Abstract Priming and the Lexical Boost Effect across Development in a Structurally Biased Language

Round 1

Reviewer 1 Report

This manuscript presents a syntactic priming study with German-speaking 3-8 year olds and adults. The authors examined children’s production of dative structures and specifically the effect of abstract priming and lexically-based priming on children’s production of the rarer PO structure across development. They also explored the increase in production relative to baselines and across the experiment. Whilst abstract priming was present at all ages, the lexical boost to priming was only significant in adults, and approached significance in the oldest children. By contrast, the authors report stronger abstract priming, surprisal and cumulative priming in the younger children. The results are discussed in relation to different models of syntactic priming.

This study replicates the design of previous priming studies with English children but extends this work to a new language and wider age range of children. The study design and analyses are clearly detailed and appropriate. There were, however, elements of writing and framing that could be revised, as detailed below (many of the points are rather minor details), to improve the manuscript.

General comments

Overall, I felt the link to the topic of the special issue could be more clearly made - the study addresses issues of syntactic adaptation and has interesting findings that are in line with adaptation accounts of syntactic priming however the discussion of implicit learning accounts glosses rather quickly over this aspect. For readers not familiar with syntactic priming, it would be helpful to cover surprisal accounts in more detail and why, for example, priming effects are predicted to be greater in younger participants. For example, page 4 states that more infrequent sentences lead to greater priming and below that cumulative priming is predicted to be stronger in children but on page 5 predictions about immediate priming being stronger in younger children are made as well - this step feels missed in the set up.

One aspect of novelty that the authors highlight is that the study tests priming in German to explore the extent to which previous findings generalise. Priming has been tested in children from a number of different language backgrounds, as the authors state (though further evidence has been reported in Dutch (van Beijsterveldt & van Hell, 2009); Russian (Vasilyeva et al., 2006); French (Havron et al., 2020); and Japanese (Arai & Mazuka, 2013)) so it would strengthen the case here to expand on why German or German datives are a good test of generalisation specifically or how these results in German build on other cross-linguistic data.

The authors examine their results by distinct age categories but in reality, their age groups run continuously from 3 to 8 years. As such, does it make sense to artificially break this age span (given each age group covers two years, children at the boundaries of adjacent age groups are likely to be more similar to each other than the children at the other end of their age group) and have they looked at all their analysis with age as a continuous predictor? The age groups are of quite different sizes, 37 - 60 participants, is this due to exclusions or different recruitment criteria? It would be helpful to add some information here.

One nice contribution of this paper is the inclusion of the baseline condition - this has been missing from most, if not all, child priming studies published so far but confirms what I suspect often happens in these studies which is that with exposure to primes, the dispreferred structure occurs more often after the alternative prime than you would expect, or find if you included a baseline. As the authors propose, this is good evidence of a cumulative/lasting effect of priming, but it can also have a detrimental effect on the measure of immediate priming effects as it reduces the difference between conditions; see Messenger et al., 2022 (TiLAR 31, chapter 4 - email KM for a copy) for discussion of this issue and recommendations for studies to include baseline conditions! I think the authors’ interpretation of this pattern in their data is entirely appropriate; it’s nice that they clearly distinguish the exploratory analyses that were conducted to explore this hypothesis.

One explanation that I did not follow was the suggestion that children’s representations might not be lexically-grounded. This is provided as an explanation for the lack of a lexical boost early on in development. I’m not sure I can see how children would have abstract representations of structure dissociated from verb lemmas but still be able to use those verbs in those structures. Some explanation and clarification as to what this means (and how this would be instantiated within the implicit learning model of priming) is really needed to show that this is a viable interpretation.

Specific comments

Pages 2-3 - this discussion is a bit clunky here - the first paragraph on page 2 wanders from starting to talk about child findings to listing key adult findings. The subsequent paragraphs provide some further detail but it feels rather piecemeal. This discussion could be better organised and integrated with the rest of the introduction, for example in relation to the accounts described after.

Page 3 - the authors state that the residual activation account cannot explain the absence of a lexical boost in young children - it would be helpful to explain why, especially as this is not a model of language development.

Page 3 - correct ‘structural activation is taking place’ - ‘takes place’

Page 3 - suggest you add ‘some’ or similar to ‘the absence of a lexical boost effect in children which has been previously shown in child studies’ since not all child studies report an absence of lexical boost.

Page 3 - last line and over the page - the subject switches from ‘the model’ to ‘we’ - make this consistent.

Page 4 - Lehmden 2013 is not in the references list.

Page 4 - the authors state that the dative alternation has not been tested with children in a language other than English but Wolleb et al examined datives in Norwegian, albeit with older (adolescent) participants.

Page 4 - last sentence - the authors report that a corpus analysis revealed that children rarely hear or produce PO structures - can this be quantified, e.g. as a % of dative structures, at all?

Page 6 - what is a university bread box?! Just curious!

Page 6 - Design section, please also state which variables were manipulated within-items.

Page 6, last sentence, italicise ‘an’ in line with italicised verbs and zu.

Page 7  - could you clarify in the text how many items there were per combined conditions (prime x verb condition) - I think there were 4 items per combination of structure and verb type but it would be useful to be clear.

Page 9 - baseline results - are the % DOs the proportion of DO+PO or DO+PO+other responses?  Essentially, can we assume that 90% DOs meant only 10% POs? It would be useful to know as in the rest of the results, it is the PO responses that are examined so knowing definitely what the baseline PO response rates were would be helpful.

Page 10  - please define the sum contrast coding applied to age group - the text implies that each age group is compared to the rest of the group - it would be useful to see how this was done.

Page 11 - there is a significant interaction between prime structure and the 5-6 year old age group contrast which wasn’t referred to in the text - please also describe what this indicates.

Page 12 - sorry, I did not understand what the data in Table 3 are - where do these figures come from and how should they be interpreted as an effect size?

Page 13 - could Figure 2 be plotted as a box plot or violin plot to show the distribution of scores within each age group?

Page 13 - the authors state that the the 3-way interaction of prime, verb and age confirms that the lexical boost effect gradually increased with age - being nit-picky, the interaction shows that there was a difference in the lexical boost effect with age, it’s the data which tell you the direction (e.g. figure 2) - might be worth wording this more carefully. Relatedly, it could be helpful to present some mean priming effects to illustrate some of the significant effects reported.

Page 14 - are the PO/DO prime condition data plotted in Figure 3 the same as Figure 1? The difference between PO and DO bars looks smaller for 5-6 year olds and adults and the error bars on most bars look more overlapping in Figure 1 - is there a reason for these differences?

Page 16 - middle paragraph - correct ‘table 2’ to ‘table 3’ in “the effect sizes shown in Table 2”

Page 17 - Branigan, McLean and Jones, 2005 is missing from the references.

Appendix A - none of the items have the preposition ‘zu’ but the text states that half the verbs used ‘zu’ in the PO primes.

In general the quality of English language is good but see comments above for some specific points and proof-read for erroneous 'the's occurring in the odd place.

Author Response

Responses to Reviewer 1:

“This manuscript presents a syntactic priming study with German-speaking 3-8 year olds and adults. The authors examined children’s production of dative structures and specifically the effect of abstract priming and lexically-based priming on children’s production of the rarer PO structure across development. They also explored the increase in production relative to baselines and across the experiment. Whilst abstract priming was present at all ages, the lexical boost to priming was only significant in adults, and approached significance in the oldest children. By contrast, the authors report stronger abstract priming, surprisal and cumulative priming in the younger children. The results are discussed in relation to different models of syntactic priming.

This study replicates the design of previous priming studies with English children but cinextends this work to a new language and wider age range of children. The study design and analyses are clearly detailed and appropriate. There were, however, elements of writing and framing that could be revised, as detailed below (many of the points are rather minor details), to improve the manuscript.

General comments

Overall, I felt the link to the topic of the special issue could be more clearly made - the study addresses issues of syntactic adaptation and has interesting findings that are in line with adaptation accounts of syntactic priming. However, the discussion of implicit learning accounts glosses rather quickly over this aspect. For readers not familiar with syntactic priming, it would be helpful to cover surprisal accounts in more detail and why, for example, priming effects are predicted to be greater in younger participants. For example, page 4 states that more infrequent sentences lead to greater priming and below that cumulative priming is predicted to be stronger in children but on page 5 predictions about immediate priming being stronger in younger children are made as well - this step feels missed in the set up.”

RESPONSE: The revised version of the section on theoretical frameworks of structural priming on pp. 3-4 contains a considerably more detailed description of the Implicit Learning Account, which assumes that priming effects are based on long-term adaptation. We have also included several references to adaptation throughout the revised manuscript. As for surprisal effects,we have revised both the basic description of what surprisal effects are (p. 3, ll. 94 ff.) and our coverage of surprisal effects in the context of the Implicit Learning Account (p. 4, ll. 137 ff.) to make this a bit clearer.

“One aspect of novelty that the authors highlight is that the study tests priming in German to explore the extent to which previous findings generalise. Priming has been tested in children from a number of different language backgrounds, as the authors state (though further evidence has been reported in Dutch (van Beijsterveldt & van Hell, 2009); Russian (Vasilyeva et al., 2006); French (Havron et al., 2020); and Japanese (Arai & Mazuka, 2013)) so it would strengthen the case here to expand on why German or German datives are a good test of generalisation specifically or how these results in German build on other cross-linguistic data.”

RESPONSE: A particularly interesting property of the German dative alternation is that, while German PO and DO sentences are structurally similar to the dative alternation in other European languages, German is heavily biased towards the DO. This property of German allows us to investigate to what extent priming effects for the dispreferred PO structure are boosted by surprisal. Also, the fact that the PO structure is only rarely encountered in everyday language, but occurs in 50% of all experimental prime sentences during the experimental test session makes it possible to study both immediate and cumulative priming effects within the same study. Finally, due to the fact that the German dative alternation is biased towards the DO, but otherwise structurally similar to the English dative alternation, a comparison between our results for German and previous studies for English is potentially informative with respect to the influence of structural biases on priming. We have made minor changes throughout the paper to make the novelty value of our study a bit clearer.

With respect to the additional priming studies mentioned above, thank you for pointing us towards these studies. We have included those in the respective paragraph.

“The authors examine their results by distinct age categories but in reality, their age groups run continuously from 3 to 8 years. As such, does it make sense to artificially break this age span (given each age group covers two years, children at the boundaries of adjacent age groups are likely to be more similar to each other than the children at the other end of their age group) and have they looked at all their analysis with age as a continuous predictor? The age groups are of quite different sizes, 37 - 60 participants, is this due to exclusions or different recruitment criteria? It would be helpful to add some information here.”

RESPONSE: In addition to the age group analysis, a complementary analysis of the data from the three child groups which includes ‘age in months’ as a continuous predictor is reported on page 18 in the manuscript. This analysis shows both a main effect of ‘prime type’ and a three-way interaction between ‘prime type’, ‘verb condition’ and ‘age in months’. In this respect, the key results from this complementary analysis are quite similar to the results from the group-based analysis. We decided to include both analyses in the manuscript, but to report the group-based analysis as the main analysis, for three reasons. First, age differences within the adult group do not contain the same developmental information as in the child groups. While an age difference of, for instance, three months between two children is certainly important with regard to maturation, the same three-month age difference between two adults is essentially meaningless. Thus, including the data from the adult group in an analysis with ‘age in months’ as a continuous predictor is potentially problematic. However, we consider it good practice to report a main analysis which includes the results from all groups and conditions. Second, the group-based analysis gives us a straightforward option to visualise the results and the respective developmental trajectories (see Figures 1 and 2 in the manuscript). This also allows for a direct comparison between our study and previous studies which have tested children in only one particular age group, for instance only 5- to 6-year-olds. Finally, our study design is based on two previous studies on English dative alternation (Rowland et al. (2012) and Peter et al. (2015)), in which the authors also decided to divide their sample into subgroups of 3- to 4-year-olds and 5- to 6-year-olds. Dividing our sample in the same way gives us the opportunity to directly compare the developmental trajectories in these studies.

“One nice contribution of this paper is the inclusion of the baseline condition - this has been missing from most, if not all, child priming studies published so far but confirms what I suspect often happens in these studies which is that with exposure to primes, the dispreferred structure occurs more often after the alternative prime than you would expect, or find if you included a baseline. As the authors propose, this is good evidence of a cumulative/lasting effect of priming, but it can also have a detrimental effect on the measure of immediate priming effects as it reduces the difference between conditions; see Messenger et al., 2022 (TiLAR 31, chapter 4 - email KM for a copy) for discussion of this issue and recommendations for studies to include baseline conditions! I think the authors’ interpretation of this pattern in their data is entirely appropriate; it’s nice that they clearly distinguish the exploratory analyses that were conducted to explore this hypothesis. One explanation that I did not follow was the suggestion that children’s representations might not be lexically-grounded. This is provided as an explanation for the lack of a lexical boost early on in development. I’m not sure I can see how children would have abstract representations of structure dissociated from verb lemmas but still be able to use those verbs in those structures. Some explanation and clarification as to what this means (and how this would be instantiated within the implicit learning model of priming) is really needed to show that this is a viable interpretation.”

RESPONSE: We agree that our original explanation for the lack of a lexical boost in the younger age groups was too speculative. In the revised version of the Discussion section on pp. 17-18, we instead discuss the observed developmental trajectory with reference to the predictions made by the Implicit Learning Account, which explains the lexical boost as a working-memory effect. Following this rationale, the lack of a lexical boost in the younger groups may be due to limited working memory capacity in young children. The developmental trajectory for the lexical boost in our study is consistent with this account. We have also added a more fine-grained comparison of how our results for the lexical boost relate to the findings from previous studies, and discuss to what extent a working-memory-based explanation can account for the differential pattern of results which emerged in these studies. 

“Pages 2-3 - this discussion is a bit clunky here - the first paragraph on page 2 wanders from starting to talk about child findings to listing key adult findings. The subsequent paragraphs provide some further detail but it feels rather piecemeal. This discussion could be better organised and integrated with the rest of the introduction, for example in relation to the accounts described after.”

RESPONSE: We have made a number of changes to the paragraphs on pp. 2-3 to make them easier to read. In the revised version, we introduce key effects in structural priming research (immediate priming, cumulative priming, lexical boost, prime surprisal), and describe the relevance of each effect in research on language processing and development, as well as findings from key previous studies on adults and children.

“Page 3 - the authors state that the residual activation account cannot explain the absence of a lexical boost in young children - it would be helpful to explain why, especially as this is not a model of language development.”

RESPONSE: Based on the suggestions from Reviewer 2, we have made a number of substantial changes to the respective paragraph. We believe the revised version contains a considerably clearer description of how each account explains lexical boost effects.

“Page 3 - correct ‘structural activation is taking place’ - ‘takes place’”

RESPONSE: We have corrected the sentence accordingly.

“Page 3 - suggest you add ‘some’ or similar to ‘the absence of a lexical boost effect in children which has been previously shown in child studies’ since not all child studies report an absence of lexical boost.”

RESPONSE: We have made substantial revisions to the respective section of the manuscript. In the revised version, we explicitly mention that studies investigating the lexical boost in children have come to different conclusions. We also get back to the complex pattern of results in the revised Discussion section on pp.17-18.

“Page 3 - last line and over the page - the subject switches from ‘the model’ to ‘we’ - make this consistent.”

RESPONSE: We have changed ‘we’ to ‘individuals’. (p.4; line 162, 163)

“Page 4 - Lehmden 2013 is not in the references list.”

RESPONSE: We have added the study to the references.

“Page 4 - the authors state that the dative alternation has not been tested with children in a language other than English but Wolleb et al examined datives in Norwegian, albeit with older (adolescent) participants.”

RESPONSE: We have included a reference to Wolleb et al. (2018)’s study on Norwegian to the ‘present study’ section on page 4.

“Page 4 - last sentence - the authors report that a corpus analysis revealed that children rarely hear or produce PO structures - can this be quantified, e.g. as a % of dative structures, at all?”

RESPONSE: The respective corpus study showed 6% PO structures and 94% DO structures in spontaneous speech data from German children.

“Page 6 - what is a university bread box?! Just curious!”

RESPONSE: After the experiment, each child received a ‘Thank you’ snack box with the logo of the study on it. They really liked it.

“Page 6 - Design section, please also state which variables were manipulated within-items.”

RESPONSE: Both ‘prime type’ and ‘verb condition’ were manipulated within-items. We have added this information to the paragraph.

“Page 6, last sentence, italicise ‘an’ in line with italicised verbs and zu.”

RESPONSE:. We have fixed this accordingly.

“Page 7  - could you clarify in the text how many items there were per combined conditions (prime x verb condition) - I think there were 4 items per combination of structure and verb type but it would be useful to be clear.”

RESPONSE: During the experimental session, each participant encountered four items from each of the four combined conditions (PO-DV, DO-DV, PO-SV, DO-SV). We have made changes to the paragraph to make this clearer (p.8, line 327).

“Page 9 - baseline results - are the % DOs the proportion of DO+PO or DO+PO+other responses?  Essentially, can we assume that 90% DOs meant only 10% POs? It would be useful to know as in the rest of the results, it is the PO responses that are examined so knowing definitely what the baseline PO response rates were would be helpful.”

RESPONSE: All analyses refer to the proportion of PO responses out of the sum of all PO and DO responses combined. ‘Other’ responses (i.e. grammatically incorrect responses, incomplete responses, or grammatically correct responses which are neither a PO nor a DO and therefore not informative with respect to priming effects) were excluded from the analyses. We have made minor changes to the description of the coding procedure on page 9 to clarify this.

“Page 10  - please define the sum contrast coding applied to age group - the text implies that each age group is compared to the rest of the group - it would be useful to see how this was done.”

RESPONSE: The sum contrast assigned to the ‘age group’ variable compares each age group with the grand mean of all four groups. For instance, the main effect of ‘3- to 4-year olds’ reported in the table contains information about whether the overall proportion of POs produced by 3- to 4-year olds differs from the average proportion of POs produced by the whole sample. We have added a footnote to the table to make this a bit clearer.

“Page 11 - there is a significant interaction between prime structure and the 5-6 year old age group contrast which wasn’t referred to in the text - please also describe what this indicates.”

RESPONSE: We have added a short description of the two-way interaction in the summary of the model results on page 11, which now states that the model showed “a significant two-way interaction between ‘prime type’ and the age group of 5- to 6-year-olds, with weaker priming for this age group than for the sample as a whole”.

“Page 12 - sorry, I did not understand what the data in Table 3 are - where do these figures come from and how should they be interpreted as an effect size?”

RESPONSE: Table 3 provides information about the size of the priming effect in the DV and SV conditions, separately for each age group. The values reported in the table are beta-coefficients extracted from model analyses for each condition and age group, which can be interpreted as effect size measures. For instance, the table shows that the priming effect for the adult group, while significant in both the DV and SV conditions, is substantially stronger in the SV condition (ß=3.26) than in the DV condition (ß=0.93). We have made several changes to the description of the Table 3 on p. 13 to make this a bit clearer.

“Page 13 - could Figure 2 be plotted as a box plot or violin plot to show the distribution of scores within each age group?”

RESPONSE: We have added a violin plot, which visualises the developmental trajectory for the lexical boost effect, on p.13.

“Page 13 - the authors state that the the 3-way interaction of prime, verb and age confirms that the lexical boost effect gradually increased with age - being nit-picky, the interaction shows that there was a difference in the lexical boost effect with age, it’s the data which tell you the direction (e.g. figure 2) - might be worth wording this more carefully. Relatedly, it could be helpful to present some mean priming effects to illustrate some of the significant effects reported.”

RESPONSE: It is correct that, when considered in isolation, the significant three-way interaction reported on page 13 only shows that interaction between ‘prime type’ and ‘verb condition’ (i.e. the lexical boost effect) changes with age. However, both the additional follow-up analyses for each age group reported in Table 2 and the descriptive statistics for the lexical boost effect by age group shown in Figure 2 show that the interaction is caused by the fact that the lexical boost only gradually develops with increasing age. We have nonetheless made minor changes to the description of the results on page 14, which now states that “This confirms that the age groups differ with regard to the lexical boost.”

“Page 14 - are the PO/DO prime condition data plotted in Figure 3 the same as Figure 1? The difference between PO and DO bars looks smaller for 5-6 year olds and adults and the error bars on most bars look more overlapping in Figure 1 - is there a reason for these differences?”

RESPONSE: The PO and DO bars in Figure 1 are based on the results from all items with PO and DO primes in the experiment. In this respect, Figure 1 reports the descriptive statistics for the main statistical analysis shown in Table 2 on p. 12. The key purpose of Figure 3, in contrast, is to compare the results from the baseline pretest with those from the PO and DO prime conditions from the experimental session. However, the baseline pretest contained only half as many IN items as there were PO and DO items in the experiment. Each of the IN items in the baseline pretest was matched to a particular item from the experiment, so that the items contained the same target stimulus. To ensure a fair comparison between IN, PO, and DO items, the PO and DO bars in Table 3 do not represent the results from all PO or DO items, but only from those PO and DO items which were matched to a particular item in the baseline pretest. We have made changes to the respective paragraph on p.14 to make this clearer.

“Page 16 - middle paragraph - correct ‘table 2’ to ‘table 3’ in “the effect sizes shown in Table 2””

RESPONSE: Done. 

“Page 17 - Branigan, McLean and Jones, 2005 is missing from the references.”

RESPONSE: 

We have added the study to the references. 

“Appendix A - none of the items have the preposition ‘zu’ but the text states that half the verbs used ‘zu’ in the PO primes.”

RESPONSE: We have corrected the description accordingly.

Reviewer 2 Report

Title: Abstract Priming and the Lexical Boost Effect Across Development in a Structurally Biased Language

Summary

The paper aims to investigate the development of abstract structural representation and the lexical boost effect in children from different age groups. A structural priming experiment with ditransitive structures in German is reported. The results show immediate and cumulative abstract structural priming effects in all age groups, which become weaker with increasing age. By contrast, the lexical boost effect becomes stronger upon increasing age. The authors explain the structural priming effects from an implicit learning account and suggest that the abstract structural representations become linked to lexical representations during development.

General comments

The reported study is carried out well and offers interesting and relevant insights in the development of structural representations in children. The research question is topical and the experiment is properly designed. Overall, the paper is well written. However, I think the hypotheses should be motivated better based on the existing literature. There are some omissions in the introduction, and I have some issues with the defined hypotheses and the conclusions drawn from the study. The main problem is the assumption that children start with abstract representations which become linked to lexical items over time, for which I am missing a convincing argumentation.

Specific points

Introduction:

-        Page 2, line 66-67: I feel that perhaps some more discussion is needed about the cumulative priming effect in relation to implicit learning, as the surprisal effect (reflected in the inverse frequency effect) will decrease upon repeated exposure.

-        In the discussion of previous literature, the authors do not distinguish between production and comprehension studies. For example, they mention Thothathiri and Snedeker (2008b) as evidence against a lexical boost effect in children, but the lexical boost effect in comprehension is not so robustly established in adult studies as well.

-        Related to the previous point, given that the authors also mention three studies that do find a lexical boost effect in children, section 1.1 is not particularly convincing and needs some rewriting. The claim that the implicit learning model is necessary to account for the absence of a lexical boost effect in children (page 3, line 132-133) seems to be too strong.

-        The introduction uses the ACT-R model by Reitter et al. to account for structural priming effects and the lexical boost effect, which I think is valid. However, in the remainder of the paper (especially in the discussion), the authors only refer to the implicit learning mechanism. The hypotheses following from a hybrid account such as the ACT-R model are not necessarily the same as those from Chang et al.’s implicit learning model regarding the lexical boost effect. According to the latter, the lexical boost effect is purely attributed to working memory, predicting an increasing lexical boost effect. Hybrid models also assume a role of residual activation. Although this is implicitly there, as the authors do suggest that the development of abstract structural representations becoming linked to lexical representations can account for the increasing lexical boost effect, I believe the different mechanisms should be disentangled more and be discussed more explicitly as being separate mechanisms.

-        The reason why I think the previous point is relevant, is because I am not convinced by the assumption that children start with abstract representations which become linked to lexical items over time. The assumption seems to be illogical to me,  as it contradicts item-based models of syntactic development (e.g., Tomasello, 2000). Crucially, I am missing a discussion on previous accounts of syntactic development in the paper, as well as a justification of this key assumption.

-        Coming back to the previous point, if one would assume the opposite developmental trajectory (in line with item-based accounts, and also with L2 development, see below), this would lead to contradictory mechanisms in a hybrid account: an increasing lexical boost effect due to the maturation of working memory/decreasing abstract structural priming due to weaker surprisal effects on the one hand, and a decreasing lexical boost effect/increasing abstract structural priming due to the decrease of reliance on item-based representations. I think this possible alternative explanation needs to be discussed in the paper, also in the light of the previous child studies reporting opposite patterns, as well as an elaboration on why the authors think a development from abstract representations to lexical representations is more plausible than the other way around.

-        In lines 144-150, the authors seem to confuse long-lasting priming (persistent priming after a number of intervening trials) and cumulative priming (stronger priming after repeated primes).

-        In the discussion of the potential role of working memory in the lexical boost effect (page 3, line 91; page 4, lines 151-161), the authors imply that it has only been suggested that the lexical boost effect involves a working memory component, but there is also direct evidence for this, e.g., Zhang, Bernolet, & Hartsuiker, 2020.

-        Lines 207-212: I think it should be defined clearer what the authors mean with “(stable) syntactic representations”. Does it refer to the relative weights of syntactic structures? Or to the link between lexical items and syntactic structures?

Materials/methods and Results:

-        Page 6, line 268: “an” should be in italics.

-        The results of the three different statistical analyses are not consistent. Some discussion on this is needed.

-        Page 13, lines 480-488 could perhaps be supported by a scatter plot.

Discussion:

-        Page 16, line 584: I am not sure whether the conclusion can be drawn that the representations are weaker in the youngest children. The PO structure is perfectly primeable (in fact, is primed strongest), suggesting that there are established syntactic representations. I think this issue will partially be solved after dealing with the previously raised points (i.e., a clearer definition of “stable syntactic representations” and a justification of the assumed developmental trajectory.

-        In the discussion on abstract priming, a link with L2 speakers is made, but this link is missing when discussing the lexical boost effect. Nevertheless, in the L2, the lexical boost effect is actually stronger in less proficient learners than in more proficient learners (e.g., Kim & McDonough, 2008; Bernolet, Hartsuiker & Pickering, 2013). In fact, models on L2 syntactic development suggest that learners start with item-specific representations which become abstract over time. Especially since the authors assume the reverse pattern in the L1, I think it is relevant to draw this comparison and explain why it makes sense to assume a development from abstract representations to lexical representations in children.

-        The larger cumulative priming effect may not necessarily be due to “less stable” representations (page 18, line 690). The proportion of DO responses was highest in the 3- to 4-year-olds, which also means that more adaptation to the relative weights was needed (the surprisal effect was larger/the difference between their relative weight of the DO and the relative frequency in the experiment was larger for them). Also in the case of equally stable representations but different production preferences, one would have expected this result. Therefore I have trouble understanding why the findings cannot be attributed purely to the maturation of working memory.

Author Response

Responses to Review 2: 

Summary

The paper aims to investigate the development of abstract structural representation and the lexical boost effect in children from different age groups. A structural priming experiment with ditransitive structures in German is reported. The results show immediate and cumulative abstract structural priming effects in all age groups, which become weaker with increasing age. By contrast, the lexical boost effect becomes stronger upon increasing age. The authors explain the structural priming effects from an implicit learning account and suggest that the abstract structural representations become linked to lexical representations during development.

General comments

“The reported study is carried out well and offers interesting and relevant insights in the development of structural representations in children. The research question is topical and the experiment is properly designed. Overall, the paper is well written. However, I think the hypotheses should be motivated better based on the existing literature. There are some omissions in the introduction, and I have some issues with the defined hypotheses and the conclusions drawn from the study. The main problem is the assumption that children start with abstract representations which become linked to lexical items over time, for which I am missing a convincing argumentation.”

RESPONSE: We agree that our original explanation for the lexical-boost results was too speculative. We therefore decided to instead discuss the observed developmental trajectory for the lexical boost with reference to the Implicit Learning Account. Given that the Implicit Learning Account attributes the lexical boost to working memory, lexical boost effects can only emerge when working memory capacity is sufficiently developed through maturation. We have added an additional paragraph to the Discussion section, in which we discuss to what extent the account can explain our lexical boost results as well as the findings from previous studies investigating this question.

“Page 2, line 66-67: I feel that perhaps some more discussion is needed about the cumulative priming effect in relation to implicit learning, as the surprisal effect (reflected in the inverse frequency effect) will decrease upon repeated exposure.”

RESPONSE: It is true that cumulative priming effects, due to decreasing surprisal, should gradually decrease when the amount of encounters with the respective structure increases. We have added a brief discussion of this issue to the description of the Implicit Learning account (p. 4, ll.146 ff.).

“In the discussion of previous literature, the authors do not distinguish between production and comprehension studies. For example, they mention Thothathiri and Snedeker (2008b) as evidence against a lexical boost effect in children, but the lexical boost effect in comprehension is not so robustly established in adult studies as well.”

RESPONSE: We agree that the existing studies on the lexical boost in children differ with regard to a range of potentially important properties, including whether they have looked at priming in comprehension or production. We have mentioned this issue in the revised version of the Introduction (p. 2, ll.69 ff.). That said, note that the developmental trajectory observed in Thothathiri and Snedeker’s comprehension priming study also emerged in several production priming studies (Peter et al., 2015; Rowland et al., 2012).

“Related to the previous point, given that the authors also mention three studies that do find a lexical boost effect in children, section 1.1 is not particularly convincing and needs some rewriting. The claim that the implicit learning model is necessary to account for the absence of a lexical boost effect in children (page 3, line 132-133) seems to be too strong.”

RESPONSE: Based on suggestions from Reviewer 1, we have made substantial revisions to Section 1.1. In particular, we have considerably extended our discussion of the predictions which can be derived from the respective theoretical accounts.

“The introduction uses the ACT-R model by Reitter et al. to account for structural priming effects and the lexical boost effect, which I think is valid. However, in the remainder of the paper (especially in the discussion), the authors only refer to the implicit learning mechanism. The hypotheses following from a hybrid account such as the ACT-R model are not necessarily the same as those from Chang et al.’s implicit learning model regarding the lexical boost effect. According to the latter, the lexical boost effect is purely attributed to working memory, predicting an increasing lexical boost effect. Hybrid models also assume a role of residual activation. Although this is implicitly there, as the authors do suggest that the development of abstract structural representations becoming linked to lexical representations can account for the increasing lexical boost effect, I believe the different mechanisms should be disentangled more and be discussed more explicitly as being separate mechanisms.”

RESPONSE: We have revised the paragraph on p.4, which now contains a more detailed description of Reitter’s ACT-R model, and also discusses what the account predicts with respect to the key effects investigated in our study.

“The reason why I think the previous point is relevant, is because I am not convinced by the assumption that children start with abstract representations which become linked to lexical items over time. The assumption seems to be illogical to me,  as it contradicts item-based models of syntactic development (e.g., Tomasello, 2000). Crucially, I am missing a discussion on previous accounts of syntactic development in the paper, as well as a justification of this key assumption. (...) if one would assume the opposite developmental trajectory (in line with item-based accounts, and also with L2 development, see below), this would lead to contradictory mechanisms in a hybrid account: an increasing lexical boost effect due to the maturation of working memory/decreasing abstract structural priming due to weaker surprisal effects on the one hand, and a decreasing lexical boost effect/increasing abstract structural priming due to the decrease of reliance on item-based representations. I think this possible alternative explanation needs to be discussed in the paper, also in the light of the previous child studies reporting opposite patterns, as well as an elaboration on why the authors think a development from abstract representations to lexical representations is more plausible than the other way around.”

RESPONSE: As mentioned above, we agree that the account based on the development of verb-structure links we proposed in the original version of the manuscript was too speculative, particularly given that the working-memory-based explanation derived from the Implicit Learning Account can largely account for our observed lexical-boost results.

“In lines 144-150, the authors seem to confuse long-lasting priming (persistent priming after a number of intervening trials) and cumulative priming (stronger priming after repeated primes).”

RESPONSE: In response to comments from Reviewer 1, we have made substantial revisions to the entire Introduction. The revised version contains a clearer description of each key effect investigated in our study, and also a more detailed discussion of how these effects are accounted for by established theoretical accounts.

“In the discussion of the potential role of working memory in the lexical boost effect (page 3, line 91; page 4, lines 151-161), the authors imply that it has only been suggested that the lexical boost effect involves a working memory component, but there is also direct evidence for this, e.g., Zhang, Bernolet, & Hartsuiker, 2020.”

RESPONSE: We have included a brief description of this study in the revised Introduction (p.4, ll.157 ff.).

“Lines 207-212: I think it should be defined clearer what the authors mean with “(stable) syntactic representations”. Does it refer to the relative weights of syntactic structures? Or to the link between lexical items and syntactic structures?”

RESPONSE: We have included the following sentence in the revised Introduction (p.4, ll.140 ff.) to make this clearer. Also see our response to a related comment below. 

“(...) children who, across their lifetime, have previously encountered a particular structure on only very few occasions (and whose respective structural representation is less stable in the sense that it is based on very few prior observations) should, when being primed by the respective structure, show particularly strong priming effects, due to surprisal.”

“Page 6, line 268: “an” should be in italics.”

RESPONSE: We have fixed this accordingly.

“The results of the three different statistical analyses are not consistent. Some discussion on this is needed.”

RESPONSE: The respective analyses show a largely consistent pattern of results: Specifically, the two key findings from the group-based immediate-priming analyses on pp. 11-16 are that (a) abstract structural priming occurs across all age groups (i.e. a significant main effect of ‘prime type’) and (b) the lexical boost is not present in 3- to 4-year-olds, but only gradually develops with increasing age (i.e. significant three-way interactions between ‘prime type’, ‘verb condition’, and ‘age group’) . Both key effects are also present in the other immediate-priming analyses: For instance, in the separate models for each age group, the main effect of ‘prime type’ is significant within all four groups. With regard to the lexical boost, the interaction between ‘prime type’ and ‘verb condition’ is non-significant in the two younger age groups, marginally significant in 7- to 8-year-olds, and fully significant in adults. Finally, the additional analysis with ‘age in months’ as a continuous predictor reported on p.14 also shows a significant main effect of ‘prime type’ (i.e. significant immediate priming across the whole sample) and  a significant three-way interaction between ‘prime type’, ‘verb condition’, and ‘age in months’ (i.e. the above-mentioned developmental trajectory for the lexical boost).

“Page 13, lines 480-488 could perhaps be supported by a scatter plot.”

RESPONSE: Following a recommendation from Reviewer 1, we have included a violin plot, which visualises the developmental trajectory for the lexical boost effect and also contains information about the degree of interindividual variation within each age group.

“Page 16, line 584: I am not sure whether the conclusion can be drawn that the representations are weaker in the youngest children. The PO structure is perfectly primeable (in fact, is primed strongest), suggesting that there are established syntactic representations. I think this issue will partially be solved after dealing with the previously raised points (i.e., a clearer definition of “stable syntactic representations” and a justification of the assumed developmental trajectory.”

RESPONSE: We fully agree that the observed structural priming effect in 3- to 4-year-olds suggest that these children already possess established abstract syntactic representations. Instead, our notion that these representations are less stable in young children serves as an explanation for why the group of 3- to 4-year-olds actually shows stronger priming than the older groups: Children at this age have previously encountered the respective structures only very few times. Thus, the respective syntactic representation is less stable in the sense that it is based on only relatively few previous encounters with this particular structure. This should lead to enhanced priming effects due to surprisal. Adults, in contrast, have already experienced a large number of previous encounters with the respective structure. As a result, their syntactic representations are already considerably more stable, and one additional exposure to the structure during processing of the prime has a somewhat smaller effect than in adults. In casual terms, it makes a difference whether a PO prime increases an individual’s total number of encounters with the PO structure from 10 to 11 (e.g. in a 3-year-old child) or from 1000000 to 1000001 (e.g. in an adult speaker). We have included a number of minor changes throughout the manuscript to make this a bit clearer. In particular, the revised version of the manuscript contains a much more straightforward explanation for what we mean by ‘less stable’ abstract representations in children (p. 4, ll.140 ff.).

“In the discussion on abstract priming, a link with L2 speakers is made, but this link is missing when discussing the lexical boost effect. Nevertheless, in the L2, the lexical boost effect is actually stronger in less proficient learners than in more proficient learners (e.g., Kim & McDonough, 2008; Bernolet, Hartsuiker & Pickering, 2013). In fact, models on L2 syntactic development suggest that learners start with item-specific representations which become abstract over time. Especially since the authors assume the reverse pattern in the L1, I think it is relevant to draw this comparison and explain why it makes sense to assume a development from abstract representations to lexical representations in children.”

RESPONSE: See our response above for the issue of verb-structure links. With respect to the lexical boost in L2, adult second-language learners differ from children with regard to a wide range of potentially relevant basic properties, such as the fact that adults already possess a fully developed working-memory capacity, as well as the overall amount of exposure to the target language. We have thus decided against including a discussion of these studies in the manuscript.

“The larger cumulative priming effect may not necessarily be due to “less stable” representations (page 18, line 690). The proportion of DO responses was highest in the 3- to 4-year-olds, which also means that more adaptation to the relative weights was needed (the surprisal effect was larger/the difference between their relative weight of the DO and the relative frequency in the experiment was larger for them). Also in the case of equally stable representations but different production preferences, one would have expected this result. Therefore I have trouble understanding why the findings cannot be attributed purely to the maturation of working memory.”

 RESPONSE: As mentioned above, structural representations in children are less stable in the sense that they are based on a very small number of previous encounters with the structure. As a result, the Implicit Learning Account predicts that exposure to a prime sentence with this structure should cause substantial weight adjustment, and thus a stronger priming effect. We cover this issue in more detail in the revised version of the Discussion section (p. 19, ll. 751 ff.). As for the possible role of maturation of working memory, see the revised version of Section 4.2.1 in the manuscript (p. 18, ll.691 ff.).

Round 2

Reviewer 1 Report

The manuscript has been extensively revised and is much improved as a result; in general the suggestions have been responded to thoroughly. In particular the introduction and discussion are much clearer and more appropriate. I have only a couple of minor comments to suggest, a couple following on previous comments, otherwise, this is almost ready for publication.

As mentioned, the introduction is much clearer and better organised. I just wondered if in the final paragraph of section 1.1, where hybrid accounts are discussed, the authors could also consider whether these models make predictions for frequency or surprisal effects and whether they predict developmental differences in lexical boost effects?

I think this comment got missed before: The age groups are of quite different sizes, 37 - 60 participants, is this due to exclusions or different recruitment criteria? It seems unlikely that these would be planned sample sizes so it would be helpful to add some information here about how the target sample was reached.

Previously, I asked if the authors could clarify in the text how many items there were per combined conditions (prime x verb condition). They confirmed there were four items per cell but I could not see this clearly stated in the text, the first paragraph of page 8 says that each verb appeared once in each condition, it would be useful to add somewhere here that each list had 4 items per combination of structure and verb type.

I also previously asked about the baseline results - are the % DOs the proportion of DO+PO or DO+PO+other responses? Essentially, can we assume that 90% DOs meant only 10% POs? It would be useful to know as in the rest of the results, it is the PO responses that are examined so knowing definitely what the baseline PO response rates were would be helpful. The authors have updated the coding procedure but where I felt this information (% PO responses) would be helpful was in the results section (3.1, page 10) - the reader could assume that 90% DOs meant only 10% POs but it would be helpful to add this information for each age group since it is the frequency of PO responses that is reported in other analyses.

I am still a bit confused about the sum contrast coding applied to age group. The added footnote doesn’t really explain how this was done (note that recent recommendations are that authors report their coding, what was coded as 1 or 0, what contrasts (e.g. 0.5/-0.5) were applied, in full for maximum clarity and interpretability (and reproducibility) - Brehm & Alday, 2022, JML). But I’m also a bit confused as to how this relates to the results in Table 1. If the age group variable compares each group to the grand mean of all four groups, where is the row for the adult results? Surely there should be an output for adults? And isn’t this odd to include the test group’s performance within the grand mean that is compared against? Given the adult results are not there, it looks like the age contrast is treating the adult group as the reference level and comparing each child group to the adult performance but I can’t be sure. If this is the case though, the interpretation of any age effects in these models (including subsequent models reported in the text not in tables but with the age group factor - baseline and cumulative analyses) would need to be updated. If this is not the case, where are the results for the adult data?

Minor comments:

Page 2, last paragraph - the authors cite Branigan & Messenger, 2016 and Messenger 2021 in reporting on cumulative effects found in blocked designs. Neither of these studies involved a blocked design - children experienced both structures within the same priming task.

Page 5, Wolleb et al., 2018 - it might be useful to the reader to briefly state whether or not they found significant priming for dative structures.

Page 14, there is a rogue ‘a’ in “the results also showed a significant a three-way interaction”

Page 14, supplementary analysis with age in months - was age centred prior to adding to the model? 

Page 15, first paragraph - the authors first report a linear-mixed effects model and then a logit mixed effect model - please be consistent. 

Page 18, it would be worth referencing Scheepers, Raffray and Myachykov (2017; JML) in contrast to the work by van Gompel and colleagues on lexical boost as they report findings that lexical boost effects emerge with items other than the lexical head.

Author Response

RESPONSES TO REVIEWER 1:

“The manuscript has been extensively revised and is much improved as a result; in general the suggestions have been responded to thoroughly. In particular the introduction and discussion are much clearer and more appropriate. I have only a couple of minor comments to suggest, a couple following on previous comments, otherwise, this is almost ready for publication.

As mentioned, the introduction is much clearer and better organised. I just wondered if in the final paragraph of section 1.1, where hybrid accounts are discussed, the authors could also consider whether these models make predictions for frequency or surprisal effects and whether they predict developmental differences in lexical boost effects?”

RESPONSE: We have added a discussion of the predictions of hybrid accounts for reversed frequency/surprisal effects and the lexical boost to the respective section on page 4.

“I think this comment got missed before: The age groups are of quite different sizes, 37 - 60 participants, is this due to exclusions or different recruitment criteria? It seems unlikely that these would be planned sample sizes so it would be helpful to add some information here about how the target sample was reached.”

RESPONSE: Our original plan was to test approximately 40 participants per age group. We expected 3- to 4-year olds to produce substantially more incomplete or grammatically incorrect target sentences than the other groups, and decided to compensate for the resulting unavoidable loss of data points by testing slightly more participants from this age group. During data collection in nurseries and schools, we decided to test any child who wanted to participate, irrespective of whether the respective age group already contained enough participants. Children would have felt excluded if we had decided not to test them. Once we had the data from these additional participants, it would not have made much sense to exclude them just to get to a similar number of participants per group. As a result, some age groups in the final sample contain more participants than others.

“Previously, I asked if the authors could clarify in the text how many items there were per combined conditions (prime x verb condition). They confirmed there were four items per cell but I could not see this clearly stated in the text, the first paragraph of page 8 says that each verb appeared once in each condition, it would be useful to add somewhere here that each list had 4 items per combination of structure and verb type.”

RESPONSE: We have added information about the number of items per condition which each participant encountered during the experimental session to the description of the stimuli (p.8, line 349-350).

“I also previously asked about the baseline results - are the % DOs the proportion of DO+PO or DO+PO+other responses? Essentially, can we assume that 90% DOs meant only 10% POs? It would be useful to know as in the rest of the results, it is the PO responses that are examined so knowing definitely what the baseline PO response rates were would be helpful. The authors have updated the coding procedure but where I felt this information (% PO responses) would be helpful was in the results section (3.1, page 10) - the reader could assume that 90% DOs meant only 10% POs but it would be helpful to add this information for each age group since it is the frequency of PO responses that is reported in other analyses.”

RESPONSE: As explained in the coding section on pp. 9-10, the values reported in Section 3.1 refer to the proportion of PO responses out of the sum of all PO and DO responses, excluding ‘other’ responses. To make this clear, we have revised the paragraph, which now states that “This bias was strongest in the 3- to 4-year-olds (90% DOs produced in the baseline relative to 10% POs), and gradually decreased in the older child groups (75% DOs vs. 25% POs in the 5- to 6-year-olds, 68% DOs vs. 32% POs in the 7- to 8-year-olds). The adult group also showed a bias towards the DO structure (79% DOs vs. 21% POs).”

“I am still a bit confused about the sum contrast coding applied to age group. The added footnote doesn’t really explain how this was done (note that recent recommendations are that authors report their coding, what was coded as 1 or 0, what contrasts (e.g. 0.5/-0.5) were applied, in full for maximum clarity and interpretability (and reproducibility) - Brehm & Alday, 2022, JML). But I’m also a bit confused as to how this relates to the results in Table 1. If the age group variable compares each group to the grand mean of all four groups, where is the row for the adult results? Surely there should be an output for adults? And isn’t this odd to include the test group’s performance within the grand mean that is compared against? Given the adult results are not there, it looks like the age contrast is treating the adult group as the reference level and comparing each child group to the adult performance but I can’t be sure. If this is the case though, the interpretation of any age effects in these models (including subsequent models reported in the text not in tables but with the age group factor - baseline and cumulative analyses) would need to be updated. If this is not the case, where are the results for the adult data?”

RESPONSE: As described on page 11, the effects of ‘age group’ were based on a sum contrast, with the results from each age group getting compared with the results from all four age groups combined. We have made minor changes to both Table 1 and to the description on page 11 to make this clearer. 

Sum contrasts constitute an established option to determine whether a particular subgroup of participants differs from the sample on the whole (see Shad, Vasishth, Hohenstein and Kliegl, 2020, for a detailed discussion of the issue). In such models, it is not possible to include all pairwise contrasts which are theoretically possible. This is why Table 1 reports effects of ‘age group’ for the three child groups, but not for the adult group. We decided to exclude the effect of the adult group rather than one of the three child groups because the comparison between the adult group and the whole sample is uninformative with regard to our main research question of how priming effects and lexical-boost effects in children change with increasing age. 

As for how exactly the sum contrast was implemented in R, see the annotated R script provided on OSF.

“Minor comments: Page 2, last paragraph - the authors cite Branigan & Messenger, 2016 and Messenger 2021 in reporting on cumulative effects found in blocked designs. Neither of these studies involved a blocked design - children experienced both structures within the same priming task.”

RESPONSE: Thank you very much for pointing this out. We have fixed the issue accordingly (p.2. line 85-86).

“Page 5, Wolleb et al., 2018 - it might be useful to the reader to briefly state whether or not they found significant priming for dative structures.”

RESPONSE: We have added a short description of the results from this study to the respective paragraph (p. 5, line 214-215). 

“Page 14, there is a rogue ‘a’ in “the results also showed a significant a three-way interaction””

RESPONSE: Fixed.

“Page 14, supplementary analysis with age in months - was age centred prior to adding to the model? 

RESPONSE: The supplementary analysis reported on p. 14 includes ‘age in months’ as a centred continuous predictor. We have made minor changes to the paragraph to make this clear (p. 14, line 555-556).

“Page 15, first paragraph - the authors first report a linear-mixed effects model and then a logit mixed effect model - please be consistent.”

RESPONSE: Fixed (p.15, line 588).

“Page 18, it would be worth referencing Scheepers, Raffray and Myachykov (2017; JML) in contrast to the work by van Gompel and colleagues on lexical boost as they report findings that lexical boost effects emerge with items other than the lexical head.”

RESPONSE: We have added a brief discussion of the Scheepers et al. (2017) study to the respective paragraph on p.18.

Reviewer 2 Report

The revised paper has improved a lot and most of my issues have been solved now that the speculative explanation on the developmental trajectory has been left out and some parts have been rewritten. The paper is much more straightforward and easier to follow now. I only have a few remarks left.

Page 2, line 62: perhaps add that the increase of the lexical boost effect with growing age may be attributed to the development of working memory

Section 1.1 is much clearer now. I am just missing a prediction on the development of the lexical boost effect in children for the hybrid account by Reitter et al. (given that it attributes the effect to spreading activation and not to memory cues).

Page 6, line 233: change into “the weights of syntactic representations should initially be less stable”

Results section: why is the priming effect weaker in 5-6- year olds than in the other group? You mention it in the result section, but I do not read an explicit explanation for this finding in the discussion.

Page 18: Just a thought, but perhaps one potential reason why the lexical boost effect emerges relatively late in the current study is that the prepositions used in the PO structures are varied (“zu” and “an”), while most studies have only one (“to” in English)? The preposition may serve as an additional lexical cue.

Page 18, line 703: I suggest to explicitly add that if the lexical boost effect only emerges when the head is repeated, this points to residual activation as a locus of the lexical boost in adults, and that any role of residual activation in priming in children should be further investigated.

As the discussion is now clearer on the inconsistency of a working-memory-based explanation for the lexical boost effect and does not mention this assumption on the transition from abstract to item-specific representation anymore, I am satisfied with your argumentation and have no objections anymore against not mentioning the reverse developmental trajectory regarding the lexical boost effect of L2 speakers if you feel that it does not contribute (although it might strengthen your conclusion that residual activation seems to play a larger role in adults than in children while working memory is the main mechanism drives the pattern observed in children).

Author Response

RESPONSES TO REVIEWER 2:

“The revised paper has improved a lot and most of my issues have been solved now that the speculative explanation on the developmental trajectory has been left out and some parts have been rewritten. The paper is much more straightforward and easier to follow now. I only have a few remarks left.

Page 2, line 62: perhaps add that the increase of the lexical boost effect with growing age may be attributed to the development of working memory”

RESPONSE: We have added a sentence on this to the paragraph (p.2, line 69-70).

“Section 1.1 is much clearer now. I am just missing a prediction on the development of the lexical boost effect in children for the hybrid account by Reitter et al. (given that it attributes the effect to spreading activation and not to memory cues).”

RESPONSE: We have extended our discussion of Reitter’s ACT-R model on page 4. The revised paragraph now contains a more detailed description of how the model accounts for surprisal effects and the lexical boost. That said, please note that Reitter’s model focuses on explaining structural priming in adults, and does not make any claims about how the assumed architecture initially develops in children during language acquisition. This makes it difficult to derive predictions for the developmental trajectory of lexical boost effects from this model. Our paragraph nonetheless contains speculative predictions for the developmental trajectory.

“Page 6, line 233: change into “the weights of syntactic representations should initially be less stable””

RESPONSE: Fixed.

“Results section: why is the priming effect weaker in 5-6- year olds than in the other groups? You mention it in the result section, but I do not read an explicit explanation for this finding in the discussion.”

RESPONSE: It is correct that our results show an unexpected two-way interaction between ‘prime type’ and the age group of 5- to 6-year olds, with slightly weaker (but still significant) priming in 5- to 6-year olds than in the whole sample. This interaction was not predicted by our hypotheses. Note, however, that the other two child groups also show a tendency in the same direction, with numerical trends towards weaker priming in all child groups than in the whole sample. This suggests that the significant two-way interaction for the 5- to 6-year olds is mainly driven by the contrast between the 5- to 6-year olds and the adults. Indeed, the adult group shows substantially stronger priming than all three child groups (as indicated by the β coefficients for the effect of ‘prime type’ shown in Table 2). A possible reason why this trend towards a two-way interaction reached significance only for the 5- to 6-year olds may be relatively more limited statistical power in the other two age groups: The group of 3- to 4-year olds produced a higher number of incomplete or grammatically incorrect target sentences than the older age groups, which led to more data loss, while the group of 7- to 8-year-olds contained slightly less participants. 

Given that the two-way interaction was not predicted, and also given that non-linear and gradual developmental trajectories are generally hard to explain by any theoretical account, we decided against including any strong claims about this interaction. However, we have added a brief explanation for the two-way interaction to the results section (p.12, ll. 522-524).

“Page 18: Just a thought, but perhaps one potential reason why the lexical boost effect emerges relatively late in the current study is that the prepositions used in the PO structures are varied (“zu” and “an”), while most studies have only one (“to” in English)? The preposition may serve as an additional lexical cue.”

RESPONSE: It is certainly possible that this property of the German dative alternation might generally play a role with regard to the comparison of priming effects in German versus English. However, with respect to the specific question of whether this can explain the developmental trajectory for the lexical boost effect in our study (i.e. the fact that the lexical boost only emerged relatively late), this strikes us as unlikely: The preposition in a German PO sentence depends on the particular head verb, in the sense that some verbs require the preposition ‘an’, while other verbs require ‘zu’. As a result, for all experimental items in the ‘same-verb’ condition, it was necessarily the case that these prime-target pairs shared both the same verb and the same preposition. Thus, if the preposition indeed served as an additional lexical cue, the increased amount of lexical overlap between prime and target (i.e. the fact that prime and target share both the same verb and the same preposition) should have actually led to more robust lexical boost effects. Even this issue aside, our experimental design is not ideally suited for investigating this issue empirically: If we split up our data into ‘zu’ and ‘an’ items and run a separate analysis for each subgroup, we would be left with only a very small amount of data points within each group. We thus decided against drawing any conclusions about this issue in the manuscript.

COMMENT: Page 18, line 703: I suggest to explicitly add that if the lexical boost effect only emerges when the head is repeated, this points to residual activation as a locus of the lexical boost in adults, and that any role of residual activation in priming in children should be further investigated.

RESPONSE: The revised version of the paragraph on page 18 discusses the findings for the lexical boost effect in adults in a bit more detail. We have also added that, if it is indeed correct that the lexical boost is largely specific to verb repetition, this argues against an explanation based on memory, and points towards an explanation based on residual activation.

As the discussion is now clearer on the inconsistency of a working-memory-based explanation for the lexical boost effect and does not mention this assumption on the transition from abstract to item-specific representation anymore, I am satisfied with your argumentation and have no objections anymore against not mentioning the reverse developmental trajectory regarding the lexical boost effect of L2 speakers if you feel that it does not contribute (although it might strengthen your conclusion that residual activation seems to play a larger role in adults than in children while working memory is the main mechanism drives the pattern observed in children).

RESPONSE: Thank you for the suggestion. We seriously considered adding a discussion of findings for structural priming and the lexical boost in L2 speakers to the manuscript, but eventually decided against this. Adult L2 speakers differ from L1 children with regard to a multitude of relevant properties, which makes the results from the two fields difficult to compare. For instance, adult L2 speakers obviously already possess fully developed working-memory capacity and substantial metalinguistic knowledge about the respective structures. On the other hand, they may have to deal with higher memory load because the sentences are in L2, and are also potentially influenced by how the respective structures work in their L1. These differences make it hard to interpret any differences in priming effects between L1 children and L2 adults.