Paradigmatic Uniformity: Evidence from Heritage Speakers of Spanish

Round 1

Reviewer 1 Report

Thos paper reports on the results of an acceptability task, applied to heritage speakers of Spanish in the US. The main aim is to test theoretical hypothesis on the acquisition/knowledge of subject-verb agreement in bilingual speakers, namely if speakers show different sensibility to finite vs non-finite mismatches. I think the paper is strong from a theoretical perspective, but the methodological dimension is weak and should be improved. I added my comments to the pdf.file.

Mainly, the study should include a control group of monolinguals, should control better for proficiency and give a clearer picture of the participants' profiles. The introduction has several generalizations that should be avoided.  

Comments for author File: Comments.pdf

Author Response

Please see attached

Author Response File: Author Response.pdf

Reviewer 2 Report

This is an interesting and original paper which examines different accounts of potential problems that heritage speakers might have with inflectional morphology, relating to agreement mismatches. The paper discusses accounts based on: (i) use of default/underspecified forms (substitution of finite 3PS or nonfinite forms for other finite forms) versus (ii) surface similarity (between different forms in the verbal paradigm), as well as potential effects of frequency in the latter case. The paper then tests the second type of account (surface similarity and effects of frequency). Spanish subject-verb agreement is investigated in an experiment involving heritage speakers. The paper is clearly written and presented in an organized way.

One concern, however, is whether the paper meets the requirements of the special issue on language contact and crosslinguistic influence. Although the paper is about heritage language (Spanish) inflectional morphology, there is no mention whatsoever about how the inflectional phenomena function in the majority language, English, and whether any influence would be expected. Indeed both theories addressed in the paper, namely the default hypothesis and the surface similarity hypothesis, would seem to have nothing to say about contact/crosslinguistic influence. Rather, these are accounts that seek to address problems with inflectional morphology regardless of the language combinations involved.

General issues that need to be addressed in more detail:

1. Motivation for looking at agreement mismatches in the heritage context needs to be provided. It is not at all clear why proficient heritage speakers should be expected to have problems in the first place. This is known to be an L2 acquisition problem but why should heritage speakers show the same difficulties as L2ers in this area?

2. Potential problems with the hypotheses.
The relationship between the default hypothesis and the surface similarity hypothesis is not clear. If there are no relevant similar forms, do defaults take over? In other words, are the two theories alternatives or complementary?

The experiment then goes on to test only the surface similarity (OO) claim. It needs to be made clear earlier in the paper that this is the issue under consideration, rather than the default question. (Or perhaps I am misunderstanding the discussion of Burzio on page 4, in which case clarification is needed.)

The OO similarity claim regarding similar finite forms is presumably specific to Spanish, since there are languages where one could get an OO correspondence between a finite and nonfinite form (e.g. German).

Hypothesis 1.a. i. Is the claim neutral about the direction of the difference or is it the claim that mismatched finite forms will be more accepted than nonfinite? In either case, this needs to be explained.
1.b. Could the claim be that mismatching nonfinite forms will be more acceptable than finite?

There is no hypothesis as to what the predictions would be under a default account and yet this has been presented as one of the accounts under consideration. I don’t think that the default account would in fact predict any differences in acceptance rates for defaults (whether finite or nonfinite) and correctly agreeing forms. So there is a problem here in that the default account potentially makes the same prediction as 7.a.ii. This all needs to be clarified.

3. Materials.
Test items involved the following: S1-V3, S2-V3, S1/2/3-Vinf.
If the default hypothesis is being tested (and, as already mentioned this is not clear), then these are not sufficient, since all items involve potential defaults (3PS or infinitives). Rather, verb forms should be included that are clearly not defaults. For example, speakers should reject ungrammatical S1-V2, S2-V1 to a greater extent than ungrammatical S1-V3, S2-V3 or S1-Vinf, S2-Vinf. In any case, grammatical forms like S1-V1, S2-V2 would be desirable, because (as can be seen in Table 2) almost all of the experimental items are ungrammatical.

4. In the L2 literature on agreement mismatches, the focus has mostly been on production, in other words trying to account for agreement errors that L2ers might produce. In the current paper, the focus is on acceptability judgments. This needs to be mentioned and highlighted more; in particular what aspects of the grammar are claimed to be potentially problematic? Presumably, heritage speakers are not going to hear incorrectly agreeing forms, so how does the issue of defaults or surface similarity arise? Why should a default theory based on DM expect any problems in interpretation? Why should an OO theory predict such problems?

5. Discussion
The last paragraph speculates on what monolinguals might do. I do not see why monolinguals would be expected to show any acceptance of agreement mismatches or what this would show if they did. The claim is made that ‘if we were to find similar contrasts but perhaps lower ratings ...., this would be related to the bilingual nature of heritage speakers’. I fail to follow this line of argument. Surely, if bilinguals were different from monolinguals, this would be more likely to be due to their bilingualism than if they are similar.

In general, the discussion needs to raise some ‘big picture’ issues. At present, it is mostly a summary of the hypotheses and the results.

More detailed comments:

p. 2, line 65 states ‘Subject-verb mismatches have not received much attention in the literature’. It needs to be made clear that this is in the heritage literature, since they have received a great deal of attention in the L2 acquisition literature (and also L1), and the author in fact discusses this literature in some detail.
Line 75-76 mentions a ‘similarity of surface forms effect’ as it this has already been demonstrated, whereas it is in fact something that is to be tested. So this need to be rephrased, perhaps as ‘...a similarity of surface forms effect would confirm accounts ....’

p.3, line 130. Not clear what is meant by ‘they correspond to a general rule and also to the absence of certain features’. In what sense do infinitives correspond to a general rule? What rule?

p. 5, lines 240-241. If 3rd person is also a default, one would not predict differences between 1st and 2nd versus 3rd person.

p. 6, 1st paragraph. The choice of fillers seems somewhat arbitrary. A justification for the sentence types chosen as fillers would be useful, as well as some examples.

I presume that grammaticality is controlled for via the fillers – this needs to be added to the paragraph at the top of page 6. In other words, we must be told how many grammatical and ungrammatical items there are overall.

Section 2 should be divided into two: 2.1 Materials; 2.2 Participants. And possibly the participants should be presented before the materials.

p. 9, figure 4. This is a useful summary of the results. I would have found it helpful to have this presented first, before the more detailed results in Figs 1 – 3.

p. 10, lines 365-369. Is an ‘elsewhere rule’ different from use of defaults? And how does markedness come into it?

p. 10-12. Appendix. Please indicate for each item whether it is grammatical or ungrammatical. Also, organize the list in terms of the different sub-types tested, or add a column which indicates this.

Minor points:
Misuse of ‘cf.’ at various points. ‘Cf.’ does not mean ‘see’. Rather, it implies a contrast. More appropriate to use ‘e.g.’ in most of the places where the author has used ‘cf.’.

There are some very minor typos, etc., indicated directly on the ms. (attached).

Comments for author File: Comments.pdf

Author Response

Please see attached

Author Response File: Author Response.pdf

Reviewer 3 Report

Review of: “Paradigmatic uniformity. Evidence from heritage speakers of Spanish”

General comments:

The idea behind this paper is extremely clever: not only does it examine an issue that has been underexplored in the literature, but it does so in an innovative way (talking into account the nature of the verbal forms involved in S-V agreement mismatches). While the theory and rationale behind the study are impeccable, the experimental design, the statistical analysis as well as the way in which certain aspects were controlled (participants’ level of proficiency, type/token frequency of the target verbs used) need to be improved before this manuscript can be accepted for publication. The reason why I suggest such changes is because their inclusion (or a more accurate control of how they are operationalized) might significantly affect the results of the study.

First of all, I recommend that the author/s rewrite the first research question driving the study (“[..] from the point of view of acquisition and bilingualism, are mismatches the result of bilingualism?”). Without further explanation, this question is too difficult to answer/operationalize. For one, its exploration would need the presence of a non-bilingual control group (which I would like to mention this study does NOT need, but to determine whether a particular outcome is a result of bilingualism, it would need to be present in some way). Instead, I suggest they reflect upon the idea behind that initial question:

1. Did they bring up the idea because they assume that heritage bilinguals might exhibit higher rates of S-V agreement mismatches that non-bilinguals? To test variability without having to include a control group, I suggest exploring the degree to which HL proficiency affected participants’ ratings. This would indirectly get at the same issue that the author/s may have been referring to with their initial RQ.
2. Seeing that mismatches were far less likely to be accepted if the verbal form was not marked (i.e. infinitival forms), it would be interesting to see that build into the study’s driving questions. By not doing so, it feels that the presence of infinitival vs. “marked” (for a lack of a better word) mismatches is rather random, when that is not the case. Thus, I suggest that the RQs include something about HSs’ potentially being more likely to accept the former (rather than agreement mismatches involving infinitives), dispelling the notion that their grammars are vastly different from those of controls. This change would also reinforce the importance of the results obtained when surface-form aspects are taken into account, as there are specific lexical characteristics of the verbs that increase/decrease the likelihood of detecting potential S-V mismatches (setting up a “continuum” of acceptability: stem-stressed > affix-stressed > infinitives) à this is hinted at in p.4-5 (when the author/s mention the connection between finiteness and OO correspondence), but it should have been mentioned before (and more explicitly).

Second, and despite the fact that I agree with the author/s that S-V agreement mismatches have not been frequently explored in HL acquisition, there are several studies that should have been mentioned in the literature review (see suggestions included at the end of the document). As the author/s will see, the observations provided in these articles/books are also very anecdotal, which reinforces their point and increases the value of the present study -while acknowledging the previous literature-.

Experimental suggestions:

• More tokens for each condition are needed, especially since the 4 items in each combination (S1-V3, S2-V3, etc) is also divided in two based on the stress (affix vs. stem). If one assumes that lexical knowledge will have an impact on HSs’ performance (as shown by ), it is important to determine that a certain finding (i.e. mismatches featuring stem-stressed verbs seen as being more acceptable than affix-stressed verbs) is not a result of participants’ knowledge of that particular item (for example familiarity with “traer”) –rather than an effect of a general category (stem-stressed forms)–.
• There are a couple of issues that are not entirely addressed in the paper:
  • How was frequency measured (or operationalized)?. It is important to provide this information because it could change the outcomes of the study (as the author/s mention, this variable has been found to significantly modulate HSs’ performance -especially in the domain of verbal morphology-). The author/s only include “high” or “low”:
    • is this referring to lemmatic or token frequency? In the case of the latter, of which forms? (I would suggest looking up token frequencies for the target forms only “trajo” vs. “metió” for example).
    • where did they obtain the ratings from? (self-ratings, corpora, intuition…)
  • Although I understand that the argument about “improved” acceptability is related to phonological factors, all stem-stressed forms are instances of irregular verbs, and this factor is not mentioned explicitly at any point in the manuscript. Bringing this to the fore would not only connect the present study to previous research on the impact of morphological regularity in HL grammars, but would also provide a better connection between irregular/regular forms and their lexical frequency. 
  • Since stem-stressed verbs are also higher in (lemmatic, token?) frequency than affix-stressed ones, how do we know if it is frequency rather than the type of stress what is modulating HSs’ rates of acceptability? This could be solved by adding more items in each condition with various ranges of lexical frequency (+/- frequent). This would allow the author/s to compare forms within the same category (and they would also have a more balanced design).
  • On a related note, how comparable are the high and the low frequency pairs? If we hope to make any claims about the effects of frequency verbs within each condition would need to be comparable in this aspect.

• • Based on the SD reported in Table 2, there does not seem to be a huge amount of variability in participants self-reported proficiency. However, I think it’s worth including this information to the model (maybe by creating an aggregate scores of all linguistic abilities) to see whether it modulated participants’ responses.

Analysis

1. There needs to be a more complete report of the modeling and the output obtained (R2 values of each of the models, coefficients (log odds and odds ratios), SE, p values). This could be included on a table.
2. From what I understood from the description, this was a design with repeated measures (participants were exposed to each condition more than once). Given the design, I think having a mixed effects models with fixed and random factors would be more appropriate. I am not familiar with R, but I see that there is a specific package (MIXOR package) that allows for the inclusion of random slopes and intercepts with ordinal dependent variables. Having a mixed effects model would allow the author/s to add frequency to this model as well (which would have 1-5 rating (ordinal dependent variable), type of match (matching, non-matching), type of form (finite, non-finite), person (1,2,3) and lexical frequency as fixed effects and subject and item as random effects. This type of modelling would allow the calculation of just one model (instead of several, as argued in the current proposal). On that note, I don’t seem to have seen any random effects included in the regression. Given the likelihood of inter-participant variability (very common among HSs) I would suggest having this factor as well as Item to account for individual and item variability.
3. I am unsure if the nested comparisons are part of the same model or not (I think it’s the latter, as the author/s mention the presence of several regressions. I think having a mixed effects model (and looking at the pairwise comparisons later across the different levels) would solve this issue. See comment #2 in this subsection.

Other comments and minor edits:

l.6 please change “hypothesis” to “hypotheses” (and maybe even adding “several” before hypotheses to anticipate the content of the paper).

l.16: consider changing “higher ratings” to “higher acceptability (ratings)”.

l.26: consider eliminating the second instance of “person” to avoid repetition.

l.31: change example number from (what I assume is) 0 to (2).

l.44: I understand the need to define HSs in that sentence, however, not all HSs exhibit low-proficiency in their HL, and this sentence could be currently interpreted as such (consider re-writing).

l.58: Something similar to what I pointed out in the previous comment occurs here. As it stands, the start of this paragraph can be interpreted as this being the only type of agreement mismatch involving finite forms that is document in HSs’ performance (and that is not the case). I would reformulate this sentence so that it reads like:

1. Subject-verb mismatches can also occur with finite verbs.
2. I will focus on those that involve 1/3p sg pairs (i.e. Yo trajo)

l.79: change “clashe” to “clash”

l.93: change “non-agreement” no “non-agreeing” to complete the parallelism with “agreeing”.

l.107: review sentence starting with “Van Patten […]”, negation missing.

l.137: “In this sense, one might expect bilingual speakers to draw on these similarities in instances of agreement mismatches.” à while I agree with the author/s, this needs more unpacking and a justification. I suggest using the work by Kim (2020) on HSs’ use of suprasegmental cues in the perception (and production) of lexical stress.

l.139: I think the claim about the potential effects of frequency needs to be unpacked a bit more by applying it to the data examined. How would we expect this variable to affect the forms analyzed in this particular study? This may not be necessary if the expansion of Bybee’s surface-network model included the information, but it is only touched upon briefly at the end of the next paragraph. Given the focus on preterit forms, I also suggest incorporating Bybee & Brewer’s (1980/2007) chapter on morphophonemics and Spanish preterite forms (see ref at the end).

l.163: I just wanted to mention that the connection between the different types of agreement mismatches found in HSs and the REH is brilliant. Very clear, testable hypothesis which elucidates what factors might be modulating performance.

l.213: I think CAN.t-o should be replaced by ME.t-o, to maintain the parallelism.

l.219: Delete “have” after surface and review sentence (it seems to be missing something).

l.284: “and those correspond to all varieties of monolingual Spanish” – since no control data is provided, I would reformulate this claim (or provide references to back it up).

l.285-293: what does the number 2.2 refer to? It is unclear whether it refers to finite or non-finite mismatches. This comment is connected to a suggestion I provided at the beginning of the review (see section on Analysis), where I mentioned that a more complete report of the results is needed.

Figure 1. I would suggest changing the type of graph (or maybe increasing the size so that the probabilities in the y axis were more visible).

l.309: more specifics about the R2 of the models is needed here to evaluate the effectiveness of the modeling. Comparisons between those that included certain fixed effects and those that do not are necessary in this type of models.

l.316-: In this line, it is mentioned that “Ratings were generally higher for stem-stressed verbs than for affix-stressed verbs […]”, however, is not clear whether statistical comparisons were run and if they were significant or not. I have a similar comment for the results in Figure 4. Are the differences between S1V3- S2V3 and infinitives different? More details about the statistics are needed.

References included in the review:

Albirini, A., Benmamoun, E., & Saadah, E. (2011). GRAMMATICAL FEATURES OF EGYPTIAN AND PALESTINIAN ARABIC HERITAGE SPEAKERS’ORAL PRODUCTION. Studies in Second Language Acquisition, 33(2), 273-303.

Bolonyai, A. (2007). (In) vulnerable agreement in incomplete bilingual L1 learners. International Journal of Bilingualism, 11(1), 3-23.

Bybee, J. L., & Brewer, M. A. (1980). Explanation in morphophonemics: changes in Provençal and Spanish preterite forms. Lingua, 52(3-4), 201-242. (also included in Bybee’s 2007 book).

Kim, J. Y. (2020). Discrepancy between heritage speakers' use of suprasegmental cues in the perception and production of Spanish lexical stress. Bilingualism: Language and Cognition, 23(2), 233-250.

Sánchez, R. (1983). Chicano discourse: Socio-historic perspectives. Arte Público Press.

Shin, N. (2018). Child heritage speakers’ morphosyntax: Rate of acquisition and crosslinguistic influence. The Routledge handbook of Spanish as a heritage language, 235-253.

Silva-Corvalán, C. (2014). Bilingual language acquisition: Spanish and English in the first six years. Cambridge University Press.

Author Response

Please see attached

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I have nothing to add. The authors have addressed my comments. 

Author Response

Thank you for your comments throughout the process, they have been very useful

Reviewer 2 Report

This revised version has covered the issues that I raised in my original review. The paper has been considerably changed, and improved as a result. The only thing that I am unsure about is the addition of the Nevins analysis in the Discussion. It is a bit distracting. Since this analysis did not work for the data, it is not clear what the point is to having it there.

There are a number of minor issues which I have indicated on the attached ms.

Comments for author File: Comments.pdf

Author Response

Thank you for your detailed comments and suggestions. I am particularly grateful for your careful reading and noticing typos, which I apologize for. I have incorporated all your suggestions.

Reviewer 3 Report

Please see attached form.

Comments for author File: Comments.pdf

Author Response

Thank you for your careful reading of the second version and for your suggestions. I have incorporated all of them and checked again the frequencies. The new "CREA anotado" allows one to search by lemma or by form, and the difference is noticeable.