Unmasking individual differences in adult reading procedures by disrupting holistic orthographic perception.

Word identification is undeniably important for skilled reading and ultimately reading comprehension. Interestingly, both lexical and sublexical procedures can support word identification. Recent cross-linguistic comparisons have demonstrated that there are biases in orthographic coding (e.g., holistic vs. analytic) linked with differences in writing systems, such that holistic orthographic coding is correlated with lexical-level reading procedures and vice versa. The current study uses a measure of holistic visual processing used in the face processing literature, orientation sensitivity, to test individual differences in word identification within a native English population. Results revealed that greater orientation sensitivity (i.e., greater holistic processing) was associated with a reading profile that relies less on sublexical phonological measures and more on lexical-level characteristics within the skilled English readers. Parallels to Chinese procedures of reading and a proposed alternative route to skilled reading are discussed.

Introduction

Visual word identification serves as the foundation for skilled reading comprehension. Word identification occurs when the phonological and semantic representations of a word are accessed in response to seeing a visual word form. In an alphabet like English, there is evidence that visual words are coded both holistically and analytically, with holistic representations capturing the mapping between a specific visual form and its lexical equivalent, and analytic coding capturing the mapping between sublexical orthographic units (e.g., graphemes, bigrams, etc.) and their phonological equivalents. The current study uses a simple behavioral marker of sensitivity to the spatial orientation of print to investigate individual differences in orthographic coding amongst native English speakers, and their relationship to the reading procedures that support word identification and comprehension.

Holistic versus analytic orthographic coding

Orientation manipulations have historically been used to differentially disrupt visual object recognition of items that are processed more holistically (e.g., faces) compared to items that are processed more analytically, or in a piecemeal manner. An atypical orientation is thought to disproportionately affect holistically processed objects. In some accounts, this is because the holistic process cannot be applied when the object is presented unconventionally [1, 2]. This forces a switch in recognition strategy towards a more analytical or feature-based approach, which is less than optimal and more error prone for holistically processed objects like faces [3]. Others have been argued that inverted faces are eventually processed holistically [4], but even in this case one would still expect that a ‘holistic’ strategy would take longer and potentially be more error prone, because the presented orientation is suboptimal for holistic processing.

Similarly, individual biases for holistic versus analytic orthographic coding can be determined by manipulating the orientation of visual word forms and measuring the impact it has on word recognition. The logic rests upon the same concepts that apply in the face processing literature [5]. For example, the idea that an atypical orientation forces a more analytical approach in word identification is supported by the emergence of a word length effect for rotated words [6, 7], but not upright words. The reasoning is that if a typical left to right decoding is disrupted due to atypical orientation of a word (Coltheart et al., 2001), an analytical approach becomes necessary (i.e., relying more on sub-word units) and longer words should take more time to decode than shorter words [8]. This approach of disrupting the typical presentation has recently been used to measure holistic coding in visual word processing [9, 10] and uncover cross-linguistic biases in orthographic coding [11-13].

Implications of orthographic coding procedures for word identification

The question of interest for the current study is whether individuals with potential systematic differing preferences for holistic versus analytic orthographic coding exhibit corresponding differences in the procedures and skills they use for word identification. Evidence for a link between orthographic coding and the procedures used for word identification comes from cross-linguistic studies comparing Chinese and Korean bilinguals reading English. Chinese is a morpho-syllabic writing system, and so sublexical orthographic coding and mapping to phonology is arguably less useful for word identification, as compared to the holistic coding and mapping of characters to their morphemic forms, although the extent of reliance on holistic processing of Chinese characters might depend on an individual’s writing experience [14]. Korean, in contrast, is a highly consistent alphabetic writing system in which reading instruction emphasizes the decoding of printed words based on sublexical orthographic-phonological correspondences. Three studies have used visual form manipulations to investigate whether Chinese and Korean indviduals bring biases from their native writing system to English reading [11-13]. In all three of these studies, Chinese-English bilinguals exhibited greater holistic orthographic coding and a bias towards lexical-level processing to support word identification, while Korean-English bilinguals exhibited greater analytic orthographic coding and a bias towards sublexical and phonological processing to support word identification. For example, Ben-Yehudah and colleagues (2018) found that Chinese-English bilinguals’ naming times were more sensitive to word inversion and lexical frequency, whereas Korean-English bilinguals were relatively unaffected by word inversion and more sensitive to spelling-to-sound consistency [8]. Thus, prior cross-linguistic research provides evidence that individuals can exhibit differing biases for holistic versus analytic orthographic coding of English, and this differing bias is associated with a preference for lexical versus sublexical reading procedures.

The current study extends this prior cross-linguistic work by testing for similar patterns of individual differences amongst native English speakers. The study uses an orientation manipulation as a functional marker defining two groups who show a bias toward either holistic or analytic orthographic coding. Then, to test for associated biases toward lexical vs. sublexical reading procedures, the two groups are compared for their sensitivity to psycholinguistic properties of words in an overt naming task. Specifically, the design of the naming task manipulates the following word properties: 1) lexical frequency (how often a word appears in databases of written text), 2) imageability, 3) consistency/regularity (whether the pronunciation of a word is predictable based upon the spelling of its rime body (consistency); whether it follows the grapheme-phoneme correspondence rules of the writing system (regularity)), 4) length, 5) bigram frequency, and 6) biphone frequency [15].

The predicted outcome is that individuals with a bias toward holistic orthographic coding (i.e., high orientation sensitivity) will have a bias toward lexical reading procedures, as indicated by a: 1) heightened sensitivity to frequency, because frequency effects are widely regarded as a measure of lexical-level influences [8, 16–18]; 2) greater sensitivity to imageability, because it is a semantic measure that is inherently processed at a larger grain size [e.g., morpheme or whole word; 18, 19], and 3) reduced sensitivity to consistency/regularity effects in addition to bigram and biphone frequency, because they are widely regarded as measures of sublexical influences on orthographic-phonological mapping [8, 16, 20]. An increased length effect is expected for atypically presented words [6, 7] for individuals with a bias toward a lexical-level reading procedure, as an atypical presentation should necessitate a greater reliance on sublexical processing (i.e., their less-preferred reading procedure). Opposite reading patterns would be expected for individuals with a bias towards sublexical reading approaches.

Implications of orthographic coding procedures for lexical representation

Another component that influences reading procedures is the structure of one’s lexical representations, or lexical integration. Lexical representations consist of three constituents: orthography, phonology, and semantics and individual differences in the quality of these knowledge components affect reading processes [21]. The structure of lexical representations can be quantitatively described using a factor analysis to capture the correlational structure of performance on tasks that emphasize orthographic, phonological, and semantic knowledge. Phonological decoding, or the correspondence between orthography and phonology that allows a reader to correctly pronounce a word [22-24], has been highlighted as a particularly important foundational skill in word identification. Phonological decoding requires knowledge of sublexical orthographic-to-phonological regularities to pronounce words, and is commonly measured using a nonword reading task. This line of reasoning leads us to test whether individuals with more holistic vs. analytic orthographic coding exhibit structural differences in their lexical representations. We hypothesize that individuals with greater holistic orthographic coding (and a bias towards lexical reading procedures) will have a lexical representational structure without tight correlations to measures of sublexical phonological decoding.

Summary of study

In summary, the current study extends what we know about the link between orthographic coding and reading procedures from cross-linguistic work to individual differences within native English readers. The overarching hypothesis is that native English readers who show more sensitivity to atypical orientations of printed word forms are more reliant on holistic orthographic coding, and in turn possess a distinctive reading profile. We expect this reading profile to be similar to Chinese-English readers, showing a greater reliance on lexical-level reading procedures and reduced reliance on phonological decoding in the representational structure of word-level processing.

Materials and methods

Group definition criterion

Participants were recruited from a database of 411 individuals interested in study opportunities. Participants completed a series of screening tasks on a computer. The screening was limited to tasks that participants could respond to with button presses. Non-monolingual individuals and participants who identified as having trouble reading or a history of reading disorder were not eligible.

Participants were initially identified using data from a lexical decision task (see below), which included words with typical and atypical (180° rotation, see Fig 1) orientation. To focus on individuals with average reading ability, participants outside the 25th to 75th percentile in their median reaction time (RT) to typically presented words were removed, leaving 203 from an initial 411 potential participants. Accuracy and reaction times were calculated. Incorrect trials were removed from the reaction time analyses. The orientation sensitivity of each participant was calculated as the ratio of median RT for inverted stimuli divided by the median RT for upright stimuli. The median ratio of the remaining participants was 1.42 and the standard error (SE) was .037. Group cutoffs were delineated as the median ± 2 SE (rounded to 1.5 and 1.35). Thus, participants with higher sensitivity to atypical orientation (HS) were defined as those individuals with RTs for atypically oriented stimuli (words and nonwords) that were at least 1.5 times greater than typical stimuli. Participants with lower sensitivity to atypical orientation (LS) were defined as those individuals with RTs for atypically oriented stimuli that were less than 1.35 times that of typical stimuli. Participant attrition, due to graduation since participation in initial screening and eligibility requirements for a parallel imaging study (beyond the scope of the current study), also reduced the potential participant pool. A total of 31 eligible participants were run in a second behavioral testing session. A subset of these participants (N = 22) was run in a companion imaging study [25].

Fig 1

Example of upright and inverted word presentations in lexical decision task.

In order to ensure robust group assignment, sensitivity to orientation was also computed in an overt word naming task (see Materials, Overt word naming) that manipulated atypical orientation. This task was completed in the second behavioral session. Six participants whose overt naming scores were neither above nor below the median orientation sensitivity that was consistent with their initial group assignment based on the lexical decision task were removed.

Participants

All final participants were native monolingual English speaking undergraduates with no reported history of hearing or vision issues, learning or reading difficulties, drug or alcohol abuse, mental illness, or neurological problems. All final participants scored above the 20th percentile of Raven’s Matrices [26] and were dominantly right-handed. All participants provided informed consent and were given class credit for their first session and were monetarily compensated for an additional second session. Fourteen Lower Sensitivity Readers (LS) (2 males; mean age = 20.1 years, SD = 2.62) and 11 Higher Sensitivity Readers (HS) (4 males; mean age = 19.3 years, SD = 0.65) participated in the final experiment (see Table 1). All participants provided informed consent of approved experimental protocols through the University of Pittsburgh IRB, and were compensated for their time.

Table 1

Group statistics.

		Mean	Std. Error	t	d	p
Word Identification	LS	547	2.09	0.68	0.23	0.51
	HS	545	2.92
Vocabulary	LS	14.43	0.47	0.78	0.31	0.44
	HS	13.82	0.66
Spelling	LS	2.06	0.09	-0.21	0.10	0.83
	HS	2.10	0.13
Phonological Awareness	LS	102	3.34	-0.25	0.09	0.81
	HS	103	2.46
Phonemic Decoding	LS	521	1.88	0.52	0.29	0.61
	HS	519	2.05
Comprehension	LS	539	1.80	1.79	0.69	0.09
	HS	534	2.32

Independent samples t-tests were performed. Effect size is reported as Cohen's d.

Materials

Lexical decision

Sensitivity to atypical orientation was initially assessed using a lexical decision task in which the stimuli were presented in upright and inverted (rotated 180 degree, see Fig 1) orientations, the same atypical configuration used by Ben-Yehudah and colleagues [11]. Ben-Yehudah and colleagues used a naming paradigm, but here a lexical decision task was initially used to collect orientation sensitivity data without overt responses due to logistical constraints. Words were blocked such that upright stimuli were presented first, followed by inverted stimuli, with words and nonwords randomized within a block. Words were chosen to neither have extremely high nor low lexical frequency (min log HAL frequency = 6.27, max = 12.85, mean = 8.89). There were 20 words and 20 nonwords, with half of each presented in each orientation.

Overt word naming

Sensitivity to atypical orientation was confirmed using an overt naming task in which half of the stimuli were presented in typical orientation and half were presented in a reversed (FLIGHT → THGILF) orientation. Rather than use the same distortion manipulation as the lexical decision task, we visually distorted the words in a different way, and reason that similar group assignment provides a robust generalization of orientation sensitivity beyond inversion, and is consistent with diverse approaches used in the cross-linguistic literature. The stimuli were the 465 monosyllabic words used by Graves and colleagues [15] in a parametric neuroimaging study of word recognition. The items vary in length and along lexical (e.g., frequency and imageability), and sublexical (e.g., consistency, bigram frequency, biphone frequency) factors, and were selected to ensure that all factors are uncorrelated with each other within the stimulus list. The set of stimuli assigned to each orientation condition was matched along each of the dimensions sampled by Graves et al. [15].

Reading skills

The reading skills of participants were assessed using three subtests of the Woodcock Reading Mastery Tests [WRMT-Revised, Form H; 27]. The WRMT subtests included word identification (Word ID), in which subjects read aloud a list of words, phonemic decoding (Word Attack), in which subjects read aloud a list of nonwords, and Passage Comprehension, which requires subjects to supply missing words that best fit the context of short passages. Phonological awareness was assessed using the Comprehensive Test of Phonological Processing (CTOPP) subtests, Elision and Blending Words, that make up the Phonological Awareness Composite Score [28]. Vocabulary was assessed using the Wechsler Adult Intelligence Scale (WAIS-IV) Vocabulary Test [29]. Spelling was tested using the Lexical Knowledge Battery developed by Perfetti and Hart [21], which consisted of a list of 70 correctly and 70 incorrectly spelled real words.

Procedure

Participants were tested individually in a quiet room during two sessions. The first session was a behavioral screening implemented in an online platform for a larger project. It included the Lexical Decision Task, which was used to identify potential participants for the current study, a Spelling Test, and Ravens Matrices [26]. In the Lexical Decision Task, participants made a two alternative forced decision about each stimulus (real word or nonword), presented in blocks (upright and inverted). The task was self-paced, with the next word appearing after each decision had been preceded by a fixation cross for 500ms. During the Spelling Test, participants saw an entire list of correctly and incorrectly spelled words, and were asked to select the ones that they were confident were spelled correctly. The remainder of this session was devoted to tasks for an ongoing database data collection, and it will not be discussed further. Participants were given a break after each task.

After participants were identified as eligible (see Group Definition Criteria above), additional testing took approximately two hours, including a mandatory 10-minute break halfway through the session. During the two-hour session, a sequence of tasks was administered in a predetermined constant order across all participants.

The WRMT subtests [27], WAIS Vocabulary Test, and CTOPP subtests were administered according to the published procedures. For each test, the experimenter gave detailed instructions and practice items, if applicable. All of the participants understood the instructions and completed the practice items successfully.

The overt naming task was administered using a Dell Dimension DIM4700 computer with a 17-inch screen. The stimuli were displayed using E-Prime software (Version 1.1, Psychology Software Tools, Inc., Pittsburgh, PA). The participants viewed the stimuli from a distance of approximately 50 cm. A voice key incorporated into a serial response box (Model 200A, Psychology Software Tools) recorded the time it took the participant to overtly pronounce each item from the moment it appeared on the screen (i.e., RT); accuracy was coded offline from a recording of the participant’s overt responses. Participants read aloud words that were presented in a typical or atypical (reversed) orientation. The stimuli appeared at the center of the screen, in black lowercase letters against a white background.

The orientation (typical or reversed) of the displayed items was blocked; therefore, each item list was associated with either the typical or the reversed condition. Across participants, the order of the display condition was counterbalanced, such that half of the participants began with the typical orientation and the other half began with the reversed orientation. We chose to block the orientation of the stimuli to avoid task switching confounds [30].

Each block began with a cue indicating the orientation of the displayed items. Within each block, the trial began with a 500 ms black fixation-cross followed by the stimulus, which appeared at the center of the screen and remained there until the participants responded. Following the overt response, the item was replaced by a fixation cross that cued participants to press a button when they were ready for the next trial. Within each block, items appeared in random order, without replacement. Participants were instructed to read the items presented aloud as quickly as they could without making any errors. Each participant completed practice trials before data collection commenced: 3 typical and 3 reversed items.

Data analysis

For WRMT subtests, W-scores were used for statistical analyses. W-scores provide an equal-interval measure of test performance and they are the preferred measure for most statistical comparisons of group differences [27]. Raw scores for Elision and Blending Words were converted to standard scores, and then combined and converted to the Phonological Awareness Composite Score. Raw scores for WAIS Vocabulary were converted to scaled scores. Participant performance for the Spelling test was measured by d′ (i.e., sensitivity to misspelled words).

Results

Group statistics on reading skills

There were no significant differences in any component of reading skill between the two groups (Table 1). No measure even approached significance except comprehension.

Factors affecting overt word naming

Reaction time

We used a linear mixed effects model to understand the impact of lexical and sublexical factors on overt word naming RTs. All incorrect trials were removed. The model included fixed effects of group (HS readers coded as 1, LS as 0), visual presentation orientation (typical = 0 or atypical = 1), lexical and sublexical factors (frequency, consistency, imageability, bigram frequency, biphone frequency, and length), and 2-way and 3-way interactions between group, visual presentation, and lexical/sublexical factors. All lexical and sublexical factors were mean-centered. The model also included random intercepts for individual words, participants, and the effect of lexical frequency across participants. The model was fit using the R software [31] and the lme4 package [32]. Of special interest were the 3-way interactions between group, visual word presentation, and each of the lexical and sublexical factors. A significant interaction would indicate that a particular lexical or sublexical factor has a larger influence on RT in one group over the other, depending on the orientation of the word. We hypothesized that the HS group should be more affected than the LS group by lexical-level factors when words are atypically oriented, whereas LS should be more affected than HS by sublexical-level factors when words are atypically oriented.

Results of all main effects, 2- and 3-way interactions are reported in Table 2. There were no significant interactions between Group and lexical/sublexical factors when looking at just typically presented words, but interactions with group emerged when looking at just atypically oriented words (see Fig 2). Most notably, there was a significant 3-way group x Presentation Orientation x Length interaction, such that the effect of length was relatively larger for HS readers when the words were presented in a reversed orientation. This suggests that HS are relatively slower for longer reversed words compared to LS readers. There were also two marginally significant 3-way interactions with Group (Group x Presentation Orientation x Imageability and Group x Presentation Orientation x Biphone Frequency). The 3-way interaction between Group, Presentation Orientation, and Imageability was such that HS were relatively faster to name words that were imageable when they were presented in a reversed orientation. The opposite pattern is seen with the interaction between Group, Presentation Orientation, and Biphone Frequency, such that it was the LS readers who were faster to namewords that had higher biphone frequency, when words were presented in a reversed orientation. A power analysis [33] suited to linear mixed models [34] was run, which estimates the power for specific effects in our model using Monte Carlo estimation. This was conducted for each of the three 3-way interactions that showed effects, set at 200 simulations. The 3-way interaction between group x orientation x length had a very large effect size, and the observed power was 96.50% (95% confidence interval: 92.92, 98.58). The two marginal effects both had 51% power (43.85, 58.12).

Table 2

Results of linear mixed model with reaction time as dependent variable.

	Estimate	Std. Error	t	P
(Intercept)	498	52.4	9.51	< .001	***
Group	-47.0	78.7	-0.60	0.56
Presentation Orientation	210	6.95	30.2	< .001	***
Frequency	-35.6	9.44	-3.77	0.00	***
Imageability	-13.2	5.38	-2.45	0.01	*
Length	20.3	8.86	2.29	0.02	*
Consistency	-3.18	0.78	-4.10	< .001	***
Bigram	-1.66	12.8	-0.13	0.90
Biphone	-1.77	1.18	-1.51	0.13
Group x Presentation Orientation	268	10.5	25.4	< .001	***
Group x Frequency	-10.2	11.8	-0.87	0.39
Group x Imageability	0.21	6.08	0.03	0.97
Group x Length	-9.49	10.0	-0.95	0.34
Group x Consistency	-0.28	0.87	-0.33	0.74
Group x Bigram	14.7	14.4	1.02	0.31
Group x Biphone	-0.33	1.33	-0.25	0.80
Presentation Orientation x Frequency	-17.8	8.48	-2.10	0.04	*
Presentation Orientation x Imageability	-8.82	5.78	-1.53	0.13
Presentation Orientation x Length	74.6	9.45	7.90	< .001	***
Presentation Orientation x Consistency	-0.28	0.83	-0.34	0.74
Presentation Orientation x Bigram	-5.54	13.7	-0.41	0.69
Presentation Orientation x Biphone	-1.28	1.25	-1.02	0.31
Group x Presentation Orientation x Frequency	4.05	12.9	0.32	0.75
Group x Presentation Orientation x Imageability	-16.9	8.75	-1.93	0.054	.
Group x Presentation Orientation x Length	55.7	14.3	3.89	< .001	***
Group x Presentation Orientation x Consistency	0.89	1.25	0.71	0.48
Group x Presentation Orientation x Bigram	-24.5	20.7	-1.18	0.24
Group x Presentation Orientation x Biphone	3.63	1.90	1.91	0.056	.

*** = p < .001

* = p < .05,. = p < .10

Effects involving lexical factors are highlighted in gray, and effects involving sublexical factors have a white background.

Fig 2

Reaction Time for Group X Presentation Orientation X Psycholinguistic Factor 3-way interactions.

Data for typical and atypical word presentations are graphed separately. Data points are reaction times from individual trials pooled across participants with outliers (± 2 standard deviation) removed.

An additional model was run to examine a potential concern that group selection should not be based on the data run in this analysis. The new model additionally included the previously removed participants and coded “group” as a continuous variable based solely on the lexical decision task (LDT) and not the overt naming data. The 3-way interaction between “group” (i.e., strength of inversion sensitivity in LDT task) x presentation orientation x length remained highly significant and even stronger (t = 6.12, p < .001). The two marginal 3-way interactions (between group x orientation x imageability and biphone frequency) were no longer significant in the new model, but another marginal 3-way interaction emerged in the predicted direction (t = 1.79, p = .07), such that bigram frequency was a better predictor of RT for atypically presented words in those whose LDT were relatively small (i.e., those with low sensitivity).

Accuracy

Naming accuracy was close to ceiling (LS: M = .96, SD = .20; HS: M = .93, SD = .25), which posed problems for convergence of a general linear mixed effects model (glmer). Therefore, a secondary analysis was conducted using a weighted empirical logit model, which is designed for cases of near floor or ceiling performances (i.e., the accuracy probability across subjects is near zero or one) [see 35]. An ‘empirical log odds’ of accuracy (log (correct trials + .5/incorrect trials + .5) was computed for each word in one of four bins, crossing group x presentation orientation (i.e., HS typical words, HS atypical words, LS typical words, LS atypical words). Each value was also weighted to account for a different number of total trials for a given word in a given bin, due to unequal sample size in each group, etc. The model was then essentially the same as the reaction time model, with the exception that there was no random effect of subject.

Results of all main effects, 2- and 3-way interactions are reported in Table 3. Of note were a significant 2-way interestion between Group X Biphone Frequency and marginally significant 3-way interaction between Group x Presentation Orientation x Biphone Frequency. In both, LS has a larger positive relationship between accuracy and biphone frequency. The marginally significant 3-way interaction suggests that this effect is slightly amplified when words were typically presented words.

Table 3

Results of weighted empirical logit model with accuracy log odds as dependent variable.

	Estimate	Std. Error	t	p
(Intercept)	2.41	0.04	59.51	< .001	***
Group	-0.28	0.04	-6.30	< .001	***
Presentation Orientation	-0.19	0.04	-4.51	< .001	***
Frequency	0.21	0.05	4.35	0.00	***
Imageability	0.09	0.03	2.72	0.01	**
Length	-0.02	0.06	-0.36	0.72
Consistency	0.01	0.00	1.58	0.11
Bigram	-0.10	0.08	-1.30	0.19
Biphone	0.02	0.01	2.79	0.01	**
Group x Presentation Orientation	-0.23	0.06	-3.77	< .001	***
Group x Frequency	-0.01	0.05	-0.19	0.85
Group x Imageability	0.00	0.04	-0.04	0.96
Group x Length	0.02	0.06	0.31	0.75
Group x Consistency	0.00	0.01	0.18	0.85
Group x Bigram	-0.07	0.09	-0.79	0.43
Group x Biphone	-0.02	0.01	-2.06	0.04	*
Presentation Orientation x Frequency	-0.04	0.05	-0.74	0.46
Presentation Orientation x Imageability	-0.02	0.03	-0.67	0.51
Presentation Orientation x Length	-0.07	0.06	-1.22	0.22
Presentation Orientation x Consistency	0.01	0.00	1.52	0.13
Presentation Orientation x Bigram	0.03	0.09	0.31	0.76
Presentation Orientation x Biphone	-0.01	0.01	-1.59	0.11
Group x Presentation Orientation x Frequency	0.11	0.07	1.58	0.11
Group x Presentation Orientation x Imageability	0.02	0.05	0.42	0.67
Group x Presentation Orientation x Length	-0.11	0.08	-1.31	0.19
Group x Presentation Orientation x Consistency	-0.01	0.01	-1.27	0.20
Group x Presentation Orientation x Bigram	0.03	0.13	0.25	0.80
Group x Presentation Orientation x Biphone	0.02	0.01	1.85	0.06	.

*** = p < .001

* = p < .05,. = p < .10

Effects involving lexical factors are highlighted in gray, and effects involving sublexical factors have a white background.

Principal component analyses

Lexical representational structure

Recognizing that group numbers are small, exploratory factor analyses were computed in each group separately [36]. Lexical representational structure was assessed using a principal component analysis (PCA) in each group separately to characterize the structure of the relationship between Word ID and three tasks that emphasize knowledge of lexical constituents (i.e., spelling, which assesses orthography; phonological awareness, which assesses phonology; vocabulary, which assesses semantics), and phonemic decoding (which assesses knowledge of the correspondence between orthography and phonology). PCA using a Varimax rotation with a Kaiser normalization was employed. Only eigenvalues greater than one were considered for component identification.

In LS, the PCA identified a single component, accounting for 55.06% of the variance. In contrast, the PCA in HS identified two components that accounted for 78.43% of the variance (49.77% by the first, and 33.92% by the second) (Table 4). The first component had high factor loadings for Word ID, spelling, vocabulary (.898, .814, and .886 respectively), medium factor loading for phonological awareness (.452), and low for phonemic decoding (-.175). The second component had high factor loadings for phonological awareness and phonemic decoding (.858, .897).

Table 4

Factor loadings for lexical representational structure.

	Component		Component
Lower Sensitivity Readers	1	Higher Sensitivity Readers	1	2
Word Identification	0.894	Word Identification	0.898	0.25
Vocabulary	0.823	Vocabulary	0.886	0.149
Spelling	0.487	Spelling	0.814	-0.266
Phonological Awareness	0.759	Phonological Awareness	0.452	0.858
Phonemic Decoding	0.681	Phonemic Decoding	-0.175	0.897

Variables highlighted in gray denote lexical factors. Variables highlighted in white denote sublexical factors.

Due to the differences in variance explained in the two PCAs, we then ran an additional analysis with LS that forced a two factor solution. The two components together accounted for 73.08% of the variance. Interestingly, Word ID (.935), phonological awareness (.750) and phonemic decoding (.728) loaded more heavily and together on the first component (Table 5). This is in contrast to the HS, where Word ID and the two phonological measures strongly loaded on separate components, with phonemic decoding specifically only loading on the second component, suggesting that it is less correlated with Word ID in HS (Fig 3). A follow-up analysis revealed that phonemic decoding was more correlated with Word ID in LS (r = .65, p = .013) than HS (r = .03, p = .84) in separate regressions for each group, although a group x phonemic decoding interaction in a combined model did not explain a significant amount of additional variance (p = .19).

Table 5

Factor Loadings for a forced 2-component model of lexcial representational structure for lower sensitivity readers.

	Component
Lower Sensitivity Readers	1	2
Word Identification	0.411	0.816
Vocabulary	0.827	0.225
Spelling	0.689	0.083
Phonological Awareness	0.714	0.204
Phonemic Decoding	0.071	0.935

Variables highlighted in gray denote lexical factors. Variables highlighted in white denote sublexical factors.

Fig 3

Correlation between word ID and phonemic decoding in low and high sensitivity readers.

Discussion

The present study examined whether differences in holistic orthographic coding, measured by sensitivity to orientation, predict differences in the reading procedures of native English readers. More specifically, orientation sensitivity was hypothesized to occur with a bias towards lexical reading procedures, which should be an indicator of a reading profile that has less reliance on sublexical processing and phonological decoding. While only one effect was highly significant, all effects including marginal effects demonstrated this overall pattern when looking at predictors of overt word naming reaction time and accuracy, as well as exploratory factor analyses examining lexical structural representation. Thus, while individual results must be regarded with caution due to the small sample size and marginal significance, they cohere together as predicted, and suggest that orientation sensitivity can be used as a marker of reading procedures and to unmask reading procedure differences in highly skilled readers.

The results from the overt naming task are consistent with our hypotheses and past studies looking at Chinese-English and Koren-English bilinguals, which found that greater orientation sensitivity covaries with a bias towards lexical reading procedures [11, 12]. Conversely, less orientation sensitivity covaries with a bias towards sublexical reading procedures. The current study expanded the scope of previous studies with a more thorough contrast of lexical vs. sublexical factors using a word list controlled for additional psycholinguistic factors [e.g., imageability, biphone frequency, etc.; 15]. The 3-way interactions (marginally significant) between Group x Presentation Orientation x Biphone Frequency in both reaction time and accuracy supported this hypothesis. LS individuals were more affected by biphone frequency than HS individuals in the reversed orientation condition (i.e., relatively slower RTs and lower accuracy for atypically oriented words with lower biphone frequency). While the marginally significant 3-way interaction with biphone frequency in the reaction time data did not hold in a follow-up analysis, a 3-way interaction with bigram frequency emerged in the same and predicted direction, supporting the same, albeit weak, overall pattern. Overall, this pattern is more similar to Korean-English bilingual performance in past studies and is consistent with a more analytical/sublexical reading procedure.

In addition to HS relying less on a sublexical factor compared to LS, there was some evidence that they also relied more on lexical-level factors. First, there was a marginally significant 3-way interaction between Group x Presentation Orientation x Imageability in reaction time data. Heightened sensitivity to imageability was a predicted outcome for individuals with a bias towards using lexical reading procedures since imageability can only be assessed at the whole-word level. Second, there was a highly significant 3-way interaction between Group x Presentation Orientation x Length in reaction time data, such that HS had relatively longer reaction times for longer words when they were atypically oriented. This pattern was robust to an alternative reanalysis. Further, a power analysis and recent simulations exploring issues of replicability suggest this is unlikely to be a spurious result, even when the small sample size is taken into account [37]. The significant 2-way interaction between presentation orientation and length provides support for the idea that atypical orientation leads to the requirement for a more effortful sublexical approach in word identification, thus leading to longer reading times for longer words [6, 7]. The fact that the length effect for atypically oriented words was larger for HS suggests that they were less efficient at utilizing the sublexical/analytical approach, and thus had relatively longer reaction times than LS for longer words when they were atypically oriented. This pattern is more similar to Chinese-English bilingual performance in past studies and is consistent with a more holistic/lexical reading procedure.

These behavioral patterns observed in HS readers are also consistent with recent fMRI results, examining a subset of the same participants in the current study [25]. Greater orientation sensitivity was associated with bilateral visual word form area (VWFA) engagement in the mid-fusiform gyrus (mFG). This pattern is also consistent with artificial orthography studies that suggest attention to sublexical decoding is linked with left hemisphere dominant mFG activity [38], whereas decoding using larger grain sizes is associated with relatively more bilateral mFG activity [39-41].

The exploratory factor analysis of lexical structural representation, including tasks that emphasize lexical constituent knowledge (i.e., spelling, phonological awareness and vocabulary) and phonemic decoding, revealed a single component for LS readers. In contrast, for HS readers, the measures loaded on two components, with spelling and vocabulary loading more on a first component with word identification, and phonemic decoding loading more on a second component. Phonological awareness was more balanced than the other factors, but had a stronger loading on the second component, with phonemic decoding. These results suggest that the two groups’ lexical representations may be structured differently (see Tables 4 & 5). When the models were run without restrictions, LS readers have a more cohesive structure of all factors, with all measures weighing on a single component. In contrast, for HS readers, phonological measures, especially phonemic decoding, weigh more heavily on a second component, separate from a measure of word identification, spelling, and vocabulary. In contrast, when the LS model was forced to have two components, phonological measures and word identification loaded heavily on the same component.

These exploratory factor analysis results are intriguing because the structure of lexical representations has been linked with overall reading skill. The Lexical Quality Hypothesis (Perfetti & Hart, 2001; 2002) proposes that the quality of the lexical representation can be measured by the degree to which all factors are highly redundant or correlated, leading to specific, coherent, and reliable word identification. The constituents of skilled readers tend to load on one or two components in a factor analysis, while less skilled readers’ constituents load on more components, indicating a less cohesive representation. What is interesting about the current exploratory results is that the coherence of the two groups’ lexical structural representations differed even though the groups were matched along measures of reading skill. Thus, the results suggest that less coherent structure in one’s lexical representation does not necessarily result in less skilled reading, in turn raising questions about whether more than one profile of lexical integration can support skilled reading.

Taken together, the HS profile is similar to previously observed differences amongst small subpopulations of English readers. There are documented subgroups of readers that similarly have shown high levels of comprehension with a weaker link between comprehension and phonological decoding: resilient readers [42, 43] and deaf native signers [44-47]. However, these subgroups differ from our HS group in that they have lower levels of phonological decoding, leading to the inference that these groups use lexical procedures to ‘compensate for’ their poor phonology skill and thereby achieve high levels of literacy. However, one could argue that our HS readers (and previously reported Chinese-English readers) did not need to compensate for poor phonological processing, since they have normal ranges of phonological skills. Future research is required to assess whether orientation sensitivity would distinguish skilled resilient readers and deaf native signers from appropriate control groups matched for component skills (e.g., phonological decoding, etc.), but with poor comprehension.

In future work, it will be important to determine whether the results observed in this study generalize to other samples and other orientation manipulations. There is ongoing debate, especially in the face processing literature [48], about which tasks encompass the construct of ‘holistic’ processing. A general consensus is that various tasks (e.g., inversion, composite task [10, 49]) have some unique variance, highlighting the importance of using a combination of measures, including a composite task. An open question also remains as to what leads to variation in orientation sensitivity in skilled English readers. Based on our initial sample, higher sensitivity to atypical orientation (i.e., a ratio of above 1.5 on the lexical decision task) is fairly common, although more research is needed to assess how prevalent it is in a wider population and how that relates to reading procedures. It is possible that we are simply observing a natural variation in cognitive biases present in the population. Alternatively, some variation could result from differences in foundational instructional methods. There is evidence for differences in some reading procedures (e.g., nonword decoding) in children who were taught with a phonics approach vs. a story/text-centered approach that focuses on context cues and analogies rather than sounding words out [8, 50]. Future research is needed to determine whether the ability to manipulate reading biases through instruction could benefit individuals with poor phonological decoding who did not naturally rely on a reading procedure that best fits their cognitive abilities.

(CSV)

Click here for additional data file.

(CSV)

Click here for additional data file.

25 Sep 2019

PONE-D-19-17391

Unmasking individual differences in adult reading procedures by disrupting holistic orthographic perception

PLOS ONE

Dear Dr. Hirshorn,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

I have sent your manuscript to three external reviewers. You will see below that there were several concerns raised in the course of the review that led to the decision that the research article is not acceptable in its current form. Reviewers raised some concerns about the small sample size and about the experimental design and procedure. I therefore invite you to revise and resubmit your manuscript, taking into account the points raised by the reviewers. Please also consider the other comments of the reviewers as you prepare your revision.

We would appreciate receiving your revised manuscript by Nov 09 2019 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Yafit Gabay

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: Partly

Reviewer #3: Partly

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: I Don't Know

Reviewer #2: Yes

Reviewer #3: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: OVERALL IMPRESSION:

My overall impression is somewhat positive re. the study overall but I am worried about the possibility of low power. I think that the main idea is interesting, and it seems that the study was well executed. Sample size is small (N = 14 and N = 11) which made me question the replicability of certain outcomes, both positive and negative. In any case, a small sample makes it particularly important IMO to show the underlying data points explicitly and graphically in the manuscript.

COMMENTS AND SUGGESTIONS:

-- Introduction --

“An atypical orientation is thought to disproportionately affect holistically processed objects because the holistic process cannot be applied when the object is presented unconventionally.” While the former part is widely accepted, I wanted to point the authors to this counterargument to the latter part:

Richler, J. J., Mack, M. L., Palmeri, T. J., & Gauthier, I. (2011). Inverted faces are (eventually) processed holistically. Vision Research, 51(3), 333-342.

“Chinese is a morpho-syllabic writing system, and so sublexical orthographic coding and mapping to phonology is less useful for word identification, as compared to the holistic coding and mapping of characters to their morphemic forms.” Whether or not expert Chinese readers rely on holistic processing of Chinese characters might depend on their writing experience: “Compared with Chinese nonreaders, Chinese readers who had limited writing experience showed increased holistic processing, whereas Chinese readers who could write characters fluently showed reduced holistic processing.”

Tso, R. V. Y., Au, T. K. F., & Hsiao, J. H. W. (2014). Perceptual expertise: can sensorimotor experience change holistic processing and left-side bias? Psychological Science, 25(9), 1757-1767.

-- Materials and Methods –

---- Lexical decision and Overt word naming----

As already mentioned, the sample size was quite small. The initial 2AFC (I am assuming, not much info is given – I suggest giving more detail about tasks and stimuli, randomization/blocking etc.) lexical decision screening task also only included 40 trials, which might lead to unreliable RT estimates and therefore unreliable group membership assignment.

I was therefore happy to see that the author reassessed orientation sensitivity. This manipulation is however different from the one used in lexical decision. In lexical decision, parts and wholes were both inverted (akin to the manipulation for the face inversion effect), while in overt word naming, the whole was disturbed while parts (i.e. letters) were kept in their original orientation. The authors however provide no explanation of why they used this other way of experimentally defining orientation sensitivity.

-- Results –

---- Table 1 ----

I suggest including Cohen’s d effect size estimates. Provide units (e.g. ms). I think that at least some of the stats might be wrong. E.g. I calculated Comprehension using https://www.graphpad.com/quickcalcs/ttest2/ I came up with a similar p-value (0.10 instead of 0.09) but a completely different t-value (1.73 compared to 0.51). More generally, please say explicitly in the table heading whether you are presenting regular t-tests or something else in that table.

---- Table 2, table 3, and figure 2 ----

I suggest clearly marking factors as lexical and sublexical as this was a major part of your hypothesis.

In order to interpret certain factors in the models, the reader needs to know how they were coded. Was e.g. LS 0 and HS 1? Same with others, e.g. presentation orientation.

What would help even more with interpretation of these, frankly, quite complicated models with several factors and multiple interactions, is if you would actually present the underlying data graphically. This is done in figure 2 for a few of the variables, but I strongly suggest that you show correlations in a different way that highlights the underlying data points more explicitly (e.g. scatterplots, correlograms: https://www.r-graph-gallery.com/correlogram/). This makes it much easier to compare the groups, and for scatterplots can simultaneously show overall group differences (group main effects) as well as interactions (e.g. to show that the association between orientation and RT is different for the groups). I also suggest that this should not only be shown for the significant interactions, as you expected other differences as well that did not come out as significantly different between the two groups.

---- Principal components analysis ----

The authors rightfully point out that the groups are small for an exploratory factor analysis. I have seen estimates of at least 50 participants, or of at least 10 times the number of participants as there are variables. I found one paper that claims that exploratory factor analysis can be done for very small sample sizes under certain circumstances:

de Winter*, J. D., Dodou*, D. I. M. I. T. R. A., & Wieringa, P. A. (2009). Exploratory factor analysis with small sample sizes. Multivariate behavioral research, 44(2), 147-181.

However, I think that given the small sample sizes, any differences in factor structure might be uninterpretable. If the authors want to show possible group differences as an exploratory or descriptive analysis, I again suggest that showing two correlograms, one per group, would be more likely to give the reader a gist of what might be going on.

Reviewer #2: This paper presents the results of a study that examined the existence of different reading styles in English. The majority of studies have been done on reading in English, and the consensual model, the dual route model, holds that written words can be identified via the lexical, or holistic/orthographic route, or a sub-lexical, phonological decoding route. A number of cross language/writing system studies have shown that the orthographic transparency and morphological structure of specific writing systems result in more efficient processing weighting the lexical or the sub-lexical route. In English, it has been shown that both characteristics of the words and of the reader (e.g., skill level) affect the relative use of the lexical or sublexical routes. I have several major concerns:

1. I am not exactly sure of the goal of this paper: participants were divided into a ‘holistic’ or ‘sub-lexical’ category by the degree to which reading words upside-down differed from reading words in the canonical orientation. Then, after a very extensive sifting procedure, these groups were compared on reading words forwards and backwards. I understand that the point was to see if performance on this task, using words for which the holistic and sub-lexical aspects are documented, would go in the same direction. Here lies the problem – the fact that the groups differed on the backwards reading task is trivial, as they were chosen to differ on an upside-down reading task. The fact that they only differ on performance on the upside-down words is problematic, as it suggests that in normal word identification tasks, they do not differ. Thus, when the task is made more difficult in a specific manner, then the groups differ in how well they can compensate for the spatial distortion. But that is how the groups were created to begin with – there is no theoretical reason to think that the two spatial manipulations (upside down and backwards) are inherently different.

2. The finding that the factor structure of the RTs is different in a predictable manner could be even more supporting evidence that this division is reasonable, but the very small number of participants in the groups (11 and 14) really makes it hard to believe it…-- and, it is also true by the way that the groups were defined.

3. There seem to be some missed opportunities here. Although it is not mentioned, the division of word characteristics into lexical and sub-lexical categories follows results shown by many studies of reading by split-brain patients, reading by unilateral brain damaged patients, and healthy participants identifying words in divided visual field paradigms. This has been done across languages (e.g., Rao & Vaid, 2017; Zhou et al, 2019; Ibrahim & Eviatar, 2012; and many more). Given the very large initial sample, it would have been interesting to see if left handed participants tended to fall into one or the other group. That would have been a novel result.

Summary: Other than the very small sample size, the study seems well done. I am just not sure what it adds to the literature.

Reviewer #3: The investigators have concluded that their results revealed that greater orientation sensitivity was associated with a reading profile that relies less on sublexical phonological measures and more on lexical-level characteristics within the skilled English readers. This is based on a statistical approach involving typical procedures such as the t-test, mixed models, the empirical logit model and principal component analysis. There are some concerns. Specifically:

1. The authors note that participants were recruited from a database of 411 individuals interested in study opportunities. It is not clear what ‘study opportunities’ means here or if it is even relevant. They explain that a number of subjects were removed leaving 203 from an initial 411 potential participants. This appears to be a convenience sample. A hypothesis is stated on page 7 in the form that native English readers who show more sensitivity to atypical orientation do so because they make greater use of holistic orthographic coding, which should be reflected in a distinctive reading profile. This appears to be a comparison of the low to high sensitivity group which they define. The problem is that there is no statistical design motivation for the sample size. Is 203 individuals statistically sufficient to test their hypothesis with reasonable statistical power? Also on the bottom of Page 8 the investigators note that participant attrition, due to graduation since participation in initial screening and eligibility requirements for a parallel imaging study (beyond the scope of the current study), also reduced the potential participant pool. What exactly does this mean and what was the final number of participants for each of the Tables 1 to 5?

2. As minor points please define RT on page 8. It appears to be 'reaction time'. On the top line of page 20 the term ‘ phonological decoding (.728)’ should be ‘Phonemic Decoding (0.728)’ as per Table 5.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

Reviewer #3: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

21 Nov 2019

Dear Dr. Yafit Gabay and Reviewers,

We very much appreciate the helpful comments and believe they have greatly improved the paper. Responses to individual comments are below and also uploaded as "ResponsetoReviewers_Final.docx".

Sincerely,

Elizabeth Hirshorn, PhD

Reviewer #1: OVERALL IMPRESSION:

My overall impression is somewhat positive re. the study overall but I am worried about the possibility of low power. I think that the main idea is interesting, and it seems that the study was well executed. Sample size is small (N = 14 and N = 11) which made me question the replicability of certain outcomes, both positive and negative. In any case, a small sample makes it particularly important IMO to show the underlying data points explicitly and graphically in the manuscript.

We appreciate this point and have included figures for the most critical results of the paper (see new Fig 2 and 3). We understand the concern of small sample size and have made a point to be transparent in describing strong results that are statistically unlikely due to chance, and versus those that are more tentative. We have also added text to our discussion section (pgs. 23 and 27) that explicitly notes our relatively low sample size and that recommends future work to test whether the results generalize to other groups.

COMMENTS AND SUGGESTIONS:

-- Introduction --

“An atypical orientation is thought to disproportionately affect holistically processed objects because the holistic process cannot be applied when the object is presented unconventionally.” While the former part is widely accepted, I wanted to point the authors to this counterargument to the latter part:

Richler, J. J., Mack, M. L., Palmeri, T. J., & Gauthier, I. (2011). Inverted faces are (eventually) processed holistically. Vision Research, 51(3), 333-342.

Thank you for this point. We have added this reference to the introduction at the bottom of page 3 of the manuscript. We would argue that even if perception is ultimately holistic, inversion sensitivity could still be used as a marker of reading procedures.

“Chinese is a morpho-syllabic writing system, and so sublexical orthographic coding and mapping to phonology is less useful for word identification, as compared to the holistic coding and mapping of characters to their morphemic forms.” Whether or not expert Chinese readers rely on holistic processing of Chinese characters might depend on their writing experience: “Compared with Chinese nonreaders, Chinese readers who had limited writing experience showed increased holistic processing, whereas Chinese readers who could write characters fluently showed reduced holistic processing.”

Tso, R. V. Y., Au, T. K. F., & Hsiao, J. H. W. (2014). Perceptual expertise: can sensorimotor experience change holistic processing and left-side bias? Psychological Science, 25(9), 1757-1767.

This point is well taken. We have added this reference and made the point that the degree of holistic processing varies in Chinese readers, although Chinese readers may still on average rely more on holistic processing than alphabetic readers (see pg. 4-5 of manuscript).

-- Materials and Methods –

---- Lexical decision and Overt word naming----

As already mentioned, the sample size was quite small. The initial 2AFC (I am assuming, not much info is given – I suggest giving more detail about tasks and stimuli, randomization/blocking etc.) lexical decision screening task also only included 40 trials, which might lead to unreliable RT estimates and therefore unreliable group membership assignment.

This is correct. The initial lexical decision task was part of a larger screening battery, and therefore was meant to be very short. (In subsequent research, we have increased the number of trials.) More detail has been added to the methods on page 10 of the manuscript.

I was therefore happy to see that the author reassessed orientation sensitivity. This manipulation is however different from the one used in lexical decision. In lexical decision, parts and wholes were both inverted (akin to the manipulation for the face inversion effect), while in overt word naming, the whole was disturbed while parts (i.e. letters) were kept in their original orientation. The authors however provide no explanation of why they used this other way of experimentally defining orientation sensitivity.

Thank you for this comment, as we believe this information is important to convey clearly. By not using the same manipulation, we were able to generalize the orientation sensitivity of our groups. Other studies have used alternating font, etc., but we chose the reversed presentation because it was a larger distortion of the whole, as you point out, and would be pushing our hypotheses further. This is now communicated in the text on pg. 10.

-- Results –

---- Table 1 ----

I suggest including Cohen’s d effect size estimates. Provide units (e.g. ms). I think that at least some of the stats might be wrong. E.g. I calculated Comprehension using https://www.graphpad.com/quickcalcs/ttest2/ I came up with a similar p-value (0.10 instead of 0.09) but a completely different t-value (1.73 compared to 0.51). More generally, please say explicitly in the table heading whether you are presenting regular t-tests or something else in that table.

Thank you for this comment. Cohen’s d values have been added to Table 1 on pg. 14. Thank you for catching that typo for the t-value for the comprehension measure. We see now that it didn’t make much sense. Yes, I believe the p-value was just a rounding error, as it was .096. It is now reported as .10.

---- Table 2, table 3, and figure 2 ----

I suggest clearly marking factors as lexical and sublexical as this was a major part of your hypothesis.

Thank you for this helpful suggestion. It makes a lot of sense for clarity. In the tables (2 & 3), the effects involving lexical factors are now highlighted in light gray, and the sublexical factors have a white background. This is explained in the table footnotes.

In order to interpret certain factors in the models, the reader needs to know how they were coded. Was e.g. LS 0 and HS 1? Same with others, e.g. presentation orientation.

Point taken. Yes, LS was coded as 0, and upright/typically presented words were coded as 0. This is now explicitly noted in the text on pg. 14.

What would help even more with interpretation of these, frankly, quite complicated models with several factors and multiple interactions, is if you would actually present the underlying data graphically. This is done in figure 2 for a few of the variables, but I strongly suggest that you show correlations in a different way that highlights the underlying data points more explicitly (e.g. scatterplots, correlograms: https://www.r-graph-gallery.com/correlogram/). This makes it much easier to compare the groups, and for scatterplots can simultaneously show overall group differences (group main effects) as well as interactions (e.g. to show that the association between orientation and RT is different for the groups). I also suggest that this should not only be shown for the significant interactions, as you expected other differences as well that did not come out as significantly different between the two groups.

Thank you for this suggestion. We believe it will aid in the transparency in understanding our results. We have included scatterplot figures for the 3-way interactions of the RT data, since that is where our hypotheses were more relevant. We presented the 3-way interactions as two scatterplots, one for typical and one for atypical (reversed) stimuli, with the x-axis representing the psycholinguistic factor (e.g. frequency) and the different colored dots/lines representing the groups. We included these for all psycholinguistic factors, even non-significant ones (see pg. 18).

---- Principal components analysis ----

The authors rightfully point out that the groups are small for an exploratory factor analysis. I have seen estimates of at least 50 participants, or of at least 10 times the number of participants as there are variables. I found one paper that claims that exploratory factor analysis can be done for very small sample sizes under certain circumstances:

de Winter*, J. D., Dodou*, D. I. M. I. T. R. A., & Wieringa, P. A. (2009). Exploratory factor analysis with small sample sizes. Multivariate behavioral research, 44(2), 147-181.

Thank you for this. While recognizing the small sample size, we also found this paper helpful and had cited it in the original manuscript (under the Lexical representational structure sub-header on pg. 20, now reference 34).

However, I think that given the small sample sizes, any differences in factor structure might be uninterpretable. If the authors want to show possible group differences as an exploratory or descriptive analysis, I again suggest that showing two correlograms, one per group, would be more likely to give the reader a gist of what might be going on.

We apologize. We weren’t quite sure what you were imagining here and are less familiar with correlograms, per se. Since the largest difference in the structure between the two groups was the loading of the phonemic decoding measure, we included a scatterplot of how that correlates with Word ID in both groups. This is now Figure 3. We hope this will highlight the gist of the take-home message here in a clearer manner.

Reviewer #2: This paper presents the results of a study that examined the existence of different reading styles in English. The majority of studies have been done on reading in English, and the consensual model, the dual route model, holds that written words can be identified via the lexical, or holistic/orthographic route, or a sub-lexical, phonological decoding route. A number of cross language/writing system studies have shown that the orthographic transparency and morphological structure of specific writing systems result in more efficient processing weighting the lexical or the sub-lexical route. In English, it has been shown that both characteristics of the words and of the reader (e.g., skill level) affect the relative use of the lexical or sublexical routes. I have several major concerns:

1. I am not exactly sure of the goal of this paper: participants were divided into a ‘holistic’ or ‘sub-lexical’ category by the degree to which reading words upside-down differed from reading words in the canonical orientation. Then, after a very extensive sifting procedure, these groups were compared on reading words forwards and backwards. I understand that the point was to see if performance on this task, using words for which the holistic and sub-lexical aspects are documented, would go in the same direction. Here lies the problem – the fact that the groups differed on the backwards reading task is trivial, as they were chosen to differ on an upside-down reading task.

We actually agree here. We reported the results that show group differences in the backwards reading task for the sake of completeness of the model.

The fact that they only differ on performance on the upside-down words is problematic, as it suggests that in normal word identification tasks, they do not differ. Thus, when the task is made more difficult in a specific manner, then the groups differ in how well they can compensate for the spatial distortion. But that is how the groups were created to begin with – there is no theoretical reason to think that the two spatial manipulations (upside down and backwards) are inherently different.

Here we respectfully disagree. It is true that the groups only seem to differ when the word presentation is disrupted, but that was actually hypothesized. We interpret these data patterns as suggesting that while both groups are skilled readers, they are actually using different underlying procedures for word identification. Since both groups’ procedures are efficient for them, we wouldn’t necessarily expect differences in a typical word presentation. We predicted that differences would likely emerge when we made the task more difficult, as you stated. However, it isn’t the fact that one group is worse or slower when the words were distorted that we hoped to focus on, because as you stated that was part of the group definitions. We hoped to highlight the fact that different psycholinguistic factors were better predictors of reading speed in the distorted trials. Indeed, we found that high sensitivity readers were significantly more affected by length (and marginally by imageability), both considered lexical factors, when words were distorted, but low sensitivity readers were marginally more affected by biphone frequency, considered to be a sublexical factor. While some of the results are weaker, we hoped to highlight that all observed effects are in the predicted direction, as you state below.

To your other point regarding there being no theoretical reason that the two manipulations are inherently different, we actually agree. We wanted to use two different manipulations to strengthen the argument that different manipulations should similarly disrupt reading ability and lessen a potential concern that these effects are specific to one manipulation. We highlight this point on pg. 10.

2. The finding that the factor structure of the RTs is different in a predictable manner could be even more supporting evidence that this division is reasonable, but the very small number of participants in the groups (11 and 14) really makes it hard to believe it…-- and, it is also true by the way that the groups were defined.

We hope the response to the above comment will address some of these concerns.

Lastly, while indeed some effects are quite small, but in the predicted directions, the 3-way (group x presentation orientation x length) interaction is highly significant (p=.000102), which is very unlikely to be due to chance, and added a reference to support that point on pg. 24 (Ioannidis JP. The proposal to lower P value thresholds to. 005. Jama. 2018;319(14):1429-30). We do not believe this effect is directly linked with how the groups were defined, as our criteria for inversion sensitivity did not include length as a factor. But we do agree with your concern about the strength of the other effects. We have tried to be more transparent in the text regarding this issue, both in how we describe the effects, and on page 23 of the discussion, where we explicitly note that some of our results are marginal.

3. There seem to be some missed opportunities here. Although it is not mentioned, the division of word characteristics into lexical and sub-lexical categories follows results shown by many studies of reading by split-brain patients, reading by unilateral brain damaged patients, and healthy participants identifying words in divided visual field paradigms. This has been done across languages (e.g., Rao & Vaid, 2017; Zhou et al, 2019; Ibrahim & Eviatar, 2012; and many more). Given the very large initial sample, it would have been interesting to see if left-handed participants tended to fall into one or the other group. That would have been a novel result.

We did not include left-handed subjects in our sample, and so cannot pursue this suggestion. However, we appreciate this thoughtful comment and would be interesting in pursuing these ideas in future work.

Summary: Other than the very small sample size, the study seems well done. I am just not sure what it adds to the literature.

Reviewer #3: The investigators have concluded that their results revealed that greater orientation sensitivity was associated with a reading profile that relies less on sublexical phonological measures and more on lexical-level characteristics within the skilled English readers. This is based on a statistical approach involving typical procedures such as the t-test, mixed models, the empirical logit model and principal component analysis. There are some concerns. Specifically:

1. The authors note that participants were recruited from a database of 411 individuals interested in study opportunities. It is not clear what ‘study opportunities’ means here or if it is even relevant.

We apologize for any unwanted ambiguity. The Perfetti Lab recruits several hundred participants each year for a behavioral screening in order to drive several more specific research studies where participants are needed with specific behavioral characteristics. We were conducting one of those many studies.

They explain that a number of subjects were removed leaving 203 from an initial 411 potential participants. This appears to be a convenience sample.

We first filtered based on overall reading ability, in the hopes to avoid testing participants at the extremes of the distribution. We’re not sure this would necessarily be considered convenience sampling, which to our understanding would be recruiting only participants that were available or nearby.

A hypothesis is stated on page 7 in the form that native English readers who show more sensitivity to atypical orientation do so because they make greater use of holistic orthographic coding, which should be reflected in a distinctive reading profile. This appears to be a comparison of the low to high sensitivity group which they define. The problem is that there is no statistical design motivation for the sample size. Is 203 individuals statistically sufficient to test their hypothesis with reasonable statistical power?

Also on the bottom of Page 8 the investigators note that participant attrition, due to graduation since participation in initial screening and eligibility requirements for a parallel imaging study (beyond the scope of the current study), also reduced the potential participant pool. What exactly does this mean and what was the final number of participants for each of the Tables 1 to 5?

That is correct. There were several extenuating circumstances that lead to a smaller sample including that we wanted to first recruit participants with nonoverlapping inversion sensitivity scores (described and highlighted on page 9), and that they also be eligible for a companion imaging study. Along with attrition from the original database due to graduation or simply not responding, the pool that we drew from was significantly smaller. We believe future research will be important to understand how common these behavioral profiles are.

The final number of participants for all tables was 14 for LS readers and 11 for HS readers.

2. As minor points please define RT on page 8. It appears to be 'reaction time'. On the top line of page 20 the term ‘ phonological decoding (.728)’ should be ‘Phonemic Decoding (0.728)’ as per Table 5.

Thank you- these corrections have been made.

Submitted filename: ResponseToReviewers_Final2.docx

Click here for additional data file.

24 Jan 2020

PONE-D-19-17391R1

Unmasking individual differences in adult reading procedures by disrupting holistic orthographic perception

PLOS ONE

Dear Dr. Hirshorn,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

==============================

I sent your revised manuscript to one of the original reviewers and to one new reviewer as the other reviewers were no longer available. As you will see below both reviewers commented that the manuscript has been improved and  suggested minor changes to the current version. I therefore invite you to address these comments before we will able able to move forward and accept your paper for publication.

==============================

We would appreciate receiving your revised manuscript by Mar 09 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Yafit Gabay

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: (No Response)

Reviewer #4: (No Response)

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #4: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #4: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #4: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #4: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: Thank you, my comments are below:

Regarding the comment of reviewer #2: „...the fact that the groups differed on the backwards reading task is trivial, as they were chosen to differ on an upside-down reading task.“ I started looking into this, and what I missed in my first review of the paper is that the authors apparently not only assessed the orientation sensitivity for the second time, but they also redefined their groups based on this dependent measure: „Six participants whose overt naming scores were neither above nor below the median orientation sensitivity that was consistent with their initial group assignment based on the lexical decision task were removed.“ I don‘t think that this is a good idea, if I understand correctly what the authors did, as this can be considered double dipping into the data. As this was done, I don‘t think that the main effects of orientation or interactions with orientation should be trusted, although I suggest getting a second opinion from a statistician. I suggest rerunning all models with these participants included. Otherwise, group assignments and orientation effects are not independently assessed.

What is going on in the new figure 2 in terms of RTs of atypical word presentations? Almost all of the RTs for the low sensitivity (red) seem to fall under the red regression line while almost all of the RTs for the high sensitivity (cyan) seem to fall under the cyan regression line. I suspect that this has something to do with the actual drawing of the graph, i.e. perhaps almost all of the red dots on the top are underneath the cyan dots on the top. This could likely be amended by making them partially transparent.

Also, the x-axis of figure 2 just says value.z. I suggest changing this to something a bit more transparent, and saying in the figure legend that the x-axis represents each psycholinguistic factor.

Finally, for that graph, please say explicitly that these are (if they are) RTs from individual trials pooled across participants.

The statistics in Table 1 still look weird. The t-value(s) and p-value(s) still don‘t match. I am assuming an N of 14 LS and 11 HS.

Reviewer #4: The authors show that greater sensitivity to orientation/greater holistic processing is associated more with lexical than sub-lexical characteristics, thus holistic strategies seem to be involved with lexical reading strategies.

I was not one of the three original reviewers. My fellow colleagues did an excellent job in their revisions and I also consider the response of the authors adequate. I have only a few points that the authors might additional take into consideration.

(i) One question is the very small number of participants. I wonder if the authors could provide a power analysis study using e.g., g*power.

(ii) An indicator of holistic word processing—observers’ sensitivity to changes in configural/spatial jittering information of objects in an inversion paradigm has been used by Wong et al. (2019). Is this manipulation closer to what is meant by holistic processing? Just inverting the stimuli maybe not be enough to study holistic processing.

(iii) Sensitivity to atypical orientation in the overt naming task used half of the stimuli presented in typical orientation and half presented in a reversed (FLIGHT � THGILF) orientation. What has this to do with holistic strategy

(iv) Authors hypothesize that the HS group should be more affected than the LS group by lexical-level factors when words are atypically oriented, whereas LS should be more affected than HS by sublexical-level factors when words are atypically oriented. But is it equally possible to make the same type of predictions for when words are typically oriented and lexical factors more at play?

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #4: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

14 Apr 2020

This information is also in the Response to Reviewers document that has been uploaded with better visuals and formatting:

Dear Dr. Yafit Gabay and Reviewers,

First, we apologize for the delay in response as we adjusted to impending and current changes in daily life. We very much appreciate the opportunity to address the questions from past and new reviewers and believe they have greatly improved the paper again. Responses to individual comments are below in blue italics. Many thanks to all, and we hope everyone is safe and healthy.

Sincerely,

Elizabeth Hirshorn, PhD

Reviewer #1:

1. Regarding the comment of reviewer #2: „...the fact that the groups differed on the backwards reading task is trivial, as they were chosen to differ on an upside-down reading task.“ I started looking into this, and what I missed in my first review of the paper is that the authors apparently not only assessed the orientation sensitivity for the second time, but they also redefined their groups based on this dependent measure: „Six participants whose overt naming scores were neither above nor below the median orientation sensitivity that was consistent with their initial group assignment based on the lexical decision task were removed.“ I don‘t think that this is a good idea, if I understand correctly what the authors did, as this can be considered double dipping into the data. As this was done, I don‘t think that the main effects of orientation or interactions with orientation should be trusted, although I suggest getting a second opinion from a statistician. I suggest rerunning all models with these participants included. Otherwise, group assignments and orientation effects are not independently assessed.

Thank you for this comment and the chance to consider this concern. We understand the concern of ‘double dipping,’ and agree that the 2-way interaction between group and orientation is a trivial finding, since that is essentially how the groups were defined. We consulted with a statistician, who confirmed that ‘double-dipping’ should not affect the direction or strength of the 3-way interactions, since there is not anything inherent about being sensitive to inversion, per se, that would cause one group to be more influenced than another by a psycholinguistic measure when words were presented atypically. In fact, that is what we were interested in testing. Furthermore, words that were presented in the typical and atypical orientations were matched for all psycholinguistic factors (e.g., there weren’t more long words that were atypical than typical).

Nevertheless, we did take your advice and added in the previously removed participants and ran the model again with “group” as a continuous variable based solely on the lexical decision task (LDT), which eliminates the double-dipping component of group selection since it doesn’t incorporate the naming task at all. The 3-way interaction between “group” (i.e., strength of inversion sensitivity in LDT task) x presentation orientation x length was actually even more significant with this approach (see Table below- yellow highlighted row).

Specifically, the previous t-value for that 3-way interaction was 3.89, and adding the other participants led to a t-value of 6.12. We believe this provides further support that the group/sensitivity x orientation x length 3-way interaction is robust. The two marginal 3-way interactions did not hold up (see response about power for Reviewer #4 below), but another marginal 3-way interaction emerged in the predicted direction, such that bigram frequency was a marginally better predictor of RT for atypically presented words in those whose LDT were relatively small (i.e., those with low sensitivity). We previously discussed our marginal effects in a tentative manner, so this change in marginal results does not undermine our results. Instead, we believe that the fact that a conceptually related marginal effect emerges supports the same weak overall pattern of psycholinguistic differences between the two groups.

Despite the strengthening of the group x orientation x length effect in the updated model, we are proposing to keep the main reporting of results the same, as a group design, but propose to add the results of the additional analysis for transparency (see pg. 17-18). We propose to keep the main focus on the original group-based analysis for the following reasons:

• Our a-priori hypotheses and design were meant for having separate groups.

• Related to reviewer #4’s concern that inverting words alone may not be enough to measure ‘holistic processing,’ we believe that using two similar, but not identical measures actually strengthens different aspects of the study.

o For example, based on logistics and timing, the LDT experiment did not have a lot of trials and we wanted to reduce any potential noise by combining it with an additional task that was different, but conceptually similar in order to have more stability in our measurement.

o We recently published a paper (Carlos et al, 2019) that used this same grouping. In this paper the same Ss were used to look at group differences in a neural measure of lateralization, with more inversion sensitive participants exhibiting a more bilateral pattern of activation in a putative visual word form area. This result provides converging evidence that our groups exhibit different patterns of orthographic processing, with no concerns about “double-dipping” into the same data.

o Since these two papers are in essence companion pieces, we believe it is more straightforward to be able to talk about the same participants.

That being said, our current and future plans for the continuation of this research will be using more continuous measures of sensitivity.

What is going on in the new figure 2 in terms of RTs of atypical word presentations? Almost all of the RTs for the low sensitivity (red) seem to fall under the red regression line while almost all of the RTs for the high sensitivity (cyan) seem to fall under the cyan regression line. I suspect that this has something to do with the actual drawing of the graph, i.e. perhaps almost all of the red dots on the top are underneath the cyan dots on the top. This could likely be amended by making them partially transparent.

Thank you for sharing this observation. The pattern that you noticed could be partially a consequence of the group difference in RT for atypically presented words (low sensitivity readers are faster than high sensitivity readers by design), but we understand what you’re saying. At your suggestion, we did make a version with more transparent markers, but unfortunately it just ended up looking extremely blurry and harder to interpret. For the sake of perceptual clarity, we think the information is better represented with more opaque markers, albeit less than ideal. We believe that the main take-away message from this figure, to see the non-parallel lines in the atypical presentation for length (plus imageability & biphone frequency) should still be able to be communicated.

2. Also, the x-axis of figure 2 just says value.z. I suggest changing this to something a bit more transparent, and saying in the figure legend that the x-axis represents each psycholinguistic factor.

Thank you for this suggestion. This was changed, so that the x-axis is now titled, ‘z-score of Each Psycholinguistic Factor,’ thank you.

3. Finally, for that graph, please say explicitly that these are (if they are) RTs from individual trials pooled across participants.

Thank you for this suggestion. This information was added to the figure caption.

4. The statistics in Table 1 still look weird. The t-value(s) and p-value(s) still don‘t match. I am assuming an N of 14 LS and 11 HS.

Thank you for pointing this out. We regret that the previous version did indeed have errors in Table 1. The small edits needed to correct these errors have been made. Importantly, the results are essentially the same.

Reviewer #4

The authors show that greater sensitivity to orientation/greater holistic processing is associated more with lexical than sub-lexical characteristics, thus holistic strategies seem to be involved with lexical reading strategies. I was not one of the three original reviewers. My fellow colleagues did an excellent job in their revisions and I also consider the response of the authors adequate. I have only a few points that the authors might additional take into consideration.

1. One question is the very small number of participants. I wonder if the authors could provide a power analysis study using e.g., g*power.

We understand the concern about a small sample. We consulted with a statistician, and instead of using g*power, we used a function in R that is suited to linear mixed models (see powerSim in simr package). This allows us to estimate the power for specific effects in our model using Monte Carlo estimation:

Green, P., & MacLeod, C. J. (2016). SIMR: an R package for power analysis of generalized linear mixed models by simulation. Methods in Ecology and Evolution, 7(4), 493-498.

Brysbaert, M., & Stevens, M. (2018). Power analysis and effect size in mixed effects models: A tutorial. Journal of Cognition, 1(1).

We ran this for each of the three 3-way interactions that showed effects, set at 200 simulations. The 3-way interaction between group x orientation x length had a very large effect size, and the observed power was 96.50% (95% confidence interval: 92.92, 98.58). Not surprisingly, the two marginal effects had much lower power- both had 51% power (43.85, 58.12). We hope that talking about these effects as more tentative will be adequate.

We have modified the manuscript to include this power analysis in the results section (pg. 17), and have retained our previous caution in presenting and discussing our marginal effects.

2. An indicator of holistic word processing—observers’ sensitivity to changes in configural/spatial jittering information of objects in an inversion paradigm has been used by Wong et al. (2019). Is this manipulation closer to what is meant by holistic processing? Just inverting the stimuli maybe not be enough to study holistic processing.

Thank you for this comment, as this is an ongoing issue in the field. In fact, since this work has been partially inspired by the face processing literature, we share the concern that there is not necessarily one manipulation that completely encompasses the construct of ‘holistic’ processing (see Rezlescu et al, 2017). That is one of the main reasons we chose to combine slightly different measures for group assignment.

Rezlescu, C., Susilo, T., Wilmer, J. B., & Caramazza, A. (2017). The inversion, part-whole, and composite effects reflect distinct perceptual mechanisms with varied relationships to face recognition. Journal of Experimental Psychology: Human Perception and Performance, 43(12), 1961.

That being said, the current and future work in this research program uses a combination of measures, including a composite task. For this paper, we have added a paragraph to our discussion section (pg. 28) that notes this important issue, with citations to Wong et al. (2109) and Rezlescu et al. (2017) included.

3. Sensitivity to atypical orientation in the overt naming task used half of the stimuli presented in typical orientation and half presented in a reversed (FLIGHT à THGILF) orientation. What has this to do with holistic strategy?

Our thinking is that any perceptual manipulation that disrupts the typical processing of an object should also disrupt holistic processing, if that is used. For example, other studies have used case, size, and font alteration to add visual noise and thus disrupt ‘holistic’ processing (see Pae et al, 2017).

Pae, H. K., Kim, S. A., Mano, Q. R., & Kwon, Y. J. (2017). Sublexical and lexical processing of the English orthography among native speakers of Chinese and Korean. Reading and Writing, 30(1), 1-24.

While our reverse presentation is not a widely used manipulation, we consider it to be a conceptual extension of previous manipulations (such as case alternation) – that is, the idea that presenting a word in any atypical orientation should disproportionately disrupt those who rely more on a ‘holistic’ reading procedure.

We note that similar ideas are present in the face and object processing literature. For instance, Perret, Oram, and Ashbridge (1998) propose that the speed of object recognition rests upon the accumulation of activity from neurons selective for the object as experienced in a particular viewing circumstance. This neuronal tuning is presumed to be sensitive to frequency of occurrence, and so the neural activity builds more quickly for objects presented in a canonical as compared to unusual view.

4. Authors hypothesize that the HS group should be more affected than the LS group by lexical-level factors when words are atypically oriented, whereas LS should be more affected than HS by sublexical-level factors when words are atypically oriented. But is it equally possible to make the same type of predictions for when words are typically oriented and lexical factors more at play?

This is a very astute point. The reason we focused on atypically oriented words is that the effects may be smaller and not as easily measured when words are typically oriented, since both groups are skilled and efficient using whatever reading procedures are most natural. In contrast, with an atypical orientation the idea is that all participants are forced to rely more upon sublexical procedures, and so this helps to “unmask” differences in the ability to use these procedures.

However, we do agree that biases for the different procedures remain present when words are presented in their typical orientation. In support of this point, in a previous paper, we compared Chinese- and Korean-English bilinguals:

Ben-Yehudah, G., Hirshorn, E. A., Simcox, T., Perfetti, C. A., & Fiez, J. A. (2019). Chinese-English bilinguals transfer L1 lexical reading procedures and holistic orthographic coding to L2 English. Journal of Neurolinguistics, 50, 136-148.

As we discuss in the introductory section of our manuscript, in this prior work we tested our prediction that Chinese-English bilinguals would use a more holistic reading procedure than Korean-English bilinguals. In support of this hypothesis, we found that the Chinese-English bilinguals exhibited greater sensitivity to inversion than the Korean-English bilinguals.

One finding from this paper that we did not previously discuss concerned differences in the reading skill profile of the two bilingual groups. Even though we matched the groups on the basis of spoken English experience and ability, we found that the Chinese-English bilinguals exhibited performance differences on typically presented items. For instance, they performed more poorly on the Word Attack and Word ID subtests of the Woodcock Reading Mastery Tests, and we found that the Chinese-English bilinguals exhibited greater sensitivity to lexical frequency on a word-naming task that included upright and inverted items. In the discussion section of Ben-Yehudah et al. paper, we discuss the implication of these results, which are similar to those reported in previous studies. Specifically, we suggest that Chinese-English bilinguals have a bias towards a holistic procedure is a less optimal approach for developing reading skill in an alphabetic writing system. Consequently, they require more reading experience to attain the same level of skill.

These same ideas could be relevant for native English speakers to exhibit a bias towards a holistic procedure. However, we would be unlikely to find them in the present study, because we selected the two groups to be matched on measures of reading skill. If we instead could have selected the group to be matched on reading experience or randomly selected them without concern for reading skill, then it is very possible that we would have observed the pattern suggested by the reviewer.

Due to concerns about the overall length of the paper, we have not revised our discussion section to consider this point. However, if the reviewer feels that this is an issue that merits consideration, we would be pleased to add text addressing the reviewer’s comment.

Submitted filename: PLoSOne_SecondRevision_ResponseToReviewers.docx

Click here for additional data file.

28 Apr 2020

Unmasking individual differences in adult reading procedures by disrupting holistic orthographic perception

PONE-D-19-17391R2

Dear Dr. Hirshorn,

We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements.

Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication.

Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at https://www.editorialmanager.com/pone/, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

With kind regards,

Yafit Gabay

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: I Don't Know

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: (No Response)

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

4 May 2020

PONE-D-19-17391R2

Unmasking individual differences in adult reading procedures by disrupting holistic orthographic perception

Dear Dr. Hirshorn:

I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

For any other questions or concerns, please email plosone@plos.org.

Thank you for submitting your work to PLOS ONE.

With kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Yafit Gabay

Academic Editor

PLOS ONE