) influenced by frequency and distributional properties (Ambridge et al., 2015; Tomasello, 2003). By extracting similarities across these clusters, learners develop complex linguistic systems (Dąbrowska, 2008; Goldberg, 2019; Langacker, 1987), with some clusters becoming more dominant than others (Hilpert & Diessel, 2017). The impact of frequency effects on learning is also moderated by factors such as the nature of form–function mapping and the informativeness of linguistic environments (Dąbrowska, 2008; Shin, 2021; Shin & Mun, 2023; Slobin, 1985; Stromswold et al., 1985).
Processing determinism highlights two forces for language development: external pressures from language exposure and internal efficiency-related pressures from real-time information-processing mechanisms. This approach argues that the latter exerts a stronger influence on the language acquisition processes than the former (O’Grady, 2015). The processor aims to reduce working memory load (Hawkins, 2004; O’Grady, 2005) and streamline form-to-meaning/function mapping, resolving dependencies promptly (O’Grady, 2013, 2015). While frequency of occurrence remains significant, it operates within the processor–working-memory interface, where computational efficiency is enhanced through automated processing routines derived from frequent language use (O’Grady, 2013). When input alone cannot explain language acquisition, the processor guides learners towards specific developmental pathways (O’Grady, 2021), bridging the gap between exposure and linguistic knowledge.
(Non-)reflexive possessive construction in Abui and Alor Malay
Possession in Abui is expressed as a prefix on the possessed noun. There are two third-person possessive prefixes which differentiate the reflexivity of the possessor. When the possessor of the object noun is the subject of the clause, the reflexive possessive dV- is used as in (1). When the possessor of the object noun is not the subject, the non-reflexive possessive hV- is used as in (2).
1
(1) Reflexive possessive construction in Abui neng nuku di
man one 3. ‘A mani is pulling hisi friend’. (lit. ‘A man is pulling his own friend’.) (2) Non-reflexive possessive construction in Abui neng nuku di
man one 3. ‘A mani is pulling hisj friend’. (lit. ‘A man is pulling someone else’s friend’.)
The two prefixes in Abui differ in their frequency of use. Analysis of natural speech data (approximately 2,600 words) collected by Saad (2020) revealed that the non-reflexive possessive prefix hV- occurs more frequently (267 instances) than the reflexive possessive prefix dV- (30 instances). This discrepancy arises from the differing contexts in which these prefixes are used. The prefix dV- is restricted to object NPs, and specifically those expressing reflexive possessive relations as in (1) (see also Saad, 2020). In contrast, hV- appears in at least three contexts: it can be used on subject NPs (e.g.,
Notably, when describing situations involving only two participants with an unknown relationship, Abui tends to use the reflexive possessive. This implies a default assumption of a connection between the two participants rather than a connection to a third, unknown referent. For instance, when viewing clips with two participants (e.g., a man pulling another person), Abui speakers consistently describe the scene as A man is pulling his (own) friend, rather than A man is pulling his (someone else’s) friend, assuming that the two people must be friends. Without evidence of a third participant, speakers infer a relationship between the two individuals (e.g., friendship, father–son), leading to the use of the reflexive possessive prefix. This tendency potentially influences the selection of items for the reflexive possessive condition.
Possession in Alor Malay is expressed using a generic third-person possessive construction and does not differentiate between reflexive and non-reflexive relations. This possessive takes either the long form dia punya or (various) shorter ones such as dia pung, depu, and dep. These can refer either to the subject of the clause or to a referent outside the clause as the possessor as in (3).
(3) Generic possessive construction in Alor Malay Laki-laki satu ada tarik
man one ‘A mani pulls hisi/j friend’.
In sum, the possessive system in Abui is morpho-syntactically more complex compared to Alor Malay. Abui distinguishes between reflexive and non-reflexive possessive constructions using two different prefixes, whereas Alor Malay employs a single generic possessive prefix. In addition, the two prefixes in Abui differ substantially in usage frequency: the non-reflexive possessive prefix hV- is used more frequently than the reflexive possessive prefix dV-.
The current study
With these in mind, we examine how Alor Malay–Abui bilinguals comprehend and produce (non-)reflexive possessive constructions in Abui, with a specific focus on the two prefixes that manifest contrasting usage frequency and interface with the comparable structure in the dominant language of their communities. We pursue this research question by building upon Saad et al. (2019) which examined variations in the use of the (non-)reflexive possessive within the same population. Their study addressed the following three questions:
is there variation in third person possessive constructions across the four age-groups of Abui bilinguals?; how are age and gender linked to the variation?; what do differences in production and comprehension tell us about the speakers’ knowledge of the reflexivity distinction? (p. 12)
The discussion focussed on why (pre)adolescents exhibited greater variation in production than their older peers. Furthermore, a gender effect was found, with (pre)adolescent males struggling to produce the reflexive possessive compared to (pre)adolescent females.
The present study extends Saad et al. (2019) with three key differences: theoretical premise, participant cohort, and statistical analysis. Our work diverges from Saad et al.’s (2019) focus on language contact and its implications for language variation, change, and endangerment. Instead, our study is grounded in a theoretical framework, specifically addressing hypotheses derived from functionalist approaches to language acquisition, as detailed below. This enables more theory-grounded explanations of bilinguals’ comprehension and production of these constructions. Methodologically, we depart from Saad et al. (2019) by organising participants into three distinct age groups – children (formerly [pre-]adolescents), adults, and elders – to ensure clear age boundaries between these groups. We thus excluded the group previously labelled as young adults in Saad et al. (2019) from the current study. In addition, in Saad et al. (2019), not all selected participants performed both the production and comprehension task; there were some individuals who only performed the production task while others only performed the comprehension task. In this study, only participants who completed both comprehension (forced-choice sentence selection) and production (elicited production) tasks were included, albeit with a slightly smaller sample size compared to Saad et al. (2019): 17 adults in the present study versus 19 adults in Saad et al. (2019). Finally, we employed mixed-effects modelling, a common statistical method in this field, to allow more robust analyses of the data than conducted in Saad et al. (2019).
Based on the three models of functionalism and the linguistic properties of (non-)reflexive possessive constructions in Abui, several predictions could be made regarding Alor Malay–Abui bilingual children’s comprehension and production of these construction types. In comprehension, we anticipate comparable accuracy rates across both possessive constructions for children. They are likely to grasp the non-reflexive possessive well due to their frequent use in describing possession relations, aligning with the usage-based approach. Despite the relative scarcity of encountering the reflexive possessive in their linguistic input, children will demonstrate adept comprehension of this construction. Their success in comprehension can be attributed to two key factors. One factor is the high cue validity of the reflexive possessive prefix dV- in Abui, as suggested by the Competition Model. While this prefix engages in strong interference from its counterpart in Alor Malay (which dominates overall exposure), it manifests high cue validity due to its strong form–function association derived from its rather limited usage contexts. The other factor pertains to the developing processor’s inclination towards rapid resolution of dependencies – specifically the antecedent–referent relation in this case – to reduce the burden on working memory, as indicated by processing determinism. Notably, young monolingual children often exhibit higher success rates in comprehension with reflexive pronouns than plain pronouns (Chien & Wexler, 1990), despite the latter’s greater frequency in caregiver input (O’Grady, 2015, 2021). This suggests an efficiency-related internal pressure favouring reflexivity. Unlike plain pronouns, reflexive pronouns necessitate only a local antecedent within the same clause, reducing processing costs. We thus expect this processing advantage associated with interpreting reflexivity will greatly aid children in comprehending reflexive possessives.
In production, we expect children to exhibit asymmetric performance in the two constructions due to increased cognitive load and interference from the dominant language, Alor Malay. Two contrasting scenarios emerge. One scenario predicts lower success rates for the non-reflexive possessive compared to the reflexive possessive. The consistent form–function mapping of the reflexive possessive prefix dV-, as guided by the Competition Model, may help children manage the complexities involving production as well, despite its infrequent use. In addition, phonological similarities between the Abui reflexive possessive prefix de-/da- and Alor Malay generic possessive prefixes dia punya/dep could lead to overgeneralisation, as observed in other language learners (Helland, 2017). In contrast, the multifunctionality of the non-reflexive possessive prefix hV-, which covers wide usage contexts, weakens its cue validity, as explained by the Competition Model. This reduction in cue validity may hinder children’s access to the target form and complicate the management of interference from Alor Malay knowledge during production. The alternative scenario predicts that children will struggle more with the consistent use of the reflexive possessive compared to the non-reflexive possessive. Increased production demands may impair their ability to inhibit non-target linguistic knowledge, leading them to rely more on frequency-based knowledge rather than computation-based knowledge. This scenario is consistent with the usage-based account, and further fits into the good-enough processing architecture wherein the processor favours computationally less costly options drawn from memory (Ferreira, 2003). Consequently, children may over-utilise non-reflexive possessives, predominantly employing the hV- prefix.
We test these predictions by using data from Saad et al. (2019), which serves as the baseline study for our current work and is available via an open-access data repository. 3 For comprehension (‘Study 1: forced-choice sentence comprehension’ section), participants completed a forced-choice sentence comprehension task after watching a brief video prompt. For production (‘Study 2: elicited production’ section), participants engaged in an elicited production task by describing a series of video clips. We acknowledge the study’s limitations, including a small sample size, a limited number of test stimuli, and restricted data collection procedures, which stem from the small population size and reliance on the dataset provided by Saad et al. (2019).
Study 1: forced-choice sentence comprehension
Methods
Participants
A total of 34 participants – 17 children (experimental group) and 17 adults (comparison group) – joined this study (Table 1). These numbers align with the recommended sample size for cross-sectional experimental studies of language acquisition in small-scale communities, which is 10–12 participants per group (Meakins et al., 2018).
Information about participants.
Note. There was also an elder group consisting of nine individuals aged 40 to 75 years with Abui-skewed early language exposure, serving as a reference group in terms of the model use of Abui regarding the target knowledge. This group was not included in the main analysis.
Stimuli
We used 30 items, including 6 with possessive constructions (Appendix 1) and 24 fillers. Amongst the six possessive items, three involved reflexively possessed objects (reflexive condition) and three involved non-reflexively possessed objects (non-reflexive condition). Each item consisted of a pair of sentences describing the same scene, differing only in the use of a prefix, as illustrated in (1) and (2). Five test items were adapted from the Surrey Stimuli video elicitation task (Fedden & Brown, 2017), while Item X1 was specifically created to elicit a non-reflexively possessed object.
The clips for the non-reflexive condition items clearly depicted subjects acting on unlinked objects (e.g., ‘A man is lying down. A banana falls on his stomach’). Although the reflexive condition clips may not explicitly show the intended instances, there is strong evidence supporting their validity. For example, in P15, which includes only two participants – a man holding a child – it is more likely that the child belongs to the man rather than to another unknown individual, thereby eliciting a reflexive possessive. 4 This is supported by elder speakers’ use of reflexive possessive prefixes when describing this clip, suggesting that the reflexive possessive is the default choice in scenarios with two participants and one action.
There were variations in the event structures and participants depicted in the stimuli (Appendix 2). For instance, Item C01 featured two human beings with no animacy hierarchy, while Item C08 was the most complex, with 14 words in the scene description. Sentences were recorded by a male native speaker prior to the experiment and were normed by two native speakers of each language for naturalness and felicity. Fillers followed a similar format: each clip was accompanied by a pair of sentences differing only across the variable in question, and tested other linguistic features such as clause linkers, agreement prefixes, and verb choices.
The test items and fillers, varying in length and complexity, were divided into two sub-lists with pseudo-randomised presentation orders to ensure that no two test items appeared consecutively. In addition, the sentences within each pair were pseudo-randomised so that the correct sentence appeared either first or second.
Procedure and analysis
Participants were invited individually to watch each clip on a laptop, listen to two sentences aurally, and then verbally select the sentence that best described the clip. Before the main session, participants completed a practice session consisting of two trials to familiarise themselves with the experimental setup; all participants successfully passed these practice trials. All sessions were audio-recorded. The task was untimed, with an average duration of approximately 15 minutes. A representative example of the task is provided in Appendix 3.
Responses were coded as 0 (incorrect) or 1 (correct) for all the conditions. For the global statistical model concerning the mean accuracy rates, all the data were submitted to linear mixed-effects modelling using lme4 (Bates et al., 2015) in R (R Core Team, 2022), with Condition (reflexive; non-reflexive) and Group (child; adult) as fixed effects (centred around the mean and deviation-coded) and with Participant and Item as random effects, with the maximal random-effects structure allowed by the design for the model (Barr et al., 2013). We also computed each model’s R2 value using Nakagawa’s R2 (Nakagawa & Schielzeth, 2013; conditional R2 considering both fixed and random effects). Alongside the global models, we investigated whether Group influenced by-item accuracy, aiming to delve deeper into participants’ comprehension behaviour. To achieve this, we employed linear regression (using Group as a predictor and accuracy as an outcome) instead of mixed-effects modelling due to model convergence issues resulting from limited observations.
Results and discussion
Figure 1 presents the results of the forced-choice sentence comprehension task by Condition and Group. Both groups performed above chance, demonstrating their success in the task. The global statistical model (Table 2) revealed a main effect of Group, with accuracy rates for the child group being significantly lower than those for the adult group, and this trend was consistent across conditions. There were no significant differences between conditions for either group. Importantly, the child group consistently showed above chance performance in both conditions, indicating a good command of both construction types in comprehension.
Results: Comprehension. X-axis: Condition; Y-axis: Accuracy. M = mean accuracy rate; SD = standard deviation. The elder group achieved 100% accuracy for both constructions.
Statistical model (global; α = .05).
Note. R2 = 0.182. SE: standard error.
p < .001.
We also conducted by-item analyses (Appendices 4 and 5). While the children performed above chance on Items P15, C11, P19, and X01, they performed at chance on Items C01 and C08 (ps > .05). This by-item discrepancy in performance was also found in the statistical models: we found a main effect of Group only in the two clips (C01; C08), indicating that the child group performed substantially worse than the adult group. This can be attributed to the item-specific nature of each clip. The composition of the target scene in Item C01 may not have been sufficiently clear for some children to discern the intended relationship between the two participants (friends) in the given event (pulling). Item C08 was notably more complex, as the actor and undergoer of the target action were switched several times throughout the clip. These factors may have influenced the children’s interpretation of the scenes and their subsequent choices, contributing to the group differences observed in the global statistical model.
In summary, despite variability across items, the children in this study exhibited a solid understanding of both constructions. This finding supports our prediction that children would show comparable success rates for both construction types.
Study 2: elicited production
Methods
Participants and stimuli
The same participants who completed the comprehension task also joined this production task. In addition to the six clips used in the comprehension task, we included 35 additional clips, resulting in a total of 41 clips for the production task (Appendix 6). This number was guided by previous studies using the Surrey Stimuli task (Fedden & Brown, 2017). As these stimuli were originally designed to test various variables related to differential argument realisation (Fedden & Brown, 2017), not all items were intended to elicit (non-)reflexive possessive constructions. Responses were semi-spontaneous, allowing participants freedom in their choice of grammatical structures, provided they described the main characteristics of the video clip.
Procedure and analysis
Participants were invited individually to watch each scene on a laptop. They were given instructions in Alor Malay to describe what was happening in the clip using Abui (Appendix 7); they were also allowed to use Alor Malay words when they had difficulty finding the appropriate Abui words for description. The participants’ responses were video-recorded. We did not include a practice session because all participants understood the instructions clearly. The task was untimed, and the average duration was around 30 minutes.
Seven of the 41 items did not elicit any constructions involving possessed objects and were therefore excluded from the analysis (items crossed out in Appendix 6). The recordings were transcribed and annotated by six Abui native speakers trained in transcription and annotation using ELAN (Sloetjes, 2018). The annotated data were reviewed for grammaticality and felicity by three elder Abui speakers, who identified errors in (non-)reflexive possessive usage. The final data were analysed based on participants’ use of object NPs under three conditions: possession using a reflexive prefix, possession using a non-reflexive prefix, and other conditions. Responses were coded as 0 (incorrect) or 1 (correct) for each condition, and accuracy rates were then calculated. We fitted the data to linear mixed-effects models using lme4 (Bates et al., 2015) in R (R Core Team, 2022), with Condition (reflexive; non-reflexive) and Group (child; adult) as fixed effects (centred around the mean and deviation-coded) and with Participant and Item as random effects, with the maximal random-effects structure allowed by the design for the model (Barr et al., 2013). Again, we excluded the elder group from the main analysis.
Given the semi-spontaneous nature of the task, individual items could elicit more than one response per item. For example, Item C01 ‘man pulls other man’ could elicit the following four possible responses: (4a) correct use of reflexive possessive; (4b) correct use of other construction; (4c) correct use of a non-reflexive possessive; and (4d) incorrect use of a non-reflexive possessive. The difference between (4c) and (4d) is that, in (4c), the object ha-tang ‘ (4a) Correct use of reflexive possessive on object
Neng nuku de-feela ha-fik-e.
man none ‘A manj is pulling hisj friend’. (4b) Correct use of other construction
Ama ayoku ming-ta-fik.
person two ‘Two men are pulling each other’. (4c) Correct use of non-reflexive possessive on object
Di ha-tang telang-di.
‘Hej pulled hisi hand’. (4d) Incorrect use of non-reflexive possessive on object Neng nuku he-feela ha-fik-e. man one 3. ‘A manj is pulling hisi friend’.
Results and discussion
The participants’ production rates of the two construction types in the target clips were both low and varied, as shown in Table 3. This variability is attributable to the semi-spontaneous nature of the production task, which contrasts with the comprehension task requiring participants to select between two options. As a result, some participants may have generated multiple responses per item, while others might have produced none, thereby limiting our ability to draw meaningful generalisations from the results.
Results: production (by-item).
Note. The numeric values in each cell: correct responses divided by all responses. See Appendix 8 for the elder group’s performance.
We thus aggregated data from all 41 clips for analysis (Figure 2). Two major trends emerged. First, both groups produced reflexive possessives more frequently than non-reflexive possessives. Second, the child group exhibited a notably high error rate with reflexive possessives, by occasionally producing a non-reflexive possessive prefix hV- in contexts where a reflexive possessive would have been more appropriate.
Result: Production (collapsing over all items). X-axis: Condition; Y-axis: Rate of correct responses against all responses. M = mean accuracy rate; SD = standard deviation. Numeric values within each bar: correct responses divided by all responses. The elder group achieved 100% correct responses for both construction types.
The global model (Table 4) revealed main effects of Condition and Group and an interaction effect between them. Post-hoc analyses (α = .025) further revealed by-condition difference within the child group (β = 0.665, SE = 0.110, t = 6.038, p < .0005) and by-group difference within the reflexive condition (β = 0.655, SE = 0.082, t = 8.003, p < .0005). These results indicate that the child group struggled to produce the reflexive construction correctly, particularly when compared to their performance with non-reflexive constructions and the overall performance of the adult group in both conditions.
Statistical model (global; α = .05).
Note. F1 analysis due to model convergence. R2 = 0.182. SE: standard error.
p < .001.
In summary, the children in this study made numerous errors with reflexive possessives but used non-reflexive possessives accurately. This result is consistent with the second scenario of our prediction regarding children’s production of the two construction types.
Comparison of results
To investigate individual differences in children’s performance across tasks, we compared their accuracy rates in each experiment (Table 5). The comprehension task involved a fixed set of items per condition (three reflexive and three non-reflexive), while the semi-spontaneous nature of the production task allowed for varying numbers of possessive sentences to be produced.
Individual children’s performance in the two tasks.
Note. In the comprehension task, the numerator represents the number of correct responses, while the denominator indicates the total number of items. In the production task, the numerator represents the correct use of the designated prefix, and the denominator represents the total number of sentences containing possessed objects that target the constructions in question. For example, in the ‘production, reflexive’ column, Participant 47 scored 2/4, indicating correct usage of reflexive possessive prefix dV- in two out of four sentences, and non-reflexive possessive prefix hV- in the remaining two.
We found greater variability in the production of reflexive possessives amongst individual children, with accuracy ranging from none to all of the sentences produced. In contrast, variability for non-reflexive possessives ranged from zero to all sentences. Overall, accuracy rates were more variable in the production task than in the comprehension task. These trends support and extend the findings reported earlier regarding the impact of task type on performance.
General discussions and conclusion
Our study revealed that Alor Malay–Abui bilingual children demonstrated strong comprehension of both reflexive and non-reflexive possessive constructions in Abui. However, their production of these constructions varied considerably. This finding aligns with the well-documented comprehension–production asymmetry observed in language acquisition contexts (Ågren & Van de Weijer, 2013; Bornstein & Hendricks, 2012), supporting the notion that production imposes greater demands than comprehension (Litcofsky et al., 2016).
The comparable accuracy rates observed in the bilingual children’s comprehension of the two possessive constructions in Abui appear to be influenced by different factors. For the non-reflexive possessive, the frequent use of hV- in everyday speech, as noted in Saad et al. (2019), likely contributed to the children’s ability to understand this construction. This conjoins previous research emphasising the impact of exposure to linguistic environments on the development of linguistic knowledge in both monolingual (Ambridge et al., 2015) and bilingual (Unsworth et al., 2019) settings, supporting the usage-based approach. On the other hand, the children’s success in comprehending reflexive possessives is ascribable to both the high cue validity of the reflexive possessive prefix dV- and the inherent processing benefits associated with reflexivity. Despite the low usage frequency of this construction, the consistent form–function mapping of dV- strengthens its cue validity, as predicted in the Competition Model (Bates & MacWhinney, 1989), while the local nature of reflexivity facilitates efficient processing when resolving the antecedent–referent relation, as predicted in processing determinism (O’Grady, 2015). These combined factors likely contributed to the comparable success rates observed in reflexive and non-reflexive possessives in Abui. If this reasoning is valid, it suggests that the cue strength of form–function mapping and the efficiency sought by the developing processor may be as influential as usage frequency in driving bilingual children’s language acquisition patterns, pointing to the relevance of considering various – and interrelated – models of functionalist accounts to more precisely address learning outcomes.
In contrast to their comprehension performance, the bilingual children faced greater difficulty in producing reflexive possessives in Abui compared to non-reflexive ones. This discrepancy is attributable to the higher cognitive demands of production tasks relative to comprehension tasks (Bock & Griffin, 2000; Litcofsky et al., 2016). The children in our study may thus have relied heavily on frequency-based knowledge, favouring non-reflexive possessives, which are more readily accessible from memory. Consequently, they may have struggled to deploy reflexive possessives effectively when encountering scenes requiring reflexive relations. This result suggests the role of usage frequency in language acquisition, supporting the usage-based account, and implies that the cue validity of the reflexive possessive prefix dV-, along with the processing benefits associated with reflexivity, may have been compromised by the inherent demands of production tasks.
Together, these findings shed light on the developing language processor in bilingual children, which is influenced by usage frequency, cognitive skills, and task effects within the general architecture of language development. Research has long highlighted the distinct characteristics of the child processor (cf. Omaki & Lidz, 2015). It tends to favour reliable and/or available cues with a one-to-one mapping between form and function (Bates & MacWhinney, 1989; Shin & Mun, 2023). Given the global impact of language-usage experience (Ambridge et al., 2015; Tomasello, 2003), the processor is particularly sensitive to the linguistic environments in which target items appear (Dąbrowska, 2008; Shin, 2021; Shin & Deen, 2023). Furthermore, the contribution of domain-general factors to the processor’s operation is sometimes limited or less efficient (Adams & Gathercole, 2000). Especially for child bilinguals, the interplay between the two language systems and their competition, alongside their cognitive skills (sometimes in an immature state), notably influences language behaviour (Bialystok et al., 2008; Hurtado et al., 2014; Kroll et al., 2008; Unsworth et al., 2019). These factors collectively modulate how the developing bilingual processor adjusts its language activities (Ågren & Van de Weijer, 2013; Litcofsky et al., 2016; Paradis et al., 2017), aligning with the functionalist accounts for (bilingual) language development that emphasise the interplay between language-usage experience and domain-general factors (Ambridge et al., 2015; Bates & MacWhinney, 1989; Goldberg, 2019; Langacker, 1987; O’Grady, 2015; Tomasello, 2003). By combining the three models of functionalism, our findings are more properly understood, thereby highlighting their complementary nature in addressing developmental trajectories of linguistic knowledge for bilingual children.
As a final remark, since we utilised the dataset generated by Saad (2020) without modifications and did not conduct additional data collection specifically for this study, the findings should be re-evaluated in future research. First, it is essential to thoroughly investigate bilingual children’s language-usage experience. While our study highlights the significance of usage frequency in both construction types, we did not detail the children’s language-usage profiles, including the relative dominance of each language in input and use, as our focus was on experimental approaches. Future work should analyse parental interviews to assess whether aspects of child–caregiver interactions or the child-rearing environment correlate with the experimental findings. In addition, measuring children’s cognitive skills related to task performance will provide insights into their interaction with language behaviour. Examining how bilingual children apply their knowledge and skills in real-time language activities could reveal the interactive nature of their activation patterns. Our study focusses solely on the end-product of children’s language behaviour and does not address this aspect.
Despite these areas for further exploration, the implications of this study illuminate the intricate deployment of linguistic knowledge in bilingual children, interfacing with language usage, domain-general factors, and task requirements. Importantly, our study introduces a unique bilingual context, combining lesser-studied languages and linguistic items, highlighting the need for research in underexplored languages and learning environments to address sampling biases in the field.
Footnotes
Appendix
Results: Production (by-item, the elder group only).
| P15 | C01 | C11 | |||
|---|---|---|---|---|---|
| Reflexive | Non-reflexive | Reflexive | Non-reflexive | Reflexive | Non-reflexive |
| 6/6 | 0/0 | 3/3 | 1/1 | 5/5 | 0/0 |
| P19 | C08 | X01 | |||
| Reflexive | Non-reflexive | Reflexive | Non-reflexive | Reflexive | Non-reflexive |
| 4/4 | 9/9 | 1/1 | 1/1 | 2/2 | 1/1 |
Acknowledgements
We appreciate the considerable support and contribution by the Abui community, and, in particular, the Abui field station, which consists of five well-trained colleagues who contributed to the data collection and annotation. All errors and misunderstandings remain our own.
Data availability
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This publication was supported by two research grants: Operační program Jan Amos Komenský Marie Skłodowska-Curie Actions Fellowships CZ CZ.02.01.01/00/22_010/0002593 (by George Saad); Czech Science Foundation 20-18407S ‘Verb Class Analysis Accelerator for Low-Resource Languages – RoboCorp’ (by František Kratochvíl).