Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Thank you for the opportunity to review this interesting paper. I found it very well written and  theoretically interesting. I think that it makes an important contribution to the field. I have only one main concern, related to a potential confound for the "socialness" lexical rating. To me, it seems as though defining "socialness" based on the number of agents/actors might actually be identifying underlying argument structure (which is necessarily more complex for verbs that involve more than one actor). In the text, the examples of catch and throw, aside from requiring two people (i.e., more socialness), also require more complex sentence constructions than sleep for example. I wonder, do socialness ratings correlate strongly with the number of arguments required (for verbs)? There is a rather large literature on argument structure and word class that should be considered here and a stringent test of "socialness" should account for this body of work. 

Author Response

Reviewer 1 

Comment 1: Thank you for the opportunity to review this interesting paper. I found it very well written and theoretically interesting. I think that it makes an important contribution to the field. I have only one main concern, related to a potential confound for the "socialness" lexical rating. To me, it seems as though defining "socialness" based on the number of agents/actors might actually be identifying underlying argument structure (which is necessarily more complex for verbs that involve more than one actor). In the text, the examples of catch and throw, aside from requiring two people (i.e., more socialness), also require more complex sentence constructions than sleep for example. I wonder, do socialness ratings correlate strongly with the number of arguments required (for verbs)? There is a rather large literature on argument structure and word class that should be considered here and a stringent test of "socialness" should account for this body of work.  

• Response 1: Thank you for raising this insightful concern. To directly address whether socialness overlaps with argument-structure complexity, we conducted an additional analysis on pages 9 and 10 we provide the following results. Notably though, you will see that when addressing another reviewer comment, we added analyses that separately examined the predictive role of socialness scores on word acquisition data (vocabulary size of acquisition – VSoA) for nouns and verbs separately. In these new analyses, we found that socialness ratings did not predict verb acquisition data; thus, the concern of argument structure confounds are lessened when considering socialness as a word feature that is associated with vocabulary acquisition: 

• Results: “As socialness may also relate to grammatical complexity in verbs, we next examined whether socialness rating scores differed between transitive verbs (n = 21) and verbs that could be classified as transitive or intransitive or only as intransitive (n = 19). Socialness ratings did not statistically differ (p = 0.685). We also conducted a t-test to compare socialness rating scores for verbs that could be transitive or intransitive (n = 27) with socialness ratings for verbs that are ditransitive (n = 11); socialness ratings did not differ (p = 0.647). Thus, verbs that involved more complex argument structure (e.g., ditransitive verbs, e.g., show, give) did not systematically receive higher socialness ratings than verbs that can serve as intransitive verbs, with simpler argument structure (e.g., wait, wish). This indicates that socialness ratings do not simply reflect argument-structure complexity.” 

• We also explain our approach to examining argument structure in the methods section on page 7. 

“As previously noted, Horvath et al. (2018) suggested that verbs with multiple event participants may be learned later in development because they tend to appear in sentences that have more complex argument structure (i.e., intransitive verbs in sentences and transitive verbs in sentences). As such, we used Horvath et al.’s verb characteristics table to classify our 40 verbs as intransitive and transitive. Given that the vast majority of the verbs were classified as transitive (38 out of the 40 words), we also documented whether could be classified as intransitive or transitive (n = 17), depending on the sentential context. In addition, we documented whether the verbs could be classified as ditransitive (11 of the 40 verbs; Horvath et al., 2018). This allowed us to examine whether socialness features were associated with transitivity classification.” 

Reviewer 2 Report

Comments and Suggestions for Authors

Dear Authors,

I am pleased to provide my review of the manuscript titled “The Influence of Social Word Features on Early Word Learning in Autistic and Non-Autistic Children.” This is a thoughtful and well-executed study that tackles an important question in developmental psycholinguistics, how lexical “socialness” shapes word learning across neurotypes. I found the use of the Vocabulary Size of Acquisition (VSoA) conceptually innovative and methodologically elegant. Anchoring acquisition to vocabulary size rather than chronological age offers a fairer and more developmentally sensitive comparison for populations with heterogeneous learning trajectories. The analytic approach is transparent, the data sources are openly available, and the findings, showing that socialness predicts later acquisition while frequency predicts earlier learning, fit neatly with existing empirical expectations. Importantly, the authors interpret the absence of group differences with appropriate restraint, framing this similarity as evidence of overlapping mechanisms rather than a simple null effect.

That said, several conceptual and methodological issues limit the interpretive strength of the current version and should be addressed before the manuscript is suitable for publication. My comments below are intended to be constructive and to help the authors strengthen both the theoretical framing and the empirical foundation of their conclusions.

Major weaknesses

The most significant theoretical gap is the near absence of discussion of the child’s language environment. The manuscript treats the socialness effect primarily as a function of internal cognitive processing, but an equally plausible explanation is input-based: children, particularly autistic children, may simply encounter fewer highly social words in their everyday linguistic environments. The reliance on CHILDES as a general frequency corpus does not capture potential group-specific differences in caregiver input. Without addressing this, it is difficult to attribute the observed effects solely to cognitive processing differences rather than exposure disparities.

A second conceptual limitation concerns the operationalization of socialness. The authors treat the Diveica et al. (2023) ratings as a single, continuous dimension. From a cognitive neuroscience perspective, this is likely an oversimplification. “Socialness” encompasses multiple subcomponents, agency, affective valence, cooperation versus competition, that are supported by partially distinct neural systems. Reducing this multidimensional construct to a single score may obscure meaningful variability in how different kinds of social meaning influence acquisition. Additionally, the predictor is derived from adult judgments. Adults’ conceptualizations of socialness may not mirror children’s early semantic representations, which are often grounded in perceptual and action-based schemas rather than abstract social roles. This raises questions about construct validity and the developmental relevance of the measure.

Methodologically, the study omits several key lexical control variables that strongly predict age or vocabulary size of acquisition, concreteness, imageability, grammatical class, word length, phonotactic probability, and contextual diversity. Because socialness is likely correlated with many of these, especially concreteness and word class, the unique contribution of socialness cannot yet be isolated. For instance, verbs are typically both more social and acquired later than nouns, suggesting that the observed effect may partly reflect syntactic class rather than social content per se. Controlling for these lexical dimensions, or at least examining partial correlations, would strengthen confidence in the interpretation.

The lexical scope of the dataset also warrants caution. The analysis is restricted to 170 CDI words, a relatively small and early-developing subset. This narrow range may compress both socialness values and acquisition variance, reducing sensitivity to group differences. In its present form, the study offers insight into early-emerging vocabulary but not necessarily into later or more abstract lexical development.

Finally, while the authors briefly acknowledge conflicting findings from Jiménez et al. (2021), the reconciliation is superficial. The authors could make more of their methodological advantage: a word-level VSoA analysis arguably provides a finer-grained index of acquisition order than average socialness ratings across bins. This point deserves stronger elaboration to clarify how the current work advances beyond prior literature.

Weaknesses in the Discussion Section

The Discussion section is conceptually the weakest part of the manuscript. It largely restates empirical findings and offers generic explanations such as “greater cognitive load” or “event complexity.” These explanations are plausible but insufficiently grounded in established theories of semantic, social, or neural development. This section also overlooks a key limitation: that the “socialness” measure is based on adult norms. This mismatch between adult conceptualization and child experience is a non-trivial concern that should be explicitly discussed. Likewise, the authors do not sufficiently consider that the relationship between socialness and acquisition timing might reflect correlated lexical features such as abstractness or syntactic complexity. The interpretation of the null group difference could also be deepened. Rather than simply stating that autistic and non-autistic children show similar learning profiles, the authors could discuss potential mechanisms for this convergence, perhaps shared reliance on statistical or distributional learning processes, or constraints inherent to early lexicon formation that are robust across neurotypes. The current discussion does not explore these alternatives.

Minor Issues (Clarity and Presentation)

A few smaller points could improve readability and transparency. The category labeled “other word types” should be clearly defined and enumerated. Certain sentences contain minor typographical issues (e.g., Line 244). The model specification would benefit from fuller reporting, how the Group variable was coded, whether predictors were centered or standardized, and how missing data were handled.

Overall Assessment

This manuscript presents a novel and valuable approach to examining lexical acquisition across neurotypes, and the use of VSoA is a real strength. However, the study’s interpretive reach is currently limited by conceptual oversimplification of “socialness,” insufficient control for confounding lexical variables, and a discussion that does not fully engage with theoretical mechanisms or alternative explanations. With substantial revision, particularly expanding the theoretical framework, incorporating additional lexical controls, and addressing input-based explanations, the paper could make a meaningful contribution to our understanding of social and linguistic development in autism.

Author Response

Reviewer 2 

Comment 1: The most significant theoretical gap is the near absence of discussion of the child’s language environment. The manuscript treats the socialness effect primarily as a function of internal cognitive processing, but an equally plausible explanation is input-based: children, particularly autistic children, may simply encounter fewer highly social words in their everyday linguistic environments. The reliance on CHILDES as a general frequency corpus does not capture potential group-specific differences in caregiver input. Without addressing this, it is difficult to attribute the observed effects solely to cognitive processing differences rather than exposure disparities. 

• Response 1: Thank you for highlighting this important conceptual issue. We agree that children’s unique language environments are highly likely to be associated with word learning. Given that the groups did not differ in the association between group-specific VSoA data and socialness ratings, this concern is minimized. If there had been a significant interaction between socialness and group, the concern for input differences would be heightened.  

• We have added content to the discussion section to note that group differences in input could have served as a potential causal factor had we seen a difference in the socialness effect between the groups. See page 15 in section 4.2 for this added content please.  

• We also added the following statement to the limitations section on page 16 “Longitudinal studies could also explore how caregivers adjust the social complexity of their language as children’s vocabulary develops (see Perry et al., 2021 for a similar approach when examining iconicity). Such a study could clarify how caregivers adapt their input over time and identify strategies that support learning socially complex words. Such a study could also examine whether caregivers use social words at different rates when interacting with autistic versus non-autistic children.” 

• Additionally, we now specifically address the limitations of relying on CHILDES as a general frequency corpus. Because, to our knowledge, the field does not currently have a rich corpus of parent-child transcripts from dyads that include an autistic child, we are not able to add in group-specific frequency data in our analyses. Though the ASDBank resource for example does exist, it currently contains a very small number of transcripts that do not have optimal coverage of the words that appear on the CDI. This acknowledgment now appears in Limitations and Future Directions (Section 4.4).  

“Third, in the current study, we controlled for word frequency using frequency estimates derived from CHILDES; while this is a rich source to better understand the language environments that children learn within, the children in this corpus are primarily believed to be typically developing. Our attempt to control for input frequency effects could be strengthened if we could include group-specific frequency data. Currently though, a large and public corpus that is specific to autism does not exist (though we know that researchers are working to develop such a resource in the future and have started to share small sets of parent-child transcripts; e.g., ASDBank; MacWhinney & Fromm, 2022).” 

 

Comments 2: A second conceptual limitation concerns the operationalization of socialness. The authors treat the Diveica et al. (2023) ratings as a single, continuous dimension. From a cognitive neuroscience perspective, this is likely an oversimplification. “Socialness” encompasses multiple subcomponents, agency, affective valence, cooperation versus competition, that are supported by partially distinct neural systems. Reducing this multidimensional construct to a single score may obscure meaningful variability in how different kinds of social meaning influence acquisition.  

• Response 2: Thank you for this point, we have added content to the Introduction and Discussion sections to acknowledge this limitation. For instance, in the Introduction (p. 4), we now state: 

“The Diveica et al. (2023) ratings provide an adult-derived index of socialness that does not distinguish among these components. The Diveica et al. ratings were collected from adult participants, who judged each word on how socially relevant its meaning was, using a 7-point scale. Thus, highly socially relevant words – words with high socialness ratings – would include “a social characteristic of a person or group of people, a social behavior or interaction, a social role, a social space, a social institution or system, a social value or ideology, or any other socially relevant concept” (Diveica et al., 2023, p. 463). Analyses demonstrated high reliability and validity of the ratings. Importantly, these socialness ratings were only weakly correlated with affective features of words – namely arousal and valence extremity (rs < .25).”  

In addition, we added in affective variables into our extended analyses. Specifically, we examined socialness ratings (from Diveica et al., 2023) while also including arousal and valence into our models. This expanded model (Results, Section 3.5,) shows that socialness remains a significant predictor even after controlling for these affective properties.  
 

Comment 3: Additionally, the predictor is derived from adult judgments. Adults’ conceptualizations of socialness may not mirror children’s early semantic representations, which are often grounded in perceptual and action-based schemas rather than abstract social roles. This raises questions about construct validity and the developmental relevance of the measure. 

• Response 3: We agree that relying on adult norms may not fully capture how children conceptualize words. In our revision, we have added this in the limitations and future directions section.  

“This study has several limitations. First, socialness measure we used is based on adult norms (Diveica et al., 2023). Adults conceptualize socialness abstractly (e.g., “trust,” “leader”), whereas children’s semantic representations are often grounded in perceptual and motor experiences (Pruden et al., 2006; Gentner & Boroditsky, 2001). This mismatch may overestimate children’s access to social meanings early in development. Though previous findings have indicated that young children’s early word learning is more heavily influenced by input frequency and perceptible word features like concreteness (Braginsky et al., 2019), it is notable that the current findings revealed that socialness ratings remained a significant variable and explained unique variance in vocabulary acquisition data even after controlling for these other perceptible word features. Furthermore, Diveica et al.’s broad quantification of socialness glosses over subcomponents of social features. Through our extended analyses that included valence and arousal, we found that our socialness measure remained a unique predictor of word acquisition. These results provide a strong argument for considering socialness as a word feature that impacts learning. As such, future work should explore different subcomponents of socialness (e.g., agency, cooperation vs. competition) when examining word learning and word processing in autistic and non-autistic individuals to better understand this effect. Future studies could also explore latent semantic analysis procedures from language samples from autistic and non-autistic children to explore social semantic features.”  

 

Comment 4: Methodologically, the study omits several key lexical control variables that strongly predict age or vocabulary size of acquisition, concreteness, imageability, grammatical class, word length, phonotactic probability, and contextual diversity. Because socialness is likely correlated with many of these, especially concreteness and word class, the unique contribution of socialness cannot yet be isolated. For instance, verbs are typically both more social and acquired later than nouns, suggesting that the observed effect may partly reflect syntactic class rather than social content per se. Controlling for these lexical dimensions, or at least examining partial correlations, would strengthen confidence in the interpretation. The lexical scope of the dataset also warrants caution. 

• Response 4: To determine whether the unique contribution of socialness could be isolated after accounting for additional lexical predictors that influence word acquisition, we conducted several new analyses and expanded the manuscript accordingly. 
  Specifically, we added concreteness and iconicity to the extended regression models, as well as conducted noun and verb analyses to address potential confounding by grammatical class or syntactic class. These additions strengthen our ability to determine whether socialness predicts VSoA above and beyond other lexical dimensions. 

• To support these additions, we expanded the Introduction (p. 3) to review prior research showing that concreteness and iconicity strongly shape early lexical development (e.g., Swingley & Humphrey, 2018; Braginsky et al., 2019; Verhagen & Van Stiphout, 2022; Perry et al., 2015). We also revised the Methods (p. 6), and Data Analysis Plan (p. 8) to describe how these variables were incorporated. Updated analyses appear in the Results (Sections 3.5–3.6), and we expanded the Discussion (Section 4.1, p. 14) to address how socialness effects relate to broader lexical properties. 

 

 

Comment 5: The analysis is restricted to 170 CDI words, a relatively small and early-developing subset. This narrow range may compress both socialness values and acquisition variance, reducing sensitivity to group differences. In its present form, the study offers insight into early-emerging vocabulary but not necessarily into later or more abstract lexical development. 

• Response 5: Thank you for pointing this out. We agree that the CDI includes only a small set of early-acquired words, which limits the range of socialness and acquisition values and may make group differences harder to detect. We added this point to the Discussion (Section 4.4: Limitations and Future Directions, p. 16).  
  “A second limitation of current study was that socialness ratings were available for only 170 of the 680 words on the CDI, which narrowed the range of words we could examine. This limited dataset restricted the sensitivity to identify potential group differences. Also, our limited sample restricts our ability to explore later lexical development. Though Diveica et al. (2023) developed socialness rating scores for over 8,000 English words, currently autism-specific acquisition norms (e.g., age of acquisition or VSoA) for all of the words that have socialness ratings do not exist. The current study served as an initial step to exploring the role of socialness in early vocabulary development in autistic and non-autistic children.” 
   

Comment 6: Finally, while the authors briefly acknowledge conflicting findings from Jiménez et al. (2021), the reconciliation is superficial. The authors could make more of their methodological advantage: a word-level VSoA analysis arguably provides a finer-grained index of acquisition order than average socialness ratings across bins. This point deserves stronger elaboration to clarify how the current work advances beyond prior literature 

• Response 6: On page 14, we have added details related to the Jiménez et al. (2021) findings that provide helpful insight into their significant group difference for small vocabulary sizes.  

“Jiménez et al. found that the early verb lexicons (1-25 verbs) that typically talking toddlers produced had higher average social ratings relative to larger verb vocabulary sizes (25-50 verbs within the lexicon); however, this effect was small (Cohen’s d = 0.11) and this difference in socialness scores between vocabulary sizes was not significant in autistic children and late talkers. Additionally, Jiménez et al. (2021) found that autistic children’s early verb vocabularies (1-25 verbs) consisted of verbs that were on average lower in socialness ratings relative to those of typically talking toddlers (small effect) and late talking toddlers (medium effect). Given that this small effect of socialness in verb vocabularies between the typical talkers and autistic children was observed only for children with verb vocabulary sizes between 1 and 25 verbs, this effect may not have held if verb vocabulary size was examined on a continuous scale.  

•  We elaborate on how the current study advances beyond the prior literature by stating: 

“The current study’s examination of socialness extends the previous child literature (Haebig et al., 2021; Horvath et al., 2018; Jiménez et al., 2021) by studying socialness in a larger number of words and in words that span different syntactic classes. Additionally, the autism-specific word acquisition data were derived from a much larger sample of autistic children than were examined in Haebig et al. and Jiménez et al., which lessen the concern of encountering potential spurious findings. Also, the current study used a word-level approach that allowed for more fine-grained assessment of the association between socialness features and lexical acquisition, while also controlling for important word features that have been found to impact early word learning.” 
 

Comment 7: The interpretation of the null group difference in discussion could also be deepened. Rather than simply stating that autistic and non-autistic children show similar learning profiles, the authors could discuss potential mechanisms for this convergence, perhaps shared reliance on statistical or distributional learning processes, or constraints inherent to early lexicon formation that are robust across neurotypes. The current discussion does not explore these alternatives. 

• Response 7: Thank you for your thoughtful suggestion. We have extended this discussion on page 16  
  “This similarity is consistent with findings from Lin et al. (2022), who reported no group differences in how imageability is related to vocabulary. The current study aligns with other studies that document similarities in word learning patterns and use of mechanisms that support word learning in autistic and non-autistic children as well. For instance, we know that autistic children benefit from language facilitating strategies like high-quality follow-in commenting and prompts for communication acts (e.g., Clark-Whitney et al., 2022; Haebig et al., 2013; Siller & Signman, 2008). Additionally, autistic as well as non-autistic learning trajectories align with vocabulary growth models that prioritize the importance of learning from semantic and syntactic structure from the learning environment (e.g., Haebig et al., 2025). Autistic children also attend to syntactic distributions adjacent dependencies to support word learning (e.g., Haebig et al., 2017; Horvath, McDermott et al., 2018).”  

Comment 8: A few smaller points could improve readability and transparency. The category labeled “other word types” should be clearly defined and enumerated. Certain sentences contain minor typographical issues (e.g., Line 244). The model specification would benefit from fuller reporting, how the Group variable was coded, whether predictors were centered or standardized, and how missing data were handled. 

• Response 8: Thank you for these helpful suggestions. We revised the manuscript to improve clarity and transparency. We clarified the category labeled “other word types” by explicitly defining all-lexical subcategories in Methods (Section 2.1). and corrected all identified typographical issues. In addition, we added our model specification to report whether predictors were standardized, and how missing data are addressed in the result section (section 3.4- 3.7). The current analyses do not include mean-centered variables (e.g., mean-centered concreteness or socialness).  
   

We are grateful for the reviewers’ insightful comments, which significantly strengthened the manuscript. We believe these revisions improve conceptual framing, methodological rigor, and interpretation of findings. 

We thank you again for the opportunity to revise and resubmit. 

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

After re-reviewing, I would recommend accepting the manuscript. I thank the authors for responding most of my concerns.