Reviews

[jensroes]

I will prepare a response letter but the full email with all comments is below. It sounds positive and do-able.

Manuscript No. XGE-2024-2060 Typing in tandem: finite mixture models reveal fundamental parallelism in multi-sentence text composition Journal of Experimental Psychology: General

Dear Dr Roeser,

I have received reviews of the manuscript entitled Typing in tandem: finite mixture models reveal fundamental parallelism in multi-sentence text composition (XGE-2024-2060) that you recently submitted to Journal of Experimental Psychology: General. I was fortunate to receive comments and evaluations from individuals who are very knowledgeable and highly respected experts in the topical area you are investigating. As you will see when you read their critiques, the reviewers have offered many detailed points and constructive suggestions centered on improving the current paper.

I read the manuscript prior to receiving these reviews in order to gain an independent perspective on the paper, and then again with the reviews in hand. In the end, there turned out to be a considerable level of consensus among the majority of us with respect to the perceived strengths and limitations of the current paper. All of us found several aspects of the work appealing. For example, the paper is well written and the analytic approach is strong.

At the same time, however, the reviewers raised some concerns that prevented them from recommending acceptance of the paper in its current form. I share some of these same concerns. In particular, Reviewer 1 noted theoretical alternatives that should be considered. However, the primary points of concern are limited in number and the findings presented are promising enough that I would like to encourage you to submit a revision.

Please note that this is not a guarantee your manuscript will be published. If and when you submit the revised manuscript, I expect that you will take each of the concerns seriously and address these concerns in two ways.

First, when possible, you should make changes in the manuscript to correct shortcomings that the reviewers perceive. If there are comments that you do not find to be correct or apt, you still should consider that the incorrect perception is something that you might expect in other readers, so it would be helpful to take steps for the paper to anticipate such misperceptions and add clarifications in the text to prevent them. The goal is to make your paper as accurate, scientifically responsible, interesting, and accessible to a wide range of experimental psychologists as possible.

The second way that I would like you to respond to the reviewers is to put effort into a careful cover letter that goes through each comment in this editorial letter and each review point by point, explaining how you addressed each comment and, if you disagree with a comment, why you disagree (and, if possible, how you altered the writing in anticipation that other readers might have similar concerns). Please note that Reviewer 3's comments are in an attached annotated pdf document. Please also be sure to include a version of the manuscript that denotes the changes in some way (e.g., highlighting, alternative font color, etc.), as well as a 'clean' version of the manuscript.

ADDITIONAL REQUIREMENTS FROM THE EDITORIAL OFFICE: —Authors should include a final paragraph titled "Constraints on Generality" which "explicitly identifies and justifies the target populations for the reported findings" (Simons, Shoda, & Lindsay, 2017). The paragraph will be included in the word count for brief reports. Authors should also include a Public Significance Statement that briefly communicates the importance of the work to the general public; this statement should be written in paragraph form and will appear in a box below the abstract in published articles. Information about these journal policies are described in the editor's interview and the journal's web page. For a recent example, see: https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpsycnet.apa.org%2Ffulltext%2F2023-83147-001.pdf&data=05%7C02%7Cjens.roeser%40ntu.ac.uk%7C90c113b5397b41ca1ba008dcb4bc0de3%7C8acbc2c5c8ed42c78169ba438a0dbe2f%7C1%7C0%7C638583967343110180%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C&sdata=BJgBs1evCMtAoNZfEwjqOKgh37Fr1xV9Gu2yLpswSC0%3D&reserved=0

—Authors should include a subsection in the method section titled "Transparency and openness." Effective August 1, 2022, empirical research, including meta-analyses, submitted to the Journal of Experimental Psychology: General must meet the "requirement" level for citation, and data and code transparency, as well as the "disclosure" level for preregistration, materials transparency and design and analysis transparency. The Transparency and Openness subsection should detail the efforts the authors have made to comply with each of the seven the TOP guidelines. Please include the dataset's repository link (e.g. on ResearchBox or OSF) in this section, and include the data citation in your reference list. If you are unable to share the de-identified data and/or code due to legal or ethical reasons, this in and of itself is not a barrier to publication in the journal. In such situations, please mention the concern in the cover letter to the editor. The editorial team will consider exceptions to these requirements on a case-by-case basis.

—As part of your Author Note, please provide details (2-4 sentences) of prior dissemination of the ideas and data appearing in the manuscript (e.g., if some or all of the data and ideas in the manuscript were presented at a conference or meeting posted on a listserv, shared on a website, etc.).

—Identify the contributions of all authors using the CRediT taxonomy. If the manuscript is accepted for publication, the CRediT designations will be published as an Author Contributions Statement in the author note of the final article.

— APA guidelines recommend detailed and inclusive reporting of demographic information. For studies with participant samples, authors should report how the demographic information was collected. For example, were participants asked about their gender or sex? How was information about race and ethnicity collected? Were participants given options to choose from [If so, what were the options?] or was it a free-response box, etc.). In addition, please also report the number of participants in each group, rather than reporting the numbers for a single group. Please avoid using "female" and "male" as nouns, as per the APA style guidelines (7th ed, pp. 138-141, section 5.5).

—Please be sure to follow APA style guidelines. Double-space all content and use minimum 11-point font. For readability during review include figures and tables in the text. 


To submit a revision, go to https://eur03.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.editorialmanager.com%2Fxge%2F&data=05%7C02%7Cjens.roeser%40ntu.ac.uk%7C90c113b5397b41ca1ba008dcb4bc0de3%7C8acbc2c5c8ed42c78169ba438a0dbe2f%7C1%7C0%7C638583967343119405%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C0%7C%7C%7C&sdata=tUs3Ewff2yMHxGBW4ZeXIuytatVkPdnYn9s7AbQ0QK0%3D&reserved=0 and log in as an Author. You will see a menu item called "Submissions Needing Revision". You will find your submission record there. If the opportunity to revise is for some reason closed on the web when you are ready with a revision, please contact us to re-open it rather than submitting the paper as new. Also, at the top of the manuscript and cover letter, please write "Revision of XGE-2024-2060 as invited by the action editor, Michele Diaz, Ph.D.." If possible, I would like to receive your revision by 10/30/2024. If this is not feasible, please email our Peer Review Coordinator, Magen Speegle, at the main editorial office (consult.MSpeegle@apa.org) with an estimate of when you will resubmit. Longer timeframes are fine.

Sincerely, Michele Diaz, Ph.D. Associate Editor Journal of Experimental Psychology: General

Reviewers' comments:

Reviewer 1: This is a well-written manuscript describing an interesting and well-conducted analysis. The topic will be of high interest for those working on written language production and particularly on typing behaviour. The quality of the analytical approach is very high, but a more thorough discussion of the theoretical implications of the results and potential alternative interpretations is, in my opinion, necessary. MAJOR ISSUES:

• In my opinion, the main issue of the Introduction is the lack of clarity about what is the assumed planning unit. It seems to be the sentence, but this is not explicitly acknowledged. For example, in page 6: "Under this theory, planning of the next language unit can only start when the previous unit has been fully processed, from communicative intention to completion of the written output. Therefore interkey intervals represent periods when the writer is planning what to write next: interkey interval will be longer at linguistic edges where planning scopes over larger text spans and their duration represents the time necessary to complete that planning". More examples of lack specificity about this and assumptions of sentence-level planning are present in other parts of the manuscript (for example, see page 9 end of second paragraph or page 29 when describing predictions of the serial account). If one assumes that planning may involve a few words, a clause or a full sentence depending on the circumstances (so the unit of planning is not always the sentence), the absence of substantial differences between the pauses at the beginning of sentences and words could be reconciliated with a serial view of the production process. From this point of view, planning would happen serially but not always at the sentence level, making similarly likely to find instances of planning at the beginning of other words that do not initiate sentences. What is the proposed to be the unit of planning is then key for the argument the authors want to make, as only with clarity on this respect one can understand which pauses are not supposed to be increased in a purely serial processing (in no case, within-words pauses would be affected in a serial view, for example).
• Related to the previous point, it would be important to have a summary of the positions samples for the analysis in the between-words conditions. Pauses that separate a subject and a verb within the same clause might be different from those who are placed between clauses. More information about this (if even as a summary) and potential separate analyses in line with proposed unit of planning would be helpful.
• In page 9, first line of second paragraph: "The semi-parallel cascaded account of the composition…". The described account of the composition process it is not, as far as I can see, cascaded. Or at least is confusing on what is proposed in this respect. The model includes buffers that "provide transient storage at each processing level to accommodate unsynchronised output". I find this very difficult to integrate with cascaded processing of information, in which information flows downstream before processing has finished. The point of the buffer seems to be, in fact, to preclude cascading of information (this is, to keep the output of a level of processing "in wait" until the immediately lower level of processing has finished ongoing processing of earlier units). More clarity and specificity about the nature of flow of information needs to be provided. I can see that the account of the process proposed is parallel, but not sure if it is cascaded. Whether or not this is the case, it is key to the interpretation of the results and where hesitation in hesitant interkey intervals actually comes from.
• Page 30, second paragraph. I might have missed an explanation for this, but why exactly the authors think that parallel processing leads to increased time between words but not within-words? It is not clear why unfinished planning cannot affect any interkey interval , rather than only between-words intervals.
• Page 31, second paragraph. I more in depth discussion of what might happen in terms of processing in younger or less skilled writers should be provided. Related to my first point, if the present results just show that planning sometimes involves longer units (full sentences) and sometimes involves shorter units (a few words), then the discrepancies between data sets might arise from the fact that less skilled writers might plan for shorter units.

MINOR ISSUES:

• Page 2 first line of first paragraph. I think "data" is missing after "their".
• Page 4, first paragraph. "Translating" is a term very specifically used in writing research to refer to the processes involved in transforming the (more) abstract communicative ideas into written language. However, it does not seem to be used in this widely accepted sense in this first sentence of the article (as it seems to include communicative intent and idea generation). I recommend using a different term.
• Page 5, line 1 of second paragraph. "form" should be "from".
• Page 8, line 2 or second paragraph. "composting" should be "composing".
• Page 9, line 6 of second paragraph: "However if one or more upstream processes provide output more slowly, then inter-key intervals are determined just by time taken to complete upstream processes and not on finger movement". I might be misunderstanding this, but surely interkey interval duration in this case will depend both on the time needed to complete higher-level processes AND the time taken to move the fingers. I am not sure how one can avoid be affected but the time needed for the specific finger movements….
• Page 16, line 2: "and transitions between characters within words all transitions". I think something is missing or grammatically incorrect in this sentence.
• Page 20, line 4 "the first file". I think this might be "the first line", but I am not sure.
• Page 23, line 5. "before word" should be "before-word", I think.

Reviewer 2: The manuscript presents the analysis of a combination of datasets of typing sentences. It follows nicely from previous work of the first author (Roeser et al., 2021), thereby validating their method of using mixture models to recover several distributions in the data on large datasets. The analysis is appropriate, and support the conclusions the authors make. However, I think the theoretical impact of the work should be developed further, in particular for a venue like JEP:General. I have made several comments in that direction.

1- a. You use the term "semi-parallel" to contrast with the serial view, however that term is never clearly defined. In particular, you should state what makes your assumption not fully parallel, and what constraints you introduce from the parallel model that is mentioned in the review of the literature. Since the term "semi-parallel" is used throughout the paper, this should be made very clear. "Parallel" is still used in some places, so I would advise to be consistent in this choice of term, once it is properly justified.

b. Is the serial model still viable? Recent data have already put it into question, even the authors' last work. Are you using it as a strawman, or is there reason to believe that it could explain typing behavior in some contexts? If that is the case, these contexts should be made explicit.

c. Hypotheses of the semi-parallel cascaded model and what is expected from the statistical analysis isn't clear in the hypothesis section. Please specify more clearly what results you are expecting to be able to validate this model.

2- a. On p.9, you state: "However if one or more upstream processes provide output more slowly, then interkey intervals are determined just by time taken to complete upstream processes and not on finger movement". I don't understand why finger movement execution isn't taken into account in addition to the time to complete upstream processes. I would expect these two processes to be additive. Otherwise, what accounts for finger movement execution in the case of slower upstream processes? Actually, there are several places where you seem to consider them additive. p.19 and Appendix A, p.51: "The δ parameter was constrained to be positive, so that it captures how much longer hesitant interkey intervals are in addition to β. » And that's indeed what the model seems made to capture (p. 19: ∂ in addition to ß in equation (1)). It also seems implied in the discussion, but not really explained. Then, could you clarify what happens with keystroke execution in the case of slower upstream processes in the context of a semi-parallel model?

b. Could you hypothesize on what each of the distributions map onto? For instance, keystroke execution and planning? Or if this isn't appropriate, could you explain why we shouldn't think of each distribution mapping to specific process(es)?

3- Comparison of datasets The reader doesn't have enough information on each dataset to really appreciate the comparison of the posteriors on Figure 2, which is a shame. I wonder if plotting parameters as horizontal plots like in the tutorial would be more appropriate.

Moreover, you don't interpret the differences in proportions between locations. In some datasets, hesitations are equally likely before sentences and words. Only 2 datasets show significantly higher probability of hesitations before sentences than words. This is in contrast with your predictions and previous results! For young and L2 writers, hesitations are as probable before a sentence and before a word. Would you interpret this result as word-to-word planning during typing? Does it mean that there isn't a lot of sentence planning going on before beginning to type a sentence? This could (and should) be interpreted further.

p.31: I would also suggest reorganizing the discussion sections to be more systematic in terms of the comparison of age, task, and other factors if relevant (e.g., text masking).

Overall, you have a very rich dataset and it seems like you are not really taking advantage of it. I understand your approach of making general conclusions and showing that this analysis is able to explain several and quite different datasets. And this is great! But it seems like a missed opportunity to not discuss some of the interesting effects you have. Maybe they are in line with previous conclusions by the authors of each dataset or would make a significant addition to previous reports.

4- Isn't it likely that a mixture model is always preferred? Could you implement some sort of control? I appreciate the simulations to show that a mixture model is preferred when there is actually a mixture. However, real data is always going to be noisier than simulations. Could you run simulations or make that argument on something closer to real data? For instance, is there a case of real data where the no-mixture model is preferred?

5- For fluent intervals, you only allow the variance of interkey intervals to vary by location. Therefore, it seems logical that you don't find an effect of location on distribution means. When you write "the distributions of fluent transitions - indicated in grey - are constant across transition location" (p.22), you should specify that it is in terms of variance, but not means, if that is indeed what you find.

Minor comments

6- p.4: "Written communication is also less time constrained: speaking comes with fluency requirements because hesitating during speech has a communicative effect". This statement depends on the context of written production. If you refer to written communication, then there might be time constraints (e.g., finish typing a text before you get a new message). Maybe it would be more appropriate to use "written production" instead. Or you could move the example of writing an essay that is currently at the end of the paragraph to the beginning, so the reader gets the appropriate context right away.

7- Please include a reminder of the thresholds used for interpreting BF as a note of Table 6.

8- The manuscript contains a substantial number of typos (below), please have the whole manuscript checked for any other ones.

p.4: mentally > mental p.5: form > from p.9: the the > the Table 2: below word > before word p.20: first file > first line p.29: at durations to motor execution at durations > to motor execution at durations p.30: please check "Thus, although, …" p.31: of one the source > of the source

Reviewer 3: This paper reports an important study aiming at contributing to improve models of text production. Authors have gathered and used for their study six different datasets from different colleagues who collected on-line data (keystroke) from multi-sentence text production in different projects. These 6 datasets are used in this paper to test the validity of the the serial processing account and the parallel processing account in models of text production. To this end, interkey intervals at before-sentence, before word, and within word transitions have been analysed through two sets of Bayesian mixed effects models instantiating the assumptions of the serial and the parallel processing account. Models are then being compared, resulting in evidence in favour of the parallel processing view in text production. In the attached file, some improvements are suggested. You might consider changing the title for more straighforward understanding of the aim of the paper. Several passages are unclear and call for rewording. A rationale for the use of the 6 different datasets might be useful earlier in the paper. Clear explanations of what authors consider as hesitation and pause are needed. is there a next step to your study (with other data sets, other models, etc.). Would this methodology apply to handwriting ?

XGE-2024-2060_reviewer.pdf

[Mark-Torrance]

This is brilliant! I really wasn't expecting such a positive response. But if you send an intelligent paper to intelligent reviewers... :)

I agree that we can deal with this. What deadline are we working to? I've told Emily that I can't find time to help with revision on the writer behavioural stability paper until October. But this is a more important paper.

[RConijn]

Wohoo! Very positive and helpful comments indeed! Happy to assist (I'm on holiday from Aug 19-Sept 6).

[jensroes]

Re deadline it says "If possible, I would like to receive your revision by 10/30/2024. If this is not feasible, please email our Peer Review Coordinator, Magen Speegle, at the main editorial office (consult.MSpeegle@apa.org) with an estimate of when you will resubmit. Longer timeframes are fine."

[Mark-Torrance]

That's very possible. I won't work on this until mid September, but it should then not be a massive job.

We could have a writers' retreat in Eindhoven? Serious suggestion. Could just be two days. Jens and I put a lot of work into this when we were in Iowa, and it was really helpful actually both being in the same room, and free from other distractions. We have money to spend on this.

[jensroes]

Reviews are here as table: https://docs.google.com/document/d/1jhDl6F83iwc8YFbmpVufHf5FlWgVifO9NE-d7Aa-sw0/edit?usp=sharing