Priming of Possessive Constructions in German: A Matter of Preference Effects?

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

I think it is an interesting and relevant question to investigate how the inverse preference effect relates to any long-lasting priming effects. However, I think the authors could still improve on the organization of their introduction and the theoretical framing of their research questions. Both the inverse preference effect and cumulative priming are explained from the error-based learning account. I think the contribution of this paper is that it may question error-based learning as the source for the (short-lived) inverse preference effect.

Furthermore, I would like to note that cumulative priming and long-lasting priming are two different notions, and I don’t think this distinction is sufficiently made by the authors:  cumulative priming means that the priming effect increases over the course of the experiment, while long-lasting priming means that there is still a priming effect some time period after the experiment (but a potentially weaker effect). As is stated in the paper, I believe that error-based accounts predict that the inverse preference effect decreases over time, which obviously is the opposite direction from cumulative priming effects. But I am not so sure about the relationship between the inverse preference effect and long-lasting priming (which is probably weaker than immediate priming), or at least whether such an effect could have become visible in the current study, especially given that the “other” responses are taken into account for the analysis of the post-test. The relative weights reflect the total number of occasions that one has been exposed to a particular structure, which is still much higher for the more frequent structure. Relative to the other responses, there is a visible increase of the frequent primed structure. But the likelihood of the less frequent structure might still be so low, that a small increase due to long-lasting priming/adapted relative weights could not be observed in the limited amount of observations due to the dominance of particular other responses and the more frequent structure).  

I also have some difficulty with calling the cumulative priming of the postnominal genitive a preference effect, since it is not compared to the proportion of other genitive responses, but to “other” responses. I am not sure whether it is really a mechanism that should be explained from a priming account such as error-based learning, or whether it is a task adaptation effect (participants realize that the experimenters don’t want full sentences as a response), which participants may be reluctant to adopt for very rare structures.

Importantly, I have some concerns about the statistical power of the study. The number of observations does not meet the recommendations following from a meta-analysis (Mahowald et al., 2016). This is especially true for Experiment 1 which tests only 16 participants. In addition, it tests for an archaic construction, which seems to be not produced outside the context of the priming experiment. As a result, the number of observations is really low, making it problematic to draw any conclusions.

Major comments:

Introduction, page 1-2: It would probably make more sense to first discuss the error-based implicit learning account before mentioning the existence of other accounts. Right now, already the second paragraph is devoted to something that is not so relevant for the paper.

Page 2, lines 70-79: it is a bit abstract to talk about structure A and B. It would be useful if an example of two alternating structures was provided (e.g., actives and passives, with a strong preference for actives)

Page 2, lines 80-85: I don’t see how this is counterintuitive. Even though a structure is less frequently produced, the priming effect (so the increase in frequency) may be stronger. E.g., a priming effect from 10% to 20% (an increase of 100%, 10 percentage point) for a less frequent vs. a priming effect from 80% to 85%  (an increase of 6.25%, 5 percentage point) for a more frequent structure implies that there is stronger priming for the less frequent variant while the more frequent variant remains the dominant variant.

Page 3, line 136-139: I am not sure whether I understand what the authors mean here. Do they mean that the cumulative priming effects are predicted to increase during priming, in contrast to the inverse preference effect? I might think that the strength of cumulative priming perhaps also decreases over time, following the same reasoning (so cumulative priming may be non-linear).

Page 3, line 142: I have some trouble with the transition between topics here. What do verb-specific preferences have to do with the decrease of the inverse preference effect over time? Also, is there any evidence for this decrease?

Page 3, line 145-147: I don’t think that verb-specific predictions are not necessarily involved in the error-based learning model, since learning involves adapting the weight of the link between the concept and message. Different from lexicalist accounts, the error-based learning model does not necessarily assume generalized representations for structures (such as DO and PO nodes in the residual activation account by Pickering and Branigan). Since the error-based learning model predicts words based on the previous words, verb-specific preferences seem to be inherent to the model.

Page 4, lines 195-196: I have trouble following the reasoning here, since cumulative priming is usually also explained from the error-based learning model (it indicates that priming effects have long-lasting effects, in contrast to residual activation models which only predict short-lasting immediate priming effects).

Experiment 1

Method

Page 5, lines 198-204: is the prenominal genitive construction considered grammatical/productive by users? If not, wouldn’t that have consequences for the priming effects? There probably is a difference between priming of ungrammatical structures and priming of less preferred structures. Do the authors have any data from a grammaticality judgment task?

Page 5, lines 206-209) Experiment 1: I have some concerns about the sample size. According to Mahowald et al. (2016), the power to detect a priming effect without lexical overlap between prime and target with 24 critical items is only 39% The power to detect an interaction is somewhere between 7 and 36% (Mahowald explicitly states that neutral baseline items are excluded when counting the number of critical items, but even if all those 48 items would be included, the power would be very low). Also, why is there such a large difference between the number of participants between experiments? I would suggest testing extra participants for Experiment 1.

Results

Figures with bar plots (e.g., Figure 2 and 3): I would suggest using stacked bar plots, since the percentages of the target conditions add up to 100%. Perhaps it would also be useful to have a plot

I have some issue with the way cumulative priming is measured, since the number of occurrences of the prime structure vary throughout the paper. Right now, you only measure order effects, which could potentially be due to other reasons than cumulative priming. At some point in the experiment participants may for instance have encountered 5 primes of one structure and only 3 of the other structure, and this also differs per participant because of the different lists.  By adding the number of primes encountered you measure the effect of the number of encounters (primes) that actually drive prediction error. See (footnote 6 in) Bernolet et al. (2013). (Bernolet, S., Hartsuiker, R. J., & Pickering, M. J. (2013). From language-specific to shared syntactic representations: The influence of second language proficiency on syntactic sharing in bilinguals. Cognition, 127(3), 287-306)

Figure 3: Related to the previous point, I think the “Priming” title is not exactly what is displayed here, also because it doesn’t show an interaction with the prime condition. Of course the authors interpret the growth in the use of pre- and post-nominal genitives as a cumulative priming effect, but it might be better to use a more neutral title (e.g., the proportion of responses over the course of the priming experiment). I think this is especially relevant since the authors haven’t excluded the Other responses here yet.

Page 11/Page 17, Table 2/4: It is unclear to me whether the authors also used sum-coding here rather than dummy-coding. If they used dummy-coding here (which I think is the case), it should be stated explicitly for each analysis which coding scheme was used. If they used sum-coding, the naming of the variables is a bit misleading. The post-test phase then does not directly compare to the intercept (which is the priming experiment). In sum-coding, the comparison is relative to the sum of the preceding levels.

Discussion

Page 12, lines 488-490: This claim seems to be too strong, since there is no significant effect of order in the statistical model the authors report.

Line 501: I am not sure whether I agree that there is evidence for cumulative priming. There is no evidence for cumulative priming within the priming experiment. The comparison between the pre-test, priming experiment and the post-test includes the other responses. It makes sense that the number of Other responses decreases over time, because participants give responses that are more similar to those of the confederate. But that might be the consequence of a more deliberate strategy or alignment, rather than cumulative priming really, it’s more of a task effect.

Experiment 2

Method

Even though the number of participants is larger than in Experiment 1, I am still a bit concerned about the statistical power as there are so many other responses, especially for the pretest. In both experiments, the number of Other responses is really high (especially in the pre-test). Could this have been prevented by means of some instructions perhaps (given that the proportion of other responses is much lower in the study by Bernolet (2008) which the authors model)?

Page 16, Table 3/Lines 616-617: the significance level of Order is only 0.09 (whereas the authors write **, which is marginal significant at best.

Figure 8a/8b: since the proportions of genitives and von-phrases are complementary (they add up to 100), I would suggest to plot only one of the two percentages (e.g., proportion of von-phrases). In that way, you can plot the two priming conditions within one graph, which enhances the comparison.

Page 18, lines 688-691: I do not agree that the findings really contrast with previous findings regarding inverse preference effects. Previous findings measured the inverse preference effect within the experiment and haven’t been looking at the long-lastingness of the inverse preference effect. The authors haven’t directly tested for the immediate inverse preference effect, since the priming experiment did not include baseline items. Here, the research question seems to be a different one, namely whether the inverse preference effect is long-lasting.  Of course, the findings have important implications: the inverse preference effect is said to reflect prediction error leading to long-term changes (also see the discussion on the frequency attenuation effect in lexical processing, Forster & Davis 1984, Nievas 2010). Also interesting is Ferreira (2003) on the inverse preference effect as the reflection of a mechanism to prevent infrequent structures to become unlearned.

Page 15, 715-718 and page 20, 795-805: Self-priming is arguably to be explained from activation-based or hybrid rather than error-based models and has been modelled by Jacobs et al., 2019. (Jacobs, C. L., Cho, S. J., & Watson, D. G. (2019). Self‐priming in production: Evidence for a hybrid model of syntactic priming. Cognitive science, 43(7), e12749)

General discussion

Page 19, line 751: what do the authors mean by “uniform”? Again, normally the inverse frequency effect is measured as an overall-effect over the entire experiment, and only on the short term.

Lines 758-762: While it is a good point that it may be relevant how infrequent the infrequent structure actually is, I think the authors are measuring something else than the inverse preference effect as it is measured in previous studies (or at least in a different way).

Lines 779-780: there is evidence of the inverse frequency effect, since there is initially a large boost of von-responses (due to large surprisal), which then decreases (the surprisal effect decreases). It’s just that the strong priming of the infrequent variant did not lead to more von-responses in the long term (in the post-test).

 Minor suggestions:

Page 1, line 19: the year 2000 is not so “recently”

Page 6, lines 254: don’t repeat the year 2008 within the same paragraph

Page 8, line 334: “was to investigate” instead of “was investigate”

Page 15, Figure 6: the line of the other responses is almost the same color as the shaded areas around the orange line, which is confusing. It would be clearer if a different color was used for the line of the other responses.

 

 

 

Author Response

please see attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

I really enjoyed this paper. It addresses current issues in structural priming (different types of preference effects), presents interesting data, and raises a number of questions for follow-up research. I appreciated particularly that the authors investigated previously untested possessive constructions. Their results demonstrate clearly how important this is, given that previous findings, most of which come from the dative alternation, do not necessarily generalize to other construction types. Another strength of the paper lies in the authors’ careful interpretations of their results and their acknowledgement of relevant limitations.

While I think the paper is already in good shape, I would ask the authors to consider the following comments to further improve the clarity and soundness of their argument:

Abstract

The authors mention twice that they conducted “two [structural] priming experiments”; once may be enough?

p. 1

“Recently, it has been proposed to explain structural priming by an error-based implicit learning mechanism (Chang, Dell, Bock, & Griffin, 2000).”
> I’m not sure if 23 years ago is “recently”?

p. 2

“Our data are evidence of preference” > or “provide evidence”?

“However, if the incidence of structures A and B is complementary, this difference does not imply that primes of type A have a stronger effect than primes of structure B.”
> I may have missed something, but I couldn’t quite follow this sentence, and the subsequent conclusion that inverse preference effects may be “counter-intuitive”. Perhaps the authors could add some additional (simple) explanation?

pp. 5-6

Fig. 1 shows a gardener and a cyclist with a cookie, but the example utterances in (4)-(6) are about a cyclist with a flower. Neither do the sentences match the pictures, nor can they be part of the same trial because the same character shouldn’t appear in both the verification and the description picture. Perhaps the pictures and sentences could be matched to facilitate the reader’s understanding?

Some additional pictures might also help. E.g., in Fig. 1, the authors could show a whole trial, i.e., a combination of verification and description picture, to illustrate the use of different characters and objects (i.e., no lexical boost). They could also consider adding a second illustration with example pictures for the pre- and post-test (to highlight the difference with the priming phase).

p. 8

Fig. 2 > Here and in all subsequent figures, I would find it more helpful if the y-axis was broken into increments of 10% (rather than 25%) – this would make it easier to “read off” the approximate magnitudes.

p. 11

“of a magnitude of above 17%, cf. Figure 3a” > Is this the descriptive difference, or the (back-transformed) model estimate?

“The model did not converge” > I also wonder if the small sample size (only 16 participants) may have played a role. Is there a reason why the authors ran the first experiment with a considerably smaller sample than the second one?

p. 12

“To sum up, there is evidence of cumulative priming, but there is no evidence of inverse preference effects in this experiment, and, if anything, evidence of (positive) preference effects, in that the proportion of use of the relatively preferred structure increases during the priming phase and then stabilizes at a high level during the posttest, while the less preferred structure does not stabilize.”
> I appreciate the authors' careful discussion, but I still wonder if their conclusions are not somewhat biased by their hypotheses. On the face of it, the main reason why they fail to establish an inverse preference effect for pre-nominals is that the relevant statistical model doesn’t converge; but being unable to assess an effect (which, descriptively, may be there) seems different from there being “no evidence”. The authors instead identify a positive preference effect for post-nominals, but they might want to highlight that this effect is relative to speakers’ overall productions (especially their ‘other’ responses); it does not concern the complementary proportions of post- vs. pre-nominals (which, clearly, shift in favor of pre-nominals between the pre-test and the priming phase).

p. 13

“Which structure is preferred is related to formal properties of the attributive noun phrase: […]”
> Could the authors provide examples to illustrate this point? This would also help the reader recall the prepositional structure for Exp. 2 (since no concrete example is given in the Materials section either).

p. 17

“This contrasts with the results for the von-phrase structures, which were produced at an overall lower proportion during the posttest …”
> Could the authors make clear that these are only descriptive differences, thus avoiding potential confusion (since the comment is included in the section on statistical modeling)?

p. 18

“A separation of trials by 697 prime structure is even more informative.”
> Mention again the caveat that this interaction was only marginally significant?

p. 19

“So, not every structure can be successfully primed” 
> At least, not on a long-term basis? The experiments still show short-term priming of the rare structures during the priming phase.

p. 20

As for the structural “readiness” effect, I understand that the authors want to leave the details of how this effect could be modelled to future work. Still, it would be helpful if they could provide some more insight into how they envisage the workings of this mechanism, and how it potentially relates to other literature. E.g., the authors briefly state that higher a priori readiness leads to stronger priming, but could they explain why this is the case? Similarly, they introduce the notion of conservativity as an aside, but could they elaborate a little on how this factor influences “the system”?

p. 21

“But they allow to ask precise questions to come to more conclusive evidence.” > perhaps not the optimal phrasing: “allow [whom?]”; “come to” = rather informal

“according to Kopf and Weber” > year missing

As a final thought, I wonder whether the register differences between genitives and von-phrases (the latter being potentially more informal, as the authors note) might play a role in Exp. 2 or could be further investigated in the future. Perhaps there is some sort of cumulative “pragmatic” priming going on, whereby speakers increasingly adapt to the formal experimental setting and thus come to avoid the informal von-structure? Perhaps this adaptation operates on a more conscious level and thus unfolds more slowly over time than the (presumably automatic, unconscious) structural priming? Just some wild speculations.

Author Response

please see attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The authors have made good improvements in the organization of the text and are more precise in their use of the terminology. The study is better framed theoretically and especially the introduction is much clearer now. I am also glad that the authors performed a grammaticality judgment task which shows that the rare pre-nominal genitive is still considered grammatical.

I am still not convinced by the small sample size of Experiment 1. I think it was not necessary to abort data collection in Experiment 1, since these 16 participants did produce already 28 prenominal genitives in the priming phase. It is true that it would still be difficult to perform statistical tests for the pretest and the posttest, but at least the order effects within the priming phase could be investigated more reliably if the sample size is appropriate. For the priming phase, there was no floor effect. It would make your conclusions that priming infrequent structures does not translate into the increased use in non-primed conditions immediately after the experiment, much stronger. If the authors cannot or do not want to increase the sample size, I doubt whether it is relevant to include it in the paper, in that sense that it has little added value to the findings of Experiment 2. The findings that are significant, are also significant in Experiment 2. Differences between Experiment 1 and 2 concern effects that are non-significant in Experiment 1 but significant in Experiment 2, so no conclusions can be drawn on these.

In addition, the authors still draw some conclusions which cannot be made om the basis of the statistical tests reported and there are still some inaccuracies in the use of the terms for different types of priming effects, as I will also point out in the specific comments. The main issue is that the different types of priming effects also require different types of statistical tests. It also means the effects rely less on each other and are less contradictory to each other than the authors state in several places.

·        Inverse preference effect: priming is strongest for the least frequent structure: statistically compare priming phase to baseline from pre-test (so do not call differences between the priming phase and the post-test (inverse) preference effects)

·        Cumulative priming: priming increases for a particular structure within the priming phase: statistically test for order effects within the priming phase

·        Long-lasting priming: the priming effects persist after the experiment: statistically compare the pre-test and the post-test (which is not done right now)

My suggestion would be to make these distinctions much more consistently throughout the paper. Discussing these three effects systematically will also help structuring the paper when discussing hypotheses, reporting the statistical tests and discussing the effects.

Specific comments:

Page 1, line 26: “the development” is a bit ambiguous term and may also refer to the development of priming effects in L2 speakers for instance. In this introductory paragraph, I would make it more explicit what you are referring to (e.g., the development of priming effects over the course of an experiment) to make sure that your readers are on the same page instantly.

Page 4, line 198-199: I do not understand why cumulative priming effects in produced primes and/or with the more balanced DO/PO alternation would go against the error-based learning account.

Page 5, line 240: The aim of the study should be formulated a bit more precisely or sharply. The theoretical impact of the current study is not so clear at this stage of the paper. Why do we need more evidence?

Page 6, line 283: what does “they” refer to? To inverse preference effects or to the priming effects themselves? In the first case, the magnitude of the difference between the priming of the frequent and that of the infrequent structure decreases; in the second case, the priming effects themselves decrease (which should then probably only apply to the  infrequent structure).

Page 12, figure 3a, 3b and 4: I understand why you kept the Other responses in your plot for the pre-test and the post-test, but at least for the priming phase, please (also) provide plots without the other responses, since you also exclude them for your statistical analyses. This would make it easier to see whether there is any numerical trend for non-significant effects.

 Page 14, line 512: I have difficulty following your explanation on what a main effect of Order would mean. The two target structures are complementary to each other, so an increase of one structure automatically implies a decrease of another structure. It is always about the relative frequency. So if the relative frequency of one structure is increasing, e.g. the prenominal genitive over the course of the experiment, this automatically means that the relative frequency of the other structure is decreasing. It cannot be the case that the relative frequency of both structures are increasing at the same time.

Page 14, line 531: since figure 3b includes the other responses while the statistical model does not, you cannot relate the absence of the interaction between priming and order to Figure 3. Priming is not measured relative to the other condition. If you don’t include other responses, an increase in one target structure always goes to the expense of the other target structure. In figure 3, you plot percentages. So when talking about your statistical analysis, you cannot say whether there was an increase of one genitive structure to the expense of another genitive structure.

Page 14, line 539-540: similar to my comment on Order, I think it is more accurate to say that in the case of a significant priming effect, one structure increases between the two phases to the expense of the other structure (so the relative frequency changes), and to avoid the term priming here.

Page 14, line 534-542: It suffices to say that you could not statistically compare the pretest to the priming phase and the posttest because of the absence of one the pre-nominal genitive. It is not necessary to say that the model did not converge, because how would you have expected to compare priming effects in the absence of data points in one of the phases?

Page 15, line 565: That is not what the significant intercept means, since you used dummy-coding here. The intercept only refers to the reference level, which is the priming phase. The significant intercept thus means that the postnominal genitive is preferred over others in the priming phase.

Page 16, line 604-605: again, the formulation of more cumulative priming of one structure than the other is strange.

Page 16, line 607-609: you cannot really say that the gain differs, even if you leave aside the issue that in studies reporting the inverse preference effect, the effect is not calculated with percentages relative to the other responses. Because you do not have any prenominal genitives in the pretest stage, you cannot calculate the percentage increase. So you cannot calculate the inverse preference effect between the pretest/posttest and the priming phrase (this is why a baseline interspersed with the trials is actually useful despite its drawbacks – it is not a true baseline but rather a neutral condition within a priming context, which allows to measure the inverse preference effect in immediate priming). Also, in terms of percentage point, there is a stronger increase of the postnominal genitive, but if you express it in terms of percentage increase, then the post-nominal genitive increases with 27.5%. If you would take 0.1% as a value for the prenominal genitives, then there is an increase of 1400% in the priming phase for the prenominal genitives… That looks like an inverse preference effect.

Page 16, line 623-631: it cannot be concluded that there is an absence of an inverse preference effect. So you cannot state that priming of the prenominal structure was not stronger than that of the postnominal structure. You don’t know based on your data. Therefore it would have been interesting to test a full sample, even if you would still not have any prenominal genitives in the pretest or the posttest. You could then have focused on any order effects within the priming phase and on the lack of persistence in the posttest phase.

Page 21, line 732-733: see previous comments. As you don’t have a baseline, the increasing use (which you call cumulative priming) of one structure implies the decrease of the other structure. So without a baseline, there cannot be cumulative priming of both structures at the same time (at least you cannot observe it).

Page 22, line 764: since you are talking about the genitive priming condition and you only expect an increase of genitives after a genitive priming, I think it is a bit confusing to call the decrease of von-responses a floor effect here.  I would not mention the von-responses, just stick to the ceiling effect of genitive responses.

Page 22, line 772-779: I find it surprising that you were not able to run a model, since you do have data points of both structures in all phases. Have you tried using optimizers (Bobyqa) and/or simplifying your model (leaving out random intercepts)?

Page 23, line 795: as for the post-hoc analysis in experiment 1, the intercept only refers to the priming phase here

Page 23, line 806: it may seem obvious numerically, but you did not provide a statistical test to show that postnominal genitives were more frequent than von-phrases in the pre-test and the post-test, so strictly speaking you cannot draw conclusions on this.

Page 23, line 814-816: That is a wrong conclusion. You find a significant decrease in the use of post-genitives compared to other responses between the priming phase and the post-test phase in your post-hoc test. Also, you did not compare the pre-test and the post-test. Therefore you cannot say that they are statistically different from each other. You can do pairwise comparisons for your model to see whether there is a significant difference between the pretest and the posttest.

Page 23, line 814-821: you cannot state that, as you did not statistically compare the von-phrase between the different phases.

Page 24, line 868-871: the two phenomena do not exclude each other and actually go hand in hand, given the way you measure it. You can still have an overall stronger priming effect for less frequent structures than for more frequent structures, while having cumulative priming of the more frequent structure. In other words, if the inverse preference effect decreases over the course of the experiment (which is to be expected when the surprisal decreases), there will be an increase of the relative frequency of the more frequent structure.

Page 25, line 897-899: see my comment to page 16, line 607-609

Page 25, line 905-909: no, the difference is that these studies include a baseline within the experiment, which allows to calculate the magnitude of the (cumulative) priming effect for each structure separately. See Experiment 1 in van Lieburg, Hartsuiker & Bernolet (2023), in which a very low-frequent structure was absent in the pre-test but was produced in the baseline condition of the priming experiment. Interspersing the baseline with the prime trial does affect the baseline compared to a non-priming environment due to (short) long-lasting priming, but it depends on what you are measuring whether that is a problem.

Page 26, line 976-977: here you are interpreting the absence of effects in Experiment 1, which you cannot do based on your low power

Proposal in general discussion: the mechanisms you describe are interesting, but as follows from my comments, I think the conclusions you draw do not always follow from the data and the statistical tests you performed. Therefore the proposed mechanisms may not follow from your data. This makes it difficult for me to comment line-by-line on the reasoning why the current models on structural priming do not suffice. My suggestion would be to first reconsider your statistical tests and conclusions and then revise the general discussion.  

Minor:

Page 2, line 98: delete capital P  in “priming”

Page 4, line 155: what does “a specific production range” mean? Please clarify.

Page 7, figure 1b: the caption is the same as for figure 1a, but it is a priming trial instead of a pretest trial

Page 14, line 520: inconsistent whether spaces are around the | symbol

Page 17, line 642: “compare 2 vs. 3 above” is vague, it is not clear what it refers to

Page 23, line 781: “the above analysis”: please explicitly mention to which of the above analyses you are referring

Page 24, line 830-831: this may need some reformulation, it is not entirely clear what you mean here

Page 25, line 887: “inverse preference effects” instead of “inverse preference effect” (or insert “the”)

Page 25, 927-928: since you did find immediate priming for all structures, I think this needs to be phrased more precisely.

Author Response

please see the attachment.

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

Comments and Suggestions for Authors

I think it is a good solution to report the experiment with prenominal genitives in Appendix 1 and I really appreciate the transparency the authors are pursuing.

Now that the used terminology is more precise and corresponds more consistently to the reported analyses, the argumentation in favor of preference rather than inverse preference effects is much more coherent and convincing. I still think from the data that there is a surprisal effect leading to large von-priming at the beginning of the experiment, but it is a very interesting observation that this effect decreases very quickly already during the experiment, and that it is rather the preferred structure that shows long-term effects of priming. It is interesting that the increased production of von-phrases seems to be limited to a primed context, and that participants are reluctant to produce the structure in unprimed context even if they are still in the experimental context.

I have a few minor comments left.

Overall: you mention “as mentioned above/below” relatively often in your paper, which is a bit confusing/disturbing sometimes. I think it happens in some unnecessary cases, and in other cases it might mean that you should introduce something earlier in the text. For the references that you keep, please be as specific as possible (e.g., in which paragraph the reader can find it).

Lines 214: It would be clearer to the reader if you immediately introduce the possessive structures (at least briefly, e.g. between brackets, if you want to keep the structure in which you elaborate on it further down below).

Line 481: PowerPoint presentation instead of power point presentation

Table 1, line 760: inconsistent delimiter (use . instead of , to indicate the decimals of the percentages).

Line 766-771: remove space before % symbol

Table 2, line 795: you say “with only the random intercept of participant did not converge”, but I think you mean “item” here, since your model does seem to include a random intercept for participants. I would still advise to try a BOBYQA optimizer to see whether you can have a random intercept for items included (the same for your other statistical model). This is how you can do that:

glmer(target ~ condition*order + (1|participant)+ (1|item), family="binomial"(link="logit"), control = glmerControl(optimizer="bobyqa",optCtrl=list(maxfun=50000), tol = .0001), data = df)

Line 1044-147: Importantly, the hinge on the repetition of head nouns only follows from residual activation accounts to explain the lexical boost, not to the explicit memory component of error-based learning (See eg. Kantola, Van Gompel, & Wakeford, 2023). According to explicit memory accounts, repetition of any noun should lead to a lexical boost. But as you did not repeat any noun, this is merely a matter of formulation.

Author Response

Please see attachement

Author Response File: Author Response.docx