Cognate Effects on Bilingual Lexical–Semantic Processing in Children: Insights from ERPs

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript addresses an important gap in the bilingual (semantic) processing literature by examining whether the cognate facilitation effect (CFE) extends to semantic processing in early bilingual development. Using an auditory semantic priming paradigm and EEG recordings, the authors investigated N400 responses in Dutch-German bilingual children aged 2-6 years. The major finding is that CFE was observed only in the nDL (German), and that the magnitude of this effect is modulated by individual differences including age, proficiency, and exposure.

Overall, I found that the study is carefully designed and rigorously conducted, and the data analyses are sound. The manuscript provides valuable neurocognitive evidence extending the CFE beyond lexical access to the level of semantic processing in young bilinguals. I believe this work has the potential to make a meaningful contribution to the field. I have a few suggestions/comments that the authors may consider reflecting in their revision.

 

1. The authors are encouraged to provide a clearer rationale (or justification) for selecting the participant age range (2-6). Given that one of the motivations of the study is to clarify how age affects bilingual children’s sensitivity to CFE, it would be helpful to explain why this particular developmental window was targeted and how it aligns with theoretical or empirical expectations.
2. The primary hypothesis concerns the N400 (250–550 ms), yet the cluster-based permutation analysis did not reveal significant effects in this window. The main N400 effect emerges in ROI-based LME analyses and is statistically marginal at the omnibus level (p = .048). While the authors justify the ROI approach, the manuscript would benefit from a clearer discussion of the relative sensitivity and limitations of these analytic choices, and from more explicit acknowledgment that the N400 evidence is modest rather than robust.
3. An exploratory cluster in a later time window (approximately 700–1000 ms) is reported for the German condition, but this effect is not fully theorized. It remains unclear whether this reflects late semantic integration, reanalysis, attentional processes, or other mechanisms. Even a brief discussion of possible interpretations—or a justification for not interpreting this effect—would improve completeness.
4. The final sample size (27) used in individual-differences analyses is relatively modest given the number of predictors and interaction terms in the mixed-effects models. While understandable in developmental EEG research, this raises concerns about statistical power and potential overfitting, particularly for large interaction coefficients. The author may consider acknowledging this.

 

Minor issues

Page2, Line 72: “complementary adult evidence” should be revised (e.g., Evidence from adult studies)

 

Author Response

Comment 1: The authors are encouraged to provide a clearer rationale (or justification) for selecting the participant age range (2-6). Given that one of the motivations of the study is to clarify how age affects bilingual children’s sensitivity to CFE, it would be helpful to explain why this particular developmental window was targeted and how it aligns with theoretical or empirical expectations.

Response: We agree that clarifying the selection of the age range strengthens the theoretical framing of the study. We have revised the Introduction to explicitly state that the 2-to-6-year window was chosen to capture the developmental trajectory of the CFE during the most dynamic period of lexical acquisition. 

Revised paragraph (see page 3 of the revised manuscript; the blue color marks the revisions in the manuscript): 
The preceding review of behavioral and neural data tentatively suggests that cognates influence semantic processing; however, a systematic examination of CFE on semantic processing and development is currently lacking. To address this gap, the present study utilizes an auditory semantic-priming paradigm alongside electroencephalogram (EEG) recordings in Dutch–German bilinguals (a closely related language pair; Schepens et al., 2013). We targeted the 2-to-6-year age range to capture the developmental trajectory of the CFE during the most dynamic period of lexical acquisition (DeAnda et al., 2016). Covering this broad window allows us to examine how sensitivity to cognates evolves from early word learning through the preschool years, providing a more granular view of how bilingual experience impacts CFE and semantic processing over time. Here, we use the N400, a well-established event-related potential (ERP) index of lexical-semantic retrieval (Delogu et al., 2019), as the primary neural signature of processing. The N400 is typically larger (more negative amplitude) for semantically incongruent than for congruent word pairs and is interpreted as reflecting increased semantic-integration demands (Kutas & Federmeier, 2011). In priming paradigms, reduced N400 amplitude (i.e., less negative) for primed targets is taken as evidence of facilitated semantic access and activation of the semantic network. These features provide an online, response-independent index of cognitive processing and allow us to capture neural effects comparable to previous behavioral findings (e.g., Bice & Kroll, 2015).

Comment 2: The primary hypothesis concerns the N400 (250–550 ms), yet the cluster-based permutation analysis did not reveal significant effects in this window. The main N400 effect emerges in ROI-based LME analyses and is statistically marginal at the omnibus level (p = .048). While the authors justify the ROI approach, the manuscript would benefit from a clearer discussion of the relative sensitivity and limitations of these analytic choices, and from more explicit acknowledgment that the N400 evidence is modest rather than robust.

Response: We agree that a discussion of statistical strength is necessary. We have revised the Discussion to acknowledge this limitation.

Revised paragraphs (see page 16):
Finally, we need to acknowledge several limitations of the present study. To begin with, as we noted in the Materials and Methods, our stimulus set was drawn from a subset of items from a larger study that was not originally designed to compare semantic priming effects between cognate and non-cognate primes. As a result, the present experiment included a relatively low number of trials, which precluded strict control over the cognate status of the target words. This matters because a greater proportion of trials with non-cognate primes preceding cognate targets may have attenuated the observed cognate effect by facilitating processing in the non-cognate condition. Indeed, across both languages, the proportion of cognate targets was higher in the non-cognate prime condition (Dutch: 69% vs. 50%; German: 75% vs. 55%). 

Next, even though our final sample size (N = 27) is typical for developmental EEG research, it remains relatively modest given the complexity of the linear mixed-effects models we employed. The inclusion of multiple predictors and interaction terms increases the risk of overfitting and limits the statistical power to detect subtle effects. The modest nature of the N400 evidence (p = .048) reflects this issue; we detected this effect only in the hypothesis-driven ROI analysis and not in the more conservative cluster-based permutation test. This discrepancy highlights the trade-off between the sensitivity of hypothesis-driven ROI approaches and the strict error control of cluster-based permutation analyses, and thus, these effects should be interpreted with caution. 

Comment 3: An exploratory cluster in a later time window (approximately 700–1000 ms) is reported for the German condition, but this effect is not fully theorized. It remains unclear whether this reflects late semantic integration, reanalysis, attentional processes, or other mechanisms. Even a brief discussion of possible interpretations—or a justification for not interpreting this effect—would improve completeness.

Response: We agree that adding more details about the interpretations of the late cluster effect would make the statement clearer. We have now expanded the description of the cluster-based permutation analysis to include a theoretical interpretation of the late German effect (700–1000 ms). We now interpret this late negativity as an index of delayed semantic integration or attentional demands in the non-dominant language, which appears to be modulated by cognate status (see cited work by Sirri and Rämä( 2017) and Conboy and Mills (2006)).

Revised paragraphs (see page 8): 
As proposed in the pre-registration, we firstly performed cluster-based permutation analyses (CBPA; Maris & Oostenveld, 2007) in Fieldtrip (Oostenveld et al., 2011) to assess differences between cognate and non-cognate. In the expected N400 time window (250–550 ms), the test revealed no significant differences in either language. However, an exploratory open-window search (0–1000 ms) identified a significant negative cluster in the German condition (p = .02) from 726 to 998 ms over fronto-central electrodes (21 channels). Within this cluster, non-cognates displayed more negative amplitudes than cognates. While we initially expected the cognate effect to manifest as a reduced N400 (Kutas & Federmeier, 2011), this later time window aligns with distinct neural signatures observed in bilingual children’s non-dominant language. For instance, Sirri and Rämä (2017) found a late anterior negativity in French-Spanish bilingual children (aged 2–4), which they attributed to less efficient semantic processing. Similarly, Conboy and Mills (2006) linked late negativity to increased attentional demands in toddler word processing. Thus, the late negativity observed here may reflect delayed semantic integration in the non-dominant language, with the reduced amplitude for cognate targets indicating a reduction in these processing demands.

Despite this exploratory finding, our primary hypothesis specifically concerned the earlier N400 component. Cluster-based tests, while robust for exploratory searches, can lack sensitivity for specific effects if the signal is spatially broad or temporally fleeting (Sassenhagen & Draschkow, 2019). Therefore, to rigorously test our a priori N400 predictions, we proceeded to a region-of-interest (ROI) analysis on the mean amplitudes within the 250–550 ms window.

Comment 4: The final sample size (27) used in individual-differences analyses is relatively modest given the number of predictors and interaction terms in the mixed-effects models. While understandable in developmental EEG research, this raises concerns about statistical power and potential overfitting, particularly for large interaction coefficients. The author may consider acknowledging this.

Response: We fully agree. We have updated the section about limitations in the discussion to explicitly acknowledge that. We have combined this acknowledgment with a discussion of the modest N400 effect size to provide a comprehensive view of the statistical constraints of the study.

Revised paragraphs (see page 16):
Finally, we need to acknowledge several limitations of the present study. To begin with, as we noted in the Materials and Methods, our stimulus set was drawn from a subset of items from a larger study that was not originally designed to compare semantic priming effects between cognate and non-cognate primes. As a result, the present experiment included a relatively low number of trials, which precluded strict control over the cognate status of the target words. This matters because a greater proportion of trials with non-cognate primes preceding cognate targets may have attenuated the observed cognate effect by facilitating processing in the non-cognate condition. Indeed, across both languages, the proportion of cognate targets was higher in the non-cognate prime condition (Dutch: 69% vs. 50%; German: 75% vs. 55%). 

Next, even though our final sample size (N = 27) is typical for developmental EEG research, it remains relatively modest given the complexity of the linear mixed-effects models we employed. The inclusion of multiple predictors and interaction terms increases the risk of overfitting and limits the statistical power to detect subtle effects. The modest nature of the N400 evidence (p = .048) reflects this issue; we detected this effect only in the hypothesis-driven ROI analysis and not in the more conservative cluster-based permutation test. This discrepancy highlights the trade-off between the sensitivity of hypothesis-driven ROI approaches and the strict error control of cluster-based permutation analyses, and thus, these effects should be interpreted with caution. 

Comment 5: Minor issues: Page2, Line 72: “complementary adult evidence” should be revised (e.g., Evidence from adult studies)

Response: We have revised the sentence as follows.

Revised paragraph (see page 2):
Evidence from adult studies indicates that cognates can speed recognition and enhance downstream semantic prediction (Ito et al., 2025), and faster prime processing increases priming magnitude (Hutchison et al., 2008). 

Reviewer 2 Report

Comments and Suggestions for Authors

An important topic to research upon. Please find my feedback attached 

Comments for author File: Comments.pdf

Author Response

We have fully addressed all formatting comments (passive voice, sentence length, rephrasing, and UK/US English) in the revised manuscript (marked in blue). Below, we provide detailed responses to more elaborate theoretical, methodological and conceptual comments.

 

Abstract

• Commented [M1]: any specific qualitative or quantitative approach implemented? If yes, please mention a line in the abstract.
  
  
• Commented [M3]: a line on future research

Response: We have updated the Abstract. We now include other approaches we used and a future direction, as requested. For the future direction, we added a final sentence emphasizing the need for longitudinal research to determine if the observed cognate facilitation acts as a scaffolding mechanism during development.

Revised abstract:
This study investigates whether and, if so, how cognates facilitate lexical-semantic processing during early bilingual development. Additionally, we examine the interaction between the cognate facilitation effect (CFE) and bilingual experience factors, such as language proficiency, exposure, and age. We investigated language backgrounds and recorded event-related potentials during a semantic priming task from Dutch-German bilingual children. Most participants were Dutch-dominant, characterized by higher exposure and proficiency in Dutch. We compared the N400 response to target words preceded by semantically related cognate versus non-cognate primes. We found a reduced N400 effect (indexing cognate facilitation) only in the non-dominant language (nDL; German). Individual difference analyses further revealed that higher proficiency of nDL and increasing age attenuated the CFE. In contrast, higher cumulative exposure was associated with an amplified CFE. These findings suggest that cross-linguistic activation in lexical-semantic processing may benefit younger children with either lower proficiency or higher exposure to their non-dominant language during language processing. Together, the study offers direct neural evidence for bilingual cognate facilitation effects and highlights the importance of investigating interactions with external factors in early bilingualism. Future longitudinal research should examine whether cognate reliance serves as a temporary scaffolding mechanism for the acquisition of the non-dominant language.

 

Introduction

• Commented [M4]: It might be helpful to organize the paragraphs into specific themes and provide sub-headings for a free flow (for example, cognate facilitation effect", "age and CFE" etc)

Response: To guide the reader more effectively through the specific themes, we have restructured the Introduction and inserted distinct subheadings. The section is now organized as follows:

1. Introduction: Defines the CFE and theoretical framework.

1.1 The role of age and bilingual experience: Discusses the impact of age, exposure, and proficiency.

1.2 Cognate effects in semantic processing: Reviews the limited literature on semantic processing and indirect evidence.

1.3 The present study: Outlines the research gap and specific research questions.

We believe this thematic segmentation allows for easier navigation of the background literature.

• Commented [M8]: Consider adding brief definitions to many of these relevant terms for general audience. 

Response: We have added definitions for key technical terms to improve accessibility for a general audience. Specifically, we added a clear definition of "lexical-semantic access" in the Introduction. Furthermore, we provided explicit definitions for "cross-linguistic" and "lexical knowledge" in Section 2.2.1, where these concepts are operationally defined in the context of the Cross-linguistic Lexical Task (CLT).

Revised paragraphs:
1. Introduction (see pages 1 -2)
Cognates are translation equivalents across two languages that share meaning and overlap in phonological and/or orthographic form. Previous studies have suggested that cognates can benefit lexical development and processing in bilingual children. For example, multiple studies report that translation equivalents with form overlap tend to be learned more easily and earlier in children’s vocabularies (Bosch & Ramon-Casas, 2014; Bosma et al., 2019; Garcia-Castro et al., 2025; Goriot et al., 2021; Lindgren & Bohnacker, 2020; Mitchell et al., 2024). Experiments using picture naming, word recognition, lexical decision, and sentence-reading tasks find shorter reaction times and/or higher accuracy for cognates vs. non-cognates (Bosma & Nota, 2020; Duñabeitia et al., 2016; Iniesta et al., 2024; Koutamanis et al., 2024a; Poarch & van Hell, 2012; Schröter & Schroeder, 2016). These findings align with the Cognate Facilitation Effect (CFE). Previous studies suggested that CFE may arise due to the co-activation of overlapping semantic and phonological/orthographical representations within an integrated bilingual lexicon. Such co-activation often leads to faster lexical-semantic access, the process of retrieving words’ meanings and forms from memory (see the Bilingual Interactive Activation Plus model, BIA+; Dijkstra & Heuven, 2002).

2.2.1. Cross-linguistic Lexical Task (CLT) (see page 4)
We assessed children’s receptive vocabulary using the Cross-linguistic Lexical Task (CLT; Haman et al., 2015), part of the LITMUS test battery (https://www.bi-sli.org/litmustools). The CLT was developed for bilingual populations between 3 and 7 years of age and is available in multiple languages, making it suitable for cross-linguistic comparisons (analyses across different language systems) and within-child comparisons of lexical knowledge (the breadth and depth of a child's vocabulary). We selected the CLT because it is specifically designed for bilingual populations; the test items were carefully selected based on Age of Acquisition (AoA) and complexity to ensure cross-linguistic equivalence and minimize cultural bias. In this study, we administered the comprehension component of the Dutch (CLT-NL; Van Wonderen & Unsworth, 2021) and the German version (CLT-DE; Rinker & Gagarina, 2017). The comprehension task follows a four-picture choice format: after viewing four images, the child hears an audio prompt and selects the picture that matches the target noun or verb. Each domain (nouns, verbs) includes 32 test items. Accuracy in the CLT reflects the child’s receptive vocabulary size and provides a reliable estimate of lexical knowledge in each language, independent of production abilities. The task’s standardized design across languages enables direct comparison of lexical development across two languages. In this study, the CLT was administered in Dutch and German, respectively, following the two EEG recording sessions (see Section 2.3.2 Procedure).

 

• Commented [M11]: Consider providing specific examples of cognates and their applications or how they affect language processing 

Response: We have incorporated a concrete example to illustrate the mechanism of cognate facilitation. We introduced the English-Dutch pair rose–roos in the paragraph.

Revised paragraph (see page 2):
Some studies also addressed how cognates interact with bilingual language development by comparing CFE across age groups. Specifically, these studies examine whether the processing advantage for cognates (e.g., recognizing the word rose faster because of the Dutch cognate roos) changes as children mature. However, results have been inconsistent. For instance, two studies report that sensitivity to cognates increases with age. Bosma et al. (2019) and Goriot et al. (2021) found that in younger children (ages 5–8 and ages 4–12), the CFE tends to emerge or increase with age, which they attribute to growing linguistic awareness and greater sensitivity to cross-linguistic phonological regularities. In contrast, Duñabeitia et al. (2016) reported that in older children and adolescents (ages 8–15), the CFE decreases with age. They suggest that increased reading experience and maturing language-control mechanisms may strengthen cross-language inhibition and thus reduce CFE over time.

 

• Commented [M13]: Consider reorganizing the sections such as adding CFE within the broader context of bilingualism research

Response: We have addressed the reviewer’s concern regarding the contextualization and flow of the CFE sections through structural reorganization. As the reviewer's other comment to "organize the paragraphs into specific themes," we have introduced clear subheadings throughout the Introduction. The paragraph in question now falls specifically under Section 1.1: "The role of age and bilingual experience." We believe this new heading structure effectively achieves the reviewer's goal: it explicitly frames the subsequent paragraphs within the broader context of individual differences in bilingualism immediately before the reader encounters the specific studies. Because the section title now acts as the "contextual anchor," we opted to maintain the original paragraph wording to ensure a concise flow and avoid redundancy between the heading and the text.

Materials and Methods

• Commented [M26]: Provide more information on the demographic characteristics of participants such as their socioeconomic status or geographical location etc.
  
  
• Commented [M27]: were there any control groups such as monolingual children to be compared with the bilingual children?
  
  
• Commented [M28]: Elaborate more on the recruitment process such as informed consent from parents, if the participants met the inclusion criteria

Responses:

• Comment M26: We have updated the 2.1 Participants section to provide better demographic context.
  
  
• Comment M27: We want to clarify here that the study design does not require a monolingual control group. The primary research question focuses on the Cognate Facilitation Effect, which is, by definition, an interaction between two language systems. Furthermore, one of our research questions is to investigate whether and how bilingual experiences, like amount of language exposure and proficiency, interact with CFE. Therefore, we only focused on comparing the CFE across the participants' own two languages (Dominant vs. Non-Dominant) rather than against a monolingual baseline.
  
  
• Comment M28: We have expanded and added more procedural details in the 2.1 Participants section.

Revised paragraph (see page 4):
2.1. Participants

Forty typically developing Dutch-German bilingual children aged between 2 and 6 years (Mean = 3.99, SD = 1.07, range = 2.08 - 5.75, 15 females) participated in the study. We recruited these participants through the Baby and Child Research Centre in Nijmegen, The Netherlands. During the recruitment phase, we ensured that participants met specific inclusion criteria. Specifically, we verified that all children had normal hearing and vision, no diagnosed developmental disorders, and no significant exposure to languages other than Dutch or German. Furthermore, we confirmed that every participant received at least 10% of their language input in each of the two languages (Hoff et al., 2012; Hoff & Ribot, 2017). Geographically, most families resided in the Dutch-German border region; specifically, 37 families lived in the Netherlands, while three families lived in neighboring Germany. Caregivers reported a predominantly high socioeconomic status (SES) background. Regarding household income, 90% of families reported earning more than €3000 per month (with 75% exceeding €3800). Similarly, parental education levels were high, with 79.5% of caregivers holding a university degree. The study received approval from the Ethical Board of Social Sciences at Radboud University in Nijmegen, The Netherlands. Prior to the first experimental session, we provided caregivers with detailed information regarding the study procedures and obtained their written informed consent. Families received either €50 or €40 plus a children’s book as compensation. 

 

• Commented [M29]: how was this measured? through parent's report, observation?
  
  
• Commented [M30]: Please mention few lines on why CLT and Q-BEx was chosen

Responses:

• Comment M29: We have clarified that the Cross-Linguistic Tasks (CLT) are direct behavioral measures, not parent reports. As described in the text (Lines 177–179), the CLT uses a four-picture choice format administered directly to the child. The experimenter displays a slide with four images and presents a corresponding question (e.g., "Where is the apple?"), and the child selects the picture that best responds to the question.
  
  
• Comment M30: We have added specific justifications for the selection of our measures in the Methods section. For the CLT, we now explicitly state that we selected this tool to minimize the cultural and linguistic biases inherent in using translated monolingual instruments (Section 2.2.1). For the Q-BEx, we clarified that this tool was chosen for its ability to calculate cumulative exposure, allowing us to account for the dynamic history of language input rather than relying solely on current exposure estimates (Section 2.2.2).

Revised paragraphs (pages 4 - 5):
2.2.1. Cross-linguistic Lexical Task (CLT) 

We assessed children’s receptive vocabulary using the Cross-linguistic Lexical Task (CLT; Haman et al., 2015), part of the LITMUS test battery (https://www.bi-sli.org/litmustools). The CLT was developed for bilingual populations between 3 and 7 years of age and is available in multiple languages, making it suitable for cross-linguistic comparisons (analyses across different language systems) and within-child comparisons of lexical knowledge (the breadth and depth of a child's vocabulary). We selected the CLT because it is specifically designed for bilingual populations; the test items were carefully selected based on Age of Acquisition (AoA) and complexity to ensure cross-linguistic equivalence and minimize cultural bias. In this study, we administered the comprehension component of the Dutch (CLT-NL; Van Wonderen & Unsworth, 2021) and the German version (CLT-DE; Rinker & Gagarina, 2017). The comprehension task follows a four-picture choice format: after viewing four images, the child hears an audio prompt and selects the picture that matches the target noun or verb. Each domain (nouns, verbs) includes 32 test items. Accuracy in the CLT reflects the child’s receptive vocabulary size and provides a reliable estimate of lexical knowledge in each language, independent of production abilities. The task’s standardized design across languages enables direct comparison of lexical development across two languages. In this study, the CLT was administered in Dutch and German, respectively, following the two EEG recording sessions (see Section 2.3.2 Procedure).

2.2.2. Quantifying Bilingual EXperience (Q-BEx)

We collected children’s bilingual experience using the Q-BEx questionnaire (De Cat et al., 2023), which quantifies a bilingual child's language background and experience. We chose this tool because it captures the dynamic nature of bilingualism by estimating cumulative exposure rather than current input alone, providing a more reliable metric of the nature of the linguistic input that shapes lexical development over time. Participants' caregivers completed the questionnaire, which included modules on background information (e.g., age, family background), risk factors (e.g., language developmental delays), language exposure and use, the richness of linguistic experience, and language mixing. For the present study, we focused solely on the language exposure and use module, using the resulting cumulative estimates as an index of language exposure. To calculate cumulative exposure, caregivers first identified significant time points where a child's language exposure changed. These points defined distinct time periods, and for each period, the caregiver reported the percentage of exposure to each language. Then, we calculated the total cumulative exposure for each language by multiplying the duration of each period (in months) by the percentage of exposure to that language and summing these values across all periods. The percentage of cumulative language exposure was then determined by dividing this total cumulative exposure score for a language by the child's current age in months. In the current study, the Q-BEx was completed online by the caregiver(s) before the experiment, and took approximately 15 to 30 minutes to finish. 

 

• Commented [M36]: If any software or tools were used for assessment please specify 

Response: We have specified the software tool used for the assessment and standardization of our materials. In the Materials section (Section 2.3.1), we clarified that the pydub library in Python was used to objectively assess and normalize the acoustic intensity (RMS amplitude) of the stimuli, ensuring consistency across all experimental trials.

Revised paragraph (see page 5):
The stimuli comprised 124 nouns in Dutch and German, arranged into 62 prime-target word pairs (32 Dutch and 31 German). To ensure that the words were familiar to the children, we selected the words from the Dutch and German McArthur Communicative Development Inventory (CDI-WS) (Szagun et al., 2023; Zink & Lejaegere, 2003), with a few additional German words sourced from a Goethe Institute A1-level word list (Goethe-Institut., 2004). To ensure uniform acoustic quality and native pronunciation, stimuli were recorded in a soundproof booth by two female speakers per language (Dutch and German). All audio files were subsequently processed using the pydub library in Python to normalize intensity (RMS amplitude) (Robert, 2018).

Results

• Commented [M41]: Consider summarizing the key findings

Response: We have added a summary roadmap to the beginning of the Results section as suggested. Following the description of participant inclusion and exclusion (Section 3.1), we included a concise paragraph summarizing the three main takeaways: (1) the dominance patterns, (2) the presence of the CFE in the non-dominant language only, and (3) the modulation of this effect by individual differences. This provides readers with a clear overview before getting into the detailed statistical results.

Revised paragraph (see page 9):
This study incorporates both behavioral and neural measures, each with different sample sizes, in order to maximize participant inclusion. For the ERP analyses, we report data from 29 participants (10 female; M = 4.16 years, SD = 1.07). Eleven participants were excluded from this dataset due to an insufficient number of trials remaining after artifact removal (e.g., blink, muscle noise, and head movements). For the subsequent individual-level analyses, which combine ERP and behavioral data, further exclusions were necessary: two participants due to incomplete Q-BEx questionnaires (final N = 27) and one due to missing CLT task data (final N = 28). As a result, the analyses in Sections 3.1 and 3.3 rely on different sample sizes. We included 27 participants in the language exposure and age model, but 26 participants in the language exposure and proficiency model. This variation reflects the availability of data for each specific predictor.

In summary, the results confirmed that the majority of participants were Dutch-dominant based on language background measures. Neural data indicated a CFE, indexed by reduced N400 effect, only in the non-dominant language (German). Finally, linear mixed-effects modeling revealed that the effect was modulated by individual differences in age, language proficiency, and cumulative exposure, respectively. We detail these findings in the sections below.

 

• Commented [M42R41]: When discussing participant exclusions, provide more context about the nature of the artefacts or issues that led to their exclusion, which can help inform the reader about the robustness of the results

Response: We have clarified the nature of the artifacts that led to participant exclusion. In the Results section (Section 3.1), we now specify that exclusions were driven by physiological artifacts, specifically blinks, muscle noise, and head movements. This addition provides the requested context regarding the data quality issues encountered.

Revised paragraph (see page 9):
This study incorporates both behavioral and neural measures, each with different sample sizes, in order to maximize participant inclusion. For the ERP analyses, we report data from 29 participants (10 female; M = 4.16 years, SD = 1.07). Eleven participants were excluded from this dataset due to an insufficient number of trials remaining after artifact removal (e.g., blink, muscle noise, and head movements). For the subsequent individual-level analyses, which combine ERP and behavioral data, further exclusions were necessary: two participants due to incomplete Q-BEx questionnaires (final N = 27) and one due to missing CLT task data (final N = 28). As a result, the analyses in Sections 3.1 and 3.3 rely on different sample sizes. We included 27 participants in the language exposure and age model, but 26 participants in the language exposure and proficiency model. This variation reflects the availability of data for each specific predictor.

 

• Commented [M45]: while discussing language background and proficiency scores, may be add a sentence interpreting the dominance in Dutch 

Response: We have refined the Results section to emphasize that the observed Dutch dominance aligns with the sociolinguistic context of the sample, as the majority of participants reside in the Netherlands. This addition clarifies the link between the background measures and the environmental language exposure.

Revised paragraph (see pages 9 - 10):
Descriptive results for the participant’s cumulative language exposure, as measured by the Q-BEx, and lexical proficiency in Dutch and German, assessed using the LITMUS Cross-linguistic Lexical Task (CLT), are presented in Table 2. Overall, both language background measures reveal a clear pattern of Dutch, which aligns with the demographic context that the majority of families reside in the Netherlands, where Dutch is the societal language. Consequently, German represents the non-dominant language for our sample on average. Based on the Q-BEx questionnaire (N = 27), twenty-two participants received more cumulative exposure to Dutch than to German, three were predominantly exposed to German, and two had balanced exposure. The CLT score (N = 28) also demonstrates an overall better proficiency in Dutch. Specifically, the accuracy for nouns is better than for verbs, regardless of language. No significant difference was observed in participants' noun proficiency across languages (t(24) = 0.16, p = .88), whereas verb proficiency was significantly higher in Dutch (t(24) = 5.52, p < .001). 

 

• Commented [M46]: may be add a chart or table on the findings of language proficiency 

Response: We have clarified the location of the proficiency findings. While Table 4 details the main ERP results, we present the specific interactions between language proficiency and the EEG effects in Table 5, located within the "3.3. Individual Differences" section. To ensure readers can easily locate these findings, we added a cross-reference in the text near Table 4 pointing directly to Section 3.3 and Table 5.

Revised paragraph (see page 11):
A linear-mixed effect model was performed to investigate whether the N400 mean amplitude measured within the 250 to 550 ms time-window differed between cognates and non-cognates in both Dutch and German. The mixed-effects model showed a marginal main effect of Cognate status (model coefficient for Non-cognate − Cognate = −1.86, p = .048), which corresponded to cognates eliciting, on average, an effect that was 1.86 µV more positive than non-cognates (i.e., a smaller N400 response for cognates). In addition, the model revealed several interactions affecting N400 amplitude. The Cognate × Language interaction (β = 1.23, p = .048) indicated that the reduction in N400 for cognates versus non-cognates was reduced in Dutch. The Cognate × Topography interaction (β = 2.05, p < .001) showed that the N400 difference between conditions was smaller at posterior electrodes than at frontal–central electrodes. Language also interacted with Topography (β = 1.71, p = .004), reflecting language-dependent differences in anterior–posterior amplitude. The three-way interaction between Cognate status, Language, and Topography did not reach the significance threshold (β=−1.65, p=.053). However, the result suggested a trend where the reduction of the cognate effect in Dutch (compared to German) was stronger over posterior electrodes. Overall, cognates elicited a smaller N400 than non-cognates (≈1.9 µV), but this N400 difference is substantially modulated by both topographic distribution and language. Figure 1 displays the EEG waveforms. Table 4 details the fixed effects for the main ERP analysis of the semantic priming experiment. We report the subsequent analyses regarding how language proficiency and exposure modulate these effects in the 'Individual Differences' section (Section 3.3; see Table 5).

Discussion

• Commented [M47]: In this section, highlight on the practical implications of the findings in more depth.
  
  
• Commented [M49]: Elaborate on the future findings in more details 

Response: We have added a paragraph to the Discussion addressing these two comments together. Given the modest statistical strength of the findings (p = .048) and the sample size limitations we previously noted, we have framed the practical implications cautiously. Rather than suggesting immediate changes to educational practice, we propose that our findings serve as a theoretical basis for future translational research (e.g., testing whether cognates effectively scaffold vocabulary learning in intervention settings).

Regarding future directions, we now explicitly list: Longitudinal designs to distinguish between developmental shifts and proficiency effects. Cross-linguistic comparisons involving language pairs with varying degrees of typological distance to disentangle phonological from semantic drivers of the effect.

Revised paragraph (see pages 16 - 17):
Despite the limitations regarding statistical power and sample size, these findings offer a theoretical basis for future research. If the CFE serves as a scaffolding mechanism for the non-dominant language, as our data tentatively suggest, this dynamic could eventually inform strategies in bilingual education. Consequently, future research should expand on these findings to better understand the developmental trajectory of the CFE. Beyond the necessity for larger, high-powered replications, longitudinal designs are essential to map how the reliance on cognates shifts over time within the same individuals. Such studies could determine whether the age-related reduction in CFE we observed represents a true developmental shift in lexical organization or simply a byproduct of increasing proficiency. Furthermore, to disentangle the role of phonological overlap from semantic overlap, future work should examine these effects in bilinguals acquiring language pairs with varying degrees of language distance (e.g., English-Dutch versus English-Spanish). Comparing these populations would clarify whether the scaffolding mechanism relies strictly on form similarity or if conceptual links alone are sufficient to drive cross-language activation in the developing brain.