The Influence of Case and Word Order in Child and Adult Processing of Relative Clauses in Greek

Round 1

Reviewer 1 Report

This study explores the online processing of subject and object RCs in Greek. It compares processing of RCs in two conditions: (i) a complementizer condition (pu), where case is only marked on the head and the RC DP; and (ii) a relative pronoun condition, where case is also marked on the pronoun, disambiguating early the subject vs object status of the relative clause. It also manipulates word order by having the RC DPs either in preverbal or postverbal position, thus resulting in a combination of 2 SRCs (canonical SVO, scrambled SOV) and 2 ORCs (canonical OSV, scrambled OVS). 

This is an ambitious study which provides evidence from an under-researched language whose morphosyntactic properties make it a good testing ground for theories of RC processing. It is also novel in its attempt to explore the timecourse of SRC and ORC processing in both children and adults and the results are interesting. The authors’ argument should however be clarified in places and the hypotheses should be outlined much more clearly, both in the literature review and in the discussion. Moreover, the results should be presented more clearly addressing each of the aims in turn, i.e.: the role of word order, the role of case, and the developmental question. I think the paper should be published on condition that the following points are therefore clarified. 

 

On RCs in adult processing: 

Broad summary of universal/parametric accounts is clear, but I found the review of the studies that looked at languages with free word order (i.e. those that are directly relevant to the present study) very vague (lines 148 ff.). Unclear what ‘partial evidence for expectation-verb theories’ means. 

Please spell out clearly what predictions those theories make when the RC is introduced by a case-marked relative pronoun, and what those studies found. 

The discussion of these studies mentions case as a cue on the relative pronoun but does not mention word order at all. Did those studies also manipulate canonical vs scrambled word order?

Please end the section with clear hypotheses and predictions.

 

On RCs in acquisition:

Whereas studies in the previous section are discussed in the context of disentangling the predictions of some competing approaches, the studies in this section are presented outside of a coherent theoretical debate, which makes it hard to see what the authors’ aims are here. 

Guasti et al. (2012): what do their results suggest regarding the role of case?

Line 240=Guasti et al. (2012) again?

Line 243=Varlokosta et al (2015). Results are discussed in light of RM but what about the theories discussed in the previous section? Why is RM not introduced before among the grammatical/parametric approaches?

Line 281=genuine minimality effects in non-canonical structures. Which of those in (4-5) are non-canonical? 

The two studies which used online methods with children are only introduced at the end of this section (from line 300) and only very briefly. I think the first part of this section could be shortened significantly to leave more space for these studies.

Felser et al. (2003): tested children who are close to the age examined in this study but the results are not discussed. Which language did they test? What kind of support for continuity did they show? For a study that is aiming to look at the online time course of RC comprehension in children, it would be useful to have a much more detailed discussion of the previous findings and to have those findings integrated in the broader discussion around memory-based vs expectation-based explanations.

The aim of testing children appears to be simply to explore whether parsing strategies in older children are similar to adults’. But to me neither memory-based nor expectation-based accounts have much to say about continuity – children might have the same parsing strategies as adults and still incur in higher processing difficulties precisely because of memory limitations/higher reliance on frequency. It would be useful if the authors clarified what kind of result they would take as supporting vs challenging continuity.

According to the final section, the evidence on the role of word order is mixed (Schelstraete and Degand 1998, advantage for OVS, Guasti et al. 2018, advantage for OSV) – since this is directly relevant to the study, could the details be spelled out more clearly?

Guasti et al. (2018). In what way do the predictions of Fodor and Inoue’s analysis and repair model differ from other memory limitation theories? 

Line 335=OVS —> the lions [that hits the horse]; authors say that up to the verb the structure is analysed as SVO then a reanalysis is needed at the post-verbal DP, but doesn't the singular morphology on the verb already disambiguate?

Please end the section with clear hypotheses and predictions.

 

Methodology and analysis:

The methodology and the analysis are described in good detail, but given the complexity of the design I think there are some points which the authors could clarify to facilitate the interpretation of the results. When the stimuli are introduced in the materials section, it needs to be to clarified what the regions of interest are for the 4 structures. Typically SPR analyses focus on a critical region and the two spillover regions following it. As the authors note later on (but only in the discussion), a comparison between the pre-verbal and the post-verbal structures here is complicated by the fact that disambiguation occurs at different regions: at region 3 in the case of canonical SVO and scrambled OVS, at region 4 in the case of scrambled SOV and canonical OSV. This means that the verb is a spillover region on the SOV and OSV conditions, but actually a pre-critical region on the SVO and OVS conditions. I think that this makes it hardly surprising that LTs are higher at the verb region in the OSV condition compared to the other conditions. 

I am also not sure why adults’ and children’s LTs are compared against each other. I think the results would be easier to read if they were reported separately by group. The aim here is to see whether there is a similar timecourse of RC processing along the critical regions, regardless of which group is faster.

Please include error bars in the graphs.

To make the differences among the lines more visible, personally I would eliminate regions 1 and 2 (where nothing is happening).

 

Discussion:

Since the aim of the paper is to explore the role of both word order (canonical vs scrambled) and case in the processing of RCs, and also to explore adults and children’s sensitivity to these cues during on-line processing, it would seem more useful to me to address each of these in turn. The fact that the discussion is organised, like the results section, by comparing LTs at each region of interest for each condition makes it hard to extract the key points. From my understanding, children’s timecourse of RC interpretation was actually very different from adults’ when structures with the same word order (=hence same regions of interest) are compared, especially in the PU condition. In the preverbal conditions (SOV, OSV) adults’ processing is entirely as expected: no sig difference at the DP(region 3), but a SRC advantage in the two following regions (region4, region5). For the children, as far as I can tell, the SRC advantage never reached significance. In the postverbal conditions (SVO,OVS), my understanding is that the SRC advantage emerges for both groups at the final region. So in the post-verbal structures children’s processing seems more adult-like. On the other hand, in the relative pronoun condition, children didn’t seem to rely on DP case at all when processing SRCs and ORCs with the same word orders (no sig difference at any region, either for SVO vs OVS or for SOV vs OSV). This is quite interesting to me but it took me a while to extract this information because a direct comparison between these structures is only presented (lines 692 ff.) for the verb region. 

I think it is possible, as the authors suggest, that children’s faster reading times (and lack of sig effects) at the final regions in the relative pronoun condition (for preverbal RCs) might actually indicate a failure to build the structure in real time – but if this was the case one should find a correlation between LTs and grammaticality ratings in the children’s group. I think the authors could run a model including children’s ratings and their LTs to test this.

 

Other comments

lines 78-79=(3) and (4) should be (2a), (2b)

line 202=chaces > chases

line 240=in more recent studies

line 248 = (4) should be (5)

line 299 = remove ‘is’

line 311 = acquired by 

line 707= check sentence

line 760 = remove ‘occurred’

line 776 = repetitive sentence

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper well addresses the issue related to the different models that account for the difference in processing between subject and object relative clauses, providing new insight for the debate on the universal complexity accounts, memory limitations theory and expectation-based theory. The authors interestingly use as variables the word order in S/O relative clauses and the characteristic of the relativizer (complementizer/ pronouns) in a rich morphology language like Greek in a self-paced listening task. The results are quite interesting since, except for the overall data of acceptability judgement in which SR are preferred to OR, the advantage of Subject relative clauses is softened by the presence of a case marked preverbal NP in the relative clause, and on the processing of nominal elements within RCs (postverbal vs preverbal status).  Furthermore, the study confirm that non crucial differences are formnd between adults and children.

 

I just missed some reference to the Self-Organized Sentence Processing of Tabor & Hutchins (2014) which actually make prediction (not always compatible with cue-based model) on the role of when the disambiguation element bis found. Such a model could be useful to explain sime characteristics of the results and furthermore since the authors make reference to  when the marked pronoun/complemntizers and Marked NPs, I was surprised that not even a footnote on the SOPE model was made. I would suggest to add some footmnote on how the data fit in the SOPE model.

 

I suggest warmly to accept this paper almost as it is, which some minor comments on how the glosses were written and very very few typos in how the sentences are written. I attach the original document with some comment directly in the text on some minor remarks. 

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This study makes a new contribution to a body of research on the processing of subject and object relative clauses in Greek. However, I do not think that the manuscript is publishable in its current version. I have several concerns with the experimental design and statistical analyses that I detailed below.

 

• The authors mention in the introduction that the different word orders serve different pragmatic functions. The authors should give more information about pragmatic felicity of the different word orders and explain if these were taken into account when designing the stimuli.
• Page 7: “The children in our study are older than previous studies in children because we wanted both children and adults to do the same task.” The age range of the children selected is not fully justified. Children can participate in self-paced reading experiments starting at age 6-7.

• The listening times reported in the tables A1-A2 seem very long. Did the authors calculate residuals? To control for the length of each segment, raw RTs should be transformed into residual RTs. Residual RTs should be calculated by subtracting the participants’ RTs from the length of each segment.
• I do not think that the children and adults’ results are directly comparable, because the administration of the task was different for the two groups. I understand that adults completed a longer version of the task than children: for adults the complementizer manipulation as within-subject and for children, it is a between-subject. This factor is then analyzed separately and no justification is given.
• The authors compare the segments that appear in different sentence positions across sentences. For instance, the verb region appears in some stimuli after the complementizer and in some stimuli after the embedded subject. The same segment is going to be processed differently if it appears in different positions in a sentence, because different processes will be involved, based on the preceding material that has been encountered until that point (e.g., complementizer vs. subject noun). However, these differences do not tell us anything about where reanalysis occurs or where processing may be more effortful. The meaningful comparisons are between sentences that have similar word order, where upon encountering a disambiguation region, revision of the initial interpretation occurs (e.g., accusative vs. nominative NP; complementizer vs. relative pronoun, etc..).

Minor comments

• In the experiment, participants judged all sentences for grammaticality. How many of the fillers were grammatical/ungrammatical?
• The authors say that children could not cope with the task. Does it mean that the task given to adults was too long for children, or too difficult? This should be explained more clearly.
• Was item used in the analysis for random intercepts and slopes?
• Results of the models should be reported in tables, so that it is possible to check all the significant values for main effects and interactions.

 

 

 

Author Response

Please see attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

I would like to thank the authors for addressing my comments/questions. I think that the manuscript is now ready for publication.