Reading Left-To-Right and Right-To-Left Orthographies: Ocular Prevalence, Similarities, Differences and the Reasons for Orthographic Conventions

ABSTRACT

Purpose

We compare right-to-left and left-to-right orthographies to test the theory, derived from studying the latter, that small temporal asynchronies between the two eyes at the beginning and end of every fixation favor ocular prevalence for the left eye in the left hemifield and the right eye in the right hemifield. Ocular prevalence is the prioritizing of one eye’s input in the conscious, fused binocular percept.

Method

We analyze binocular eye-tracking data from the reading of multiline Arabic and Hebrew text by 28 Arabic (M = 28.7, SD = 7.2 years, 71% female) and 16 Hebrew (M = 30.1, SD = 7.9 years, 50% female) native speakers, respectively.

Results

Critically, the complex pattern of asynchronies in Arabic and Hebrew resembles that reported for the left-to-right orthographies, English and Chinese, but with some particular differences that we attribute to left hemisphere specialization in word recognition.

Conclusion

We conclude, first, that the oculomotor musculature plays an embodied role in the perception and cognition associated with reading. We further discuss how the evident hemispheric asymmetries in parafoveal lookahead may be reflected in the nature of the conventions of right-to-left scripts. We articulate the claim that the orthographic conventions of a language tend to reflect reading direction and hemispheric differences.

Binocular coordination in reading has been investigated intensively in recent decades (e.g. Bucci et al., 2008; Hooge et al., 2019; Jainta et al., 2010, 2010; Liversedge et al., 2006; Nuthmann & Kliegl, 2009; Nuthmann et al., 2014; Shillcock et al., 2010). Readers are assumed to coordinate their eyes to optimize higher visual processing. How is this optimization achieved? How might it be affected by the characteristics of the language and its orthography (cf. Hsiao et al., 2018; Liversedge et al., 2016)? In this paper, we ask:

1. In reading, can one eye be temporarily more important than the other? When? Why?

2. Does it matter if the direction of reading is left-to-right (English, Chinese) or right-to-left (Arabic, Hebrew)?

3. Why do left-to-right and right-to-left orthographies look different? Have they accommodated to hemispheric preferences?

To address these questions we analyze extensive corpus data from the reading of Arabic and Hebrew. These data are critical in arbitrating interpretations based on our study of English and Chinese (Zhu et al., 2021) – are the behaviors we have reported in those languages due to the peripheral musculature of eye-movements or to higher visual processing?

Ocular prevalence refers to the contribution of one or other eye to higher, conscious visual processing in which binocular fusion has already occurred. There is a subtle and fluid division of labor between the two eyes, in which the two inputs are optimally fused into a single percept, but with one or other of them necessarily having priority at any one time (Kommerell et al., 2003; see also, Khan & Crawford, 2001). Thus, that eye’s view may prevail over the other’s in perceiving relative depth of objects.

Previously we have reported small temporal asynchronies between the two eyes in beginning and ending fixations, in binocular eye-tracking data from native language-users reading multiline text in English and Chinese (Zhu et al., 2021) from a five-language corpus. We have interpreted the spatial patterns of these asynchronies across the text in terms of ocular prevalence: if one eye’s fixation begins earlier or ends later, it favors making that eye prevalent in the conscious perception of the fused visual input from the two eyes, or maintaining that eye as the prevalent eye. Overall, text in the left visual field tended to be prioritized by the left eye, text in the right visual field tended to be prioritized by the right eye. This behavior is optimal: the prevalent eye tends to be closer and more square-on to its text.

English and Chinese are very different orthographies. The fact that native-language readers produced similar fine-grain oculomotor behaviors in the two languages suggested that the behaviors were chiefly the result of the peripheral musculature responsible for horizontal eye movement – an example of “embodied cognition” (cf. Wilson, 2002), in which the physical movement of the eyes has intrinsic implications for higher, conscious visual cognition.

However, there are important asymmetries in the visual pathways and cortical areas responsible for reading. These asymmetries involve, first, ocular dominance (as distinct from ocular prevalence, e.g. Toosy et al., 2001; Walls, 1951); ocular dominance is variously defined as the eye whose input is favored longterm in sighting tasks, the eye preferred when monocular views are discrepant, or the eye manifesting physiological or refractive superiority (Porac & Coren, 1976). The second asymmetry is hemispheric specialization (e.g. Cohen et al., 2000; Coltheart, 1983; Shillcock & McDonald, 2005). Are these asymmetries at all involved in the temporal asynchronies reported by Zhu et al.? If they are, then the reading behaviors of the right-to-left and left-to-right orthographies will not simply be mirror images. To answer this question, we compare, below, the behaviors of the Chinese and English readers with binocular eye-tracking data from the same task in the five-language corpus but involving two right-to-left orthographies, Arabic and Hebrew.

Arabic and Hebrew are alphabetic, Semitic languages. They share similar morphological, semantic, and syntactic structures (Shimron, 2003). Both are read from right to left (within the word and within the sentence) and are processed through a roots-based system composed of three or four letters (Abu-Rabia, 2002; Abu-Rabia & Siegel, 2003). In both orthographies, some vowels and diacritics are ordinarily omitted in “unpointed” text intended for skilled readers. Greater processing difficulty has been reported in reading Arabic compared with Hebrew (Brysbaert, 2019; Eviatar et al., 2004, 2019; Lahoud et al. (Unpublished manuscript).

Compared with English reading behaviors, Arabic and Hebrew show both similarities and differences. For example, the size of the perceptual span is comparable in Arabic and English, although its asymmetry is reversed (Jordan et al., 2013; see also; Lallier et al., 2018); the greater informational density of the semitic orthographies quantitatively influences what is a qualitatively similar repertoire of eye-movement behaviors (see, e.g., Hermena & Reichle, 2020, for a review) (for other comparisons, cf. Frost, 2009; Frost et al., 2005; Jordan et al., 2010; Kuperman et al., Unpublished manuscript; Lahoud et al., 2023; Paterson et al., 2015; Velan & Frost, 2007a).; How might directionality affect reading? Since reports by Beaumont (1982) and Bradshaw and Nettleton (1983), it has been clear that there is typically a right visual field (i.e. left hemisphere) processing advantage for reading isolated words. In the relevant experiments with left-to-right orthographies, words are approached from the left, meaning that their relatively more informative beginnings (a phonological fact grounded in speech perception, cf. Nooteboom, 1981) are initially parafoveally projected directly to the left hemisphere. This fact initially suggests left-to-right orthographies are optimal. Deutsch et al. (2000) report a comparable parafoveal facilitation by letters from Hebrew triconsonantal roots; Hermena et al. (2021) report a similar effect in Arabic readers. Thus, Arabic and Hebrew can also benefit from the relatively informative view of the beginning of the word. It will, however, be projected to the right hemisphere. Ibrahim and Eviatar (2009) demonstrate a greater left hemisphere advantage for word recognition in L1 Arabic readers for Arabic compared with L2 Hebrew and L2 English. Does the hemispheric division of labor in word recognition mean that right-to-left scanning of Arabic and Hebrew is not optimal? We return to this question, below.

A further aspect of optimal binocular behavior involves the degree and direction of fixation disparity. The two eyes typically do not fixate exactly conjointly during reading (Liversedge et al., 2006; Shillcock et al., 2010). In our data, the majority of fixations have the right eye fixating left of the left eye (a “crossed fixation”). Crossed fixations, as opposed to “uncrossed fixations” (right eye right of the left eye) are to be expected from normal luminance conditions for reading (a light room, dark-on-light text) (cf. Kirkby et al., 2013; Shillcock et al., 2010). The disparity within a binocular fixation can be non-trivial, often involving several letters. What does this behavior mean for parafoveal preview in the two reading directions? Two facts about visual anatomy are important here. First, Toosy et al. (2001) show a stronger contralateral than ipsilateral connection between the individual eyes and hemispheres, reflecting (a) the exclusively contralateral monocular crescent of each eye’s temporal hemifield, (b) the biased crossed projection of nasal retinal ganglion cells, which drive the contralateral ocular dominance columns in V1, and (c) the blind spot representation in the ipsilateral visual cortex. Behaviorally, Obregón and Shillcock (2012) show that contralateral projection (right eye to left hemisphere, left eye to right hemisphere) is significantly better than ipsilateral projection (right eye to right hemisphere, left eye to left hemisphere) for lexical processing. Overall, these hemispheric differences mean there may be a differential reliance on the two eyes in the binocular reading of the two directions of reading.

The second fact about visual anatomy concerns the splitting of the fovea (Ellis & Brysbaert, 2010; Lavidor & Walsh, 2004; Shillcock et al., 2000); in each eye, the foveal projection is more or less precisely split, such that the two visual hemifields are projected contralaterally to the two hemispheres. All of this means that a non-conjoint binocular fixation on a line of text divides the perceptual span into distinct regions, defined by the foveal projection to the brain. They are illustrated for a typical crossed fixation in our data, by the following list, proceeding left-to-right through a left-to-right orthography:

1. The right eye fixates text that is likely to have been targeted by the previous fixation. The right eye’s left visual field is projected to the right hemisphere – an ipsilateral projection, which is likely to be less good than a contralateral projection.

2. The region between the two fixation points is projected to the left and right hemisphere by the right and left eye respectively – both optimal contralateral projections. This overlap constitutes the best visual processing, with the right eye’s right visual field going to the lexically specialized left hemisphere.

3. The left eye fixates the upcoming text and projects its right visual field input (the farthest lookahead) to the left hemisphere – an optimal ipsilateral projection to the lexically specialized left hemisphere.

In a right-to-left orthography, the difference is that it is the left eye that is fixating the “previously seen” text and it is the right eye that is obtaining the farthest lookahead in the upcoming text and projecting it contralaterally to the non-lexically specialized right hemisphere. Initially, it seems that the lexically privileged left hemisphere is not getting a direct, furthest parafoveal lookahead in right-to-left orthographies; this issue is orthogonal to the fact, reported by Jordan et al. (2013), of essentially mirror-image asymmetries of the perceptual span in the direction of reading for Arabic and English.

Cross-linguistic research on binocular coordination in reading has mainly focused on left-to-right orthographies. Hsiao (2017) explored binocular disparities in English and Chinese with temporally conjugate fixations (i.e. right and left eye both start and end their fixations at the same time) and reports similarities and differences across the two languages. Zhu et al. (2021) investigate temporal disjugacy, (i.e. the small temporal asynchronies in the two eyes’ fixations) in the same eye-tracking corpus as Hsiao and discover strikingly similar distributional patterns across the two very different orthographies, for the three most numerous types of temporal asynchrony (cf. Figure 2, below). In both English and Chinese, temporally synchronized binocular fixations accounted for slightly more than half of the binocular fixations, with over 80% of binocular fixations in all ending synchronously.

How does ocular prevalence play out in reading the right-to-left orthographies Arabic and Hebrew? We tested two hypotheses:

Hypothesis (1)

In right-to-left orthographies, as in left-to-right orthographies, the LE will tend to begin fixating earlier and stay fixating longer in the LVF, and the RE will be similarly prioritized in the right visual field, thereby facilitating the appropriate switching of ocular prevalence.

This hypothesis is based on the assumption that ocular prevalence will favor the eye that is closest and most directly square-on to the text on the screen. This analysis extends the study of the reading of the left-to-right orthographies English and Chinese (Zhu et al., 2021).

Hypothesis (2)

The pattern of binocular temporal asynchronies in right-to-left orthographies will exactly mirror the pattern in the left-to-right orthographies English and Chinese.

This strong hypothesis subsumes Hypothesis 1 and is based on the further assumption that it is the symmetrical horizontal movement of the eyes alone that is responsible for the pattern of behavior. Any departure from exact symmetry will implicate the asymmetries in projection represented by RE ocular dominance and/or by any specialization for reading behaviors in each of the hemispheres.

Method

Participants

We analyzed the data from an existing corpus of eye-tracking data, gathered with approval given by the Psychology Research Ethics Committee of the University of Edinburgh following the Code of Ethics and Conduct of The British Psychological Society (BPS). In the original study, 28 Arabic and 16 Hebrew native speakers were paid for their participation and gave informed consent. All reported having normal or corrected-to-normal vision. The data are compared with analysis from reading left-to-right orthographies from Chinese and English native speakers, previously reported as (Zhu et al., 2021). Table 1 includes the demographic information of all participants.

Table 1. Demographic characteristics of participants.

Languages	Arabic		Hebrew		English		Chinese
	n	%	n	%	n	%	n	%
Total sample	N = 28		N = 16		N = 42		N = 42
Age	Mean = 28.7		Mean = 30.1		Mean = 23.6		Mean = 26.1
Gender
Female	20	71	8	50	24	57	28	67
Male	8	29	8	50	18	43	14	33
Employment
Student	19	67	6	37	35	83	34	80
Employed	9	33	10	63	7	17	8	20

Apparatus

Participants sat in a room with diffused lighting, and watched a 22” Ilyama Vision Master Pro 514 display, at a distance of 75 cm. The screen resolution was 1024 × 768 pixels. A chin-rest and forehead support kept the head stable. The eye-tracker was an SR Research EyeLink II head-mounted video-based tracker.

Stimulus materials and procedure

Eye movements were recorded binocularly, with pupil and corneal reflection, and sampled at 500 Hz (i.e. every 2 msec), during the reading of Arabic (al naskh font) and Hebrew (Miriam font) texts, each comprising 21 newspaper stories, with a total of 5000 words for each language, presented in black characters on a light background, on consecutive pages with up to five right-justified lines of text each. The stimuli were intended to be comparable in form and content across the four languages in the overall study (English, Chinese, Arabic, Hebrew), based on the intuitions of native speakers. The maximum line length corresponded to 64 English characters. Readers were calibrated monocularly with a 9-point fixation grid while occluding the other eye with a black paper shade. Participants fixated a black fixation disc before each page of text was displayed and fixated a square at the bottom left of each page after finishing reading it. They responded on the keyboard to a yes/no question after each story, to ensure reading for meaning. Mean comprehension accuracy was 90.5% and 90.8% for Arabic and Hebrew, respectively, indicating that participants read for meaning; no data were excluded on this criterion. The grid of fixation targets was presented before the next article, to check the calibration accuracy. The whole recording process consisted of three blocks with intervening rest-breaks, lasting for around 1.5 hours in total.

Analysis

The analysis of very large data sets raises the question of the viability of one single approach. Silberzahn et al. (2018) report using 29 teams involving 61 analysts to address the same research question based on the same data set. They report significant variation in the results of the analyses. We have responded to this dilemma in two ways.

First, we present visualizations of the data based on intuitively simple behaviors of the two eyes – which eye began fixating first, which eye finished fixating first (see Figure 1). The visualizations reveal a clear major pattern in the data, as addressed in Hypothesis 1.

Figure 1. Typology of binocular fixation asynchronies. Left-priority types: T1, T6, T7. Right-priority types: T2, T3, T5. Note: lines are fixation durations. The three lines are ordered in terms of fixation onset: left, right and simultaneous.

Second, we address the choice of variables. Eye-movements in text reading have been ingeniously studied for several decades. Researchers have isolated a rich variety of physical and linguistic dimensions of text stimuli, mostly in English but increasingly in other languages, and have studied the effects on fixation time, gaze duration, length of saccade, word skipping, uptake of parafoveal information, fixation disparity, regressive fixations, semantic and phonological influences, among other behaviors. We have responded to the richness of the data by selecting variables (left eye, right eye; left of screen, right of screen) that ultimately speak to the most universal aspect of material cognitive organization, the division into two cerebral hemispheres; hemispheric division has a role in all aspects of cognition. A multitude of other analyses are possible, but we claim that the ones we employ here have a special processing status.

For each binocular fixation, the start-time offset was calculated as the fixation start-time of the RE minus the fixation start-time of the LE. The end-time offset was calculated analogously. A difference of ±2 ms between events in the two eyes was considered as simultaneous. StartTime

offset or EndTime offset < 2 means the right eye starts or ends earlier than the left eye. StartTime offset or EndTime offset > 2 means the left eye starts or ends earlier than the right eye.

Figure 1 depicts the comprehensive typology of offsets. There are nine possible types of binocular fixation, with different patterns of start-time and end-time offsets. For example, Type 1 shows both eyes starting fixation synchronously and the LE fixating for longer. We first produced demonstrative graphs to show the distributions of the types on the screen. We then analyzed the data quantitatively with GLMER models (see below for details), to further characterize the specific eye-movement behaviors across left-to-right and right-to-left orthographies.

Results

We analyzed a total of 138,493 binocular fixations (i.e. individual fixations by the left and right eye overlapping in time) for the Arabic readers and 82,591 binocular fixations for the Hebrew readers, compared with 160,567 binocular fixations from English readers and 158,794 binocular fixations for the Chinese readers previously reported in Zhu et al. (2021). Below, we first report descriptive statistics from two perspectives: (a) the overall distribution of the different types; (b) the spatial distribution of the types across the screen on which the text stimuli were displayed. Then, we report the quantitative analysis from GLMER models from two perspectives: (a) screen differences under subsets of fixation types (cf. Figure 1) in each group of right-to-left orthographies (i.e. language, Hebrew and Arabic). (b) differences between right-to-left orthographies and left-to-right orthographies with different sides of the screen. Overall, the results indicate a lawful patterning of binocular behaviors relevant to ocular prevalence, across the visual field, but with some significant differences associated with reading direction.

Overall results

Figure 2 shows the overall distribution of each type of binocular fixation and their percentages for the readers of the four languages (i.e. Arabic, Hebrew, English and Chinese, the latter two taken from Zhu et al., 2021) respectively. There are visible similarities across all four languages, with the three most numerous types of binocular fixation being Syn (synchronized pairs), T3 (right eye priority) and T6 (left eye priority); in particular, synchronized binocular fixations account for around half the binocular fixations for all languages.

Figure 2. The distribution of types of asynchrony in Arabic, Hebrew, English and Chinese readers (in left-to-right order within each set of bars).

There are asymmetries depending on reading direction. English and Chinese show more T3 than T6, Hebrew and Arabic show more T6 than T3. A less pronounced interaction, with smaller numbers of fixations, may be seen for T1 and T2: English and Chinese have more T1 than T2, Arabic and Hebrew, have more T2 than T1.

Spatial distribution

Figures 3 and 4 show the distribution of fixation types across the lines of text, for Arabic and Hebrew, respectively. The hexbin graphs show mean coordinates (during fixation) of the right eye (RE) for each binocular fixation; the choice of RE over left eye (LE) has no implications here. The distributions also show readers fixating a square at the bottom left of each page after reading it.

Figure 3. Spatial distribution of types in Arabic readers. Left-priority types: T1, T6, T7. Right-priority types: T2, T3, T5. The three lines are ordered in terms of fixation onset: left, right and simultaneous.

Figure 4. Spatial distribution of types in Hebrew readers. Left-priority types: T1, T6, T7. Right-priority types: T2, T3, T5.

Similar patterns obtain for both Semitic languages. Syn, T6 and T3 show the greatest densities. There are clear differences between left, middle and right of the text.

Figure 5 shows English data from Zhu et al. (2021), which closely resembled the associated Chinese data in that paper. Comparing the right-to-left orthographies (Figures 3 and 4) with the left-to-right orthography (Figure 5), the data show similarities, symmetries and departures from symmetry.

Figure 5. Spatial distribution of types in English readers. Left-priority types: T1, T6, T7. Right-priority types: T2, T3, T5.

As Figure 1 shows, the eight non-Syn types form four symmetrical pairs in terms of the eye to which they afford priority either early or late in the fixation: T5 (RE early and late) and T7 (LE early and late); T3 (RE early) and T6 (LE early); T4 (RE early, LE late) and T8 (LE early, RE late); T2 (RE late) and T1 (LE late). We first consider the pairs in terms of their visible distributions in Figures 3–5.

T5 is skewed to the right and T7 is skewed to the left for both reading directions.

T3 is concentrated at the right and T6 at the left for both reading directions but T3 in English and T6 in Arabic and Hebrew are more distributed across the line.

T4 and T8 are the “mixed” fixation types, which have both RE and LE priority depending on whether the start or end of fixation is concerned. T4 is skewed rightwards for Hebrew and Arabic but is more ambiguous for English. T8 is skewed leftwards for English but is more ambiguous for the Semitic languages.

T1 is skewed leftwards and T2 rightwards for both reading directions. However, T2 has more fixations persisting across the screen for the two Semitic orthographies. This pattern is apparently not reciprocated with T1 in the English data.

Modelling results

We analyzed the large volume of data in Figures 3 and 4 quantitatively. We divided the screen equally into left, middle and right regions, to investigate the distributions of fixation types across the text. We further compared the two Semitic right-to-left orthographies with the two left-to-right orthographies, English and Chinese.

We use General Linear Mixed-Effects Regression Models (GLMER), carried out in R’s lme4 software package (Bates et al., 2008). We used counts of fixation pairs as dependent variables in all the models. As our dependent variable was counts of events, all our modeling used Poisson error distributions. We defined null models with participants, articles and screen orders (i.e. page order of each article) as random factors. Predictor variables included the sides of the screen (left, middle and right, approximating left, middle and right of lines of text) and direction of reading (left-to-right for English and Chinese, right-to-left for Hebrew and Arabic) and tested as a whole as well as under subsets of types to explore the quantitative distribution of the types. All Model fit was assessed using the anova function to compare models. The results show a systematic pattern of binocular behavior across languages.

In the first part, we added Group (i.e. Hebrew and Arabic, with Arabic as reference) as fixed effect to explore the group differences in our descriptive data. The models were built under subsets of Syn, T1, T2, T3, T5 and T6 respectively. In the second part, we added sides of the screen (left, middle, right; with the middle as reference) as a predictor variable. The models were built one by one under subsets of asynchronized types to explore variation in the effects of the part of the screen on these different types, which were also tested separately within the Arabic and Hebrew data, in their own subsets. In the third part, we examined data with all four languages and added the reading direction (i.e. left-to-right and right-to-left, with left-to-right as reference) and sides of screen (reference varies according to the relevant distribution of types) as predictor variables. The models were built under the subset of the four most numerous asynchronized types (i.e. T6, T3, T2 and T1) and two ambiguous types (T4 and T8) separately.

The first part of the modeling of group differences shows both similarities and differences in the results. Group differences did not appear any qualitatively different between the two groups of readers in any other type or as a whole in the two Semitic languages, in accordance with Figure 2. The two Semitic languages elicited similar behaviors in binocular fixation-pairs in general.

In the second part of the modeling, the majority of the non-Syn types – T1, T2, T3 and T6—are similarly distributed in both groups. First, both T1 and T6 (both LE priority) have significantly more value on the left side of the screen and/or less value on the right side of the screen, compared with the middle (i.e. the reference). Specifically, T1 in Arabic (Est = 0.1308, SE = 0.0333, z(3514) = 3.923, p < .001) and in Hebrew (Est = 0.1652, SE = 0.0440, z(2239) = 3.753, p < .001) shows a statistical concentration on the left side of the screen. Table 2 further shows T6 in Hebrew readers has significantly less on the right side. Similar results can be observed in Arabic readers, while additionally Arabic T6 has significantly more value on the left side.

Table 2. GLMER analysis of Screen differences in Hebrew and Arabic T6.

GLMER analysis of Screen differences in Hebrew T6
	The number of fixations
	null model(1)	sides of screen(2)
Left		-0.011 (0.546)
Right		-0.469 (<0.001)
Constant	1.114 (<0.001)	1.236 (<0.001)
Random effects
	Variance	Std.Dev.
Participant (16)	0.0762	0.2762
Article (119)	0.0275	0.1661
Observations	4,412	4,412
Log Likelihood	-8,559.832	-8,274.856
Akaike Inf. Crit.	17,125.660	16,559.710
Bayesian Inf. Crit.	17,144.840	16,591.670
GLMER analysis of Screen differences in Arabic T6
	The number of fixations
	null model(1)	sides of screen(2)
Left		0.077 (<0.001)
Right		-0.376 (<0.001)
Constant	1.068 (<0.001)	1.134 (<0.001)
Random effects
	Variance	Std.Dev.
Participant (28)	0.1359	0.3687
Article (120)	0.0270	0.1645
Observations	7,034	7,034
Log Likelihood	-13,944.200	-13,541.870
Akaike Inf. Crit.	27,894.400	27,093.730
Bayesian Inf. Crit.	27,914.970	27,128.020

Note: p value in parentheses.

Furthermore, T2 and T3 (both RE priority) have significantly less value on the left side of the screen in both orthographies, and T3 also has additional significantly more value on the right side (Tables 3 and 4). The two Semitic languages again pattern very similarly.

Table 3. GLMER analysis of Screen differences in Hebrew and Arabic T2.

GLMER analysis of screen difference in Hebrew T2
	The number of fixations
	null model(1)	sides of screen(2)
Left		-0.340 (<0.001)
Right		-0.016 (0.515)
Constant	0.752 (<0.001)	0.840 (<0.001)
Random effects
	Variance	Std.Dev.
Participant (16)	0.0651	0.2552
Article (119)	0.0125	0.1122
Observations	3,696	3,696
Log Likelihood	-6,090.340	-6,004.389
Akaike Inf. Crit.	12,186.680	12,018.780
Bayesian Inf. Crit.	12,205.330	12,049.850
GLMER analysis of screen difference in Arabic T2
	The number of fixations
	null model(1)	sides of screen(2)
Left		-0.285 (<0.001)
Right		-0.023 (0.302)
Constant	0.600 (<0.001)	0.679 (<0.001)
Random effects
	Variance	Std.Dev.
Participant (28)	0.0828	0.2878
Article (120)	0.0084	0.0920
Observations	5,240	5,240
Log Likelihood	-8,155.412	-8,079.160
Akaike Inf. Crit.	16,316.820	16,168.320
Bayesian Inf. Crit.	16,336.510	16,201.140

Note: p value in parentheses.

Table 4. GLMER analysis of Screen differences in in Hebrew and Arabic T3.

GLMER analysis of screen difference in Hebrew T3
	The number of fixations
	null model(1)	sides of screen(2)
Left		-0.119 (<0.001)
Right		0.662 (<0.001)
Constant	0.850 (<0.001)	0.502 (<0.001)
Random effects
	Variance	Std.Dev.
Participant (16)	0.0502	0.2241
Article (119)	0.0146	0.1210
Observations	3,380	3,380
Log Likelihood	-6,160.798	-5,647.221
Akaike Inf. Crit.	12,327.600	11,304.440
Bayesian Inf. Crit.	12,345.970	11,335.070
GLMER analysis of screen difference in Arabic T3
	The number of fixations
	null model(1)	sides of screen(2)
Left		-0.223 (<0.001)
Right		0.489 (<0.001)
Constant	0.930 (<0.001)	0.712 (<0.001)
Random effects
	Variance	Std.Dev.
Participant (28)	0.0869	0.2949
Article (120)	0.0217	0.1476
Observations	5,456	5,456
Log Likelihood	-10,009.990	-9,390.262
Akaike Inf. Crit.	20,025.970	18,790.520
Bayesian Inf. Crit.	20,045.780	18,823.550

Note: p value in parentheses.

This systematic binocular pattern is further seen in T5 (RE priority) and T7 (LE priority) in both orthographies, though the data are relatively sparse. We find significantly more value at the right side in T5 and more value at the left side in T7 in both orthographies. In particular, the results of Arabic T5 shows a significant concentration at the right side (Est = 0.1442, SE = 0.0495, z(1771) = 2.908, p < .01) and the same for Hebrew T5 (Est = 0.1741, SE = 0.0531, z(1391) = 3.277, p < .01), whereas the concentration for T7 is significant at the left side for both Arabic (Est = 0.1190, SE = 0.0503, z(1794) = 2.365, p < .05) and Hebrew (Est = 0.2467, SE = 0.0594, z(1207) = 4.149, p < .001).

In the third part of the modeling, orthographies with different reading directions show significant differences in the general distribution as well as in the four main asynchronized types – T6, T3, T2 and T1, and in the two ambiguous types – T4 and T8, with respect to different sides of the screen. The result from the general distribution (with left-to-right orthography and middle of the screen as references) shows significantly less value on the left side for the right-to-left orthographies (Est= −0.1562, SE = 0.006, z(163344)= −23.248, p < .001) and also more value on the right side (Est = 0.0787, SE = 0.006, z(163344) = 11.801, p < .001).

In the specific types, Table 5 (with left-to-right orthography and right of the screen as references) shows significantly less on the left side but significantly more in the middle of the screen in the right-to-left orthographies in T6 (early-LE-priority) for the right-to-left orthographies, which is in accordance with the expanded distribution of T6 across the line in Figures 3 and 4, compared with Figure 5. On the other hand, T3 (early-RE-priority) as shown in Table 6 (with left-to-right orthography and left of the screen as references) suggests the opposite significant change, with statistically less in the middle but more on the right side of the screen, for the RTLOs. In addition, the distribution of T2 (late-RE-priority) across the line also shows significance in Table 7 (with left-to-right orthography and left of the screen as references) at both middle and the right side of the screen in the right-to-left orthographies, whereas T1 (late-LE-priority) shows significantly more value at both left and right sides of the screen in the right-to-left orthographies (with left-to-right orthography and middle of the screen as references) in Table 8. This modeling bears out the visual interpretation of Figures 2 and 3.

Table 5. GLMER analysis of differences in direction of reading T6.

	The number of fixations
	reading direction	sides of screen-reading direction
	(1)	(2)
Right to Left	0.275 (<0.001)	0.375 (<0.001)
Left		0.695 (<0.001)
Middle		0.161 (<0.001)
RTL:Left		−0.242 (<0.001)
RTL:Middle		0.248 (<0.001)
Constant	0.820 (<0.001)	0.397 (<0.001)
Random effects
	Variance	Std.Dev.
Participant(128)	0.0848	0.2912
Article(82)	0.0220	0.1483
Screen(5–9)	0.0010	0.0331
Observations	27,550	27,550
Log Likelihood	−51,161.450	−48,750.650
Akaike Inf. Crit.	102,334.900	97,521.300
Bayesian Inf. Crit.	102,384.200	97,603.540

Note: Screens differ for the article and as random slope.

Table 6. GLMER analysis of differences in direction of reading T3.

	The number of fixations
	reading direction	sides of screen-reading direction
	(1)	(2)
Right to Left	−0.055 (0.309)	−0.288 (<0.001)
Middle		0.255 (<0.001)
Right		0.328 (<0.001)
RTL:Middle		−0.070 (<0.001)
RTL:Right		0.406 (<0.001)
Constant	0.957 (<0.001)	0.744 (<0.001)
Random effects
	Variance	Std.Dev.
Participant(128)	0.0748	0.2735
Article(82)	0.0221	0.1486
Screen(5–9)	0.0019	0.0441
Observations	29,349	29,349
Log Likelihood	−53,546.840	−51,934.880
Akaike Inf. Crit.	107,105.700	103,889.800
Bayesian Inf. Crit.	107,155.400	103,972.600

Note: Screens as random slope.

Table 7. GLMER analysis of differences in direction of reading T2.

	The number of fixations
	reading direction	sides of screen-reading direction
	(1)	(2)
Right to Left	0.303 (<0.001)	0.182 (<0.001)
Middle		0.021 (0.328)
Right		0.198 (<0.001)
RTL:Middle		0.287 (<0.001)
RTL:Right		0.090 (<0.001)
Constant	0.367 (<0.001)	0.263 (<0.001)
Random effects
	Variance	Std.Dev.
Participant(128)	0.0469	0.2166
Article(82)	0.0022	0.0474
Observations	19,899	19,899
Log Likelihood	−28,822.090	−28,591.260
Akaike Inf. Crit.	57,652.190	57,198.530
Bayesian Inf. Crit.	57,683.780	57,261.720

Note: p value in parentheses.

Table 8. GLMER analysis of differences in direction of reading T1.

	The number of fixations
	reading direction	sides of screen-reading direction
	(1)	(2)
Right to Left	−0.061 (0.103)	−0.157 (<0.001)
Left		0.065 (<0.001)
Right		−0.246 (<0.001)
RTL:Left		0.081 (<0.001)
RTL:Right		0.280 (<0.001)
Constant	0.441 (<0.001)	0.460 (<0.001)
Random effects
	Variance	Std.Dev.
Participant(128)	0.0323	0.1799
Article(82)	0.0016	0.0411
Observations	18,392	18,392
Log Likelihood	−25,060.780	−24,909.780
Akaike Inf. Crit.	50,131.570	49,835.570
Bayesian Inf. Crit.	50,170.670	49,898.120

Note: p value in parentheses.

In addition, the two ambiguous types T4 (early-RE-priority and late-LE-priority) and T8 (early-LE-priority and late-RE-priority) also show differences between languages according to reading direction. Specifically, the right-to-left orthographies show significantly more fixations on the right of the screen in T4 (Est = 0.1919, SE = 0.0656, z(6960) = 2.925, p < .01). On the other hand, T8 shows a relatively greater concentration in the middle for the right-to-left orthographies, with both less on the left (Est= −0.1406, SE = 0.0527, z(7680)= −2.665, p < .01) and less on the right (Est= −0.2066, SE = 0.0584, z(7680)= −3.528, p < .001). As Figure 4 shows, this pattern is the converse of that found for the left-to-right orthographies.

Discussion

We have analyzed the distribution of small temporal asynchronies of binocular fixation in the reading of continuous multiline text in two right-to-left orthographies, Arabic and Hebrew, and we have compared them with the same type of data from two left-to-right orthographies, English and Chinese. We tested the hypothesis that the right eye would tend to begin fixating first and/or end fixating second on the right side of the text, and conversely that the left eye would tend to begin fixating first and/or end fixating second on the left side of the text. This hypothesis was based on the theory (Zhu et al., 2021) that: (a) ocular prevalence is just as present in reading as in typical scene viewing; (b) reading benefits from ocular prevalence, such that the eye that is closest and squarest-on to the text has priority in the conscious fused percept of the text; (c) small binocular timing asynchronies contribute to ocular prevalence moving fluidly between the two eyes in skilled reading. There are three aspects to our results.

Ocular prevalence

Arabic and Hebrew reading behaviors matched our predictions concerning ocular prevalence. Overall, the pattern of temporal asynchronies in the right-to-left orthographies closely resembled those of the previously reported left-to-right orthographies: the right eye’s input tends to be prioritized on the right side of the screen and the left eye’s on the left side (Figures 3–5).

Reading direction and symmetry

There is an overall quantitative symmetry of the data for the two directions of reading. Fixation types T3 and T6 in Figure 2 best demonstrate this left-right reversal of the pattern of asynchronies, but we also see this symmetry in the less numerous pairs of types. This symmetry is a critical confirmation of our claim that the observed asynchronies in left-to-right orthographies resulted from anatomical constraints interacting with the direction of reading.

Our theory of the role of ocular prevalence in reading is based on the speed and power of the lateral rectus muscle. It guarantees that abductive movement (away from the nose) of either eye has greater acceleration than adductive movement (toward the nose), in the horizontal plane (Robinson, 1964). In a left-to-right orthography, this acceleration means that the abducting right eye is likely to saccade away from a fixation more promptly, leaving the left eye still fixating (a late left eye priority). The right eye is also likely to arrive at the next fixation more promptly, allowing it to begin fixating earlier (an early right eye priority). This situation is reversed for the return sweep and for regressive saccades within a line of a left-to-right orthography, when the left eye becomes the abducting eye. The effects of the speed and strength of the lateral rectus muscle will tend to be more evident in longer saccades, which necessarily fall more toward the right or left extremes of a line (depending on reading direction) because such a landing position affords more scope for longer saccades. Figures 3–5 show just such concentrations of the non-Syn types. In right-to-left orthographies there are progressively more T6 fixations (making the left eye the prevalent eye) as the reader progresses leftwards and in left-to-right orthographies progressively more T3 fixations (making the right eye the prevalent eye) in the rightward direction of reading. The complementary T3 fixations in right-to-left orthographies and T6 fixations in left-to-right orthographies show a dense patterning largely accounted for by the respective return sweeps.

T4 (early right eye & late left eye priority) and T8 (early left eye & late right eye priority) are “mixed” types. They potentially reveal the relative importance of early and late priority in determining prevalence. T4 are skewed to the right of the screen for the right-to-left orthographies. The complementary T8 predominate on the left for left-to-right orthographies. This pattern is in line with an earlier start to fixation – as opposed to a late end – predicting prevalence; overall, binocular fixations that begin asynchronously make up some one-third of the total.

When we look at the converse interaction of these two types with reading direction, we again see symmetry: T8 in the right-to-left orthographies and T4 in the left-to-right orthographies tend to cluster in the middle of the screen, suggesting their mixed nature with regard to switching prevalence may assert itself more in the middle of the screen where pressure to switch prevalence is subtlest.

Binocular fixations with precisely synchronized starts and ends constitute the modal behavior for both directions of reading. Syn sustains the current ocular prevalence. Typically, the status quo is optimal.

Departures from symmetry

There are quantitative departures from this very salient symmetry. They are associated with direction of reading and are most clearly seen in comparing the distributions of the complementary types T1 and T2. In Figures 3–5.

T1 and T2 have fixation ends that favor the left eye and right eye respectively. The modeling showed two unexpected departures from symmetry in right-to-left orthographies: there were more T1 at both left and right sides of the screen and there were more T2 in the middle of the screen than would be expected.

There are two very relevant anatomical and behavioral asymmetries in the reading system. First, ocular dominance (chiefly of the right eye, over the population) refers to the longterm preference of one eye in sighting tasks (Porac & Coren, 1976). Second, hemispheric specialization of the left hemisphere for phonological and lexical processing. In addition, the overwhelming majority of fixations in our data were “crossed” in which the left eye fixates to the right of the right eye. These facts suggest interpretations of the observed asymmetries.

A relative concentration of T1 on the right of the screen means a left eye prevalence at the beginning of the line in a right-to-left orthography; fixations begin synchronously and the left eye stays fixating longer. (In the left-to-right orthographies T2 fixations do not exhibit the converse behavior to T1 in right-to-left orthographies.) We suggest that this T1 sub-population is adaptive because it means the right eye moving leftwards onto new text sooner. In addition, the right eye has an adaptive direct contralateral projection of its right visual field to the lexically privileged left hemisphere. Such an early movement by the right eye is adductive, meaning it is not facilitated by the powerful lateral rectus muscle, but we suggest it is facilitated by the right eye typically being the dominant eye.

In right-to-left orthographies, T2 in the middle of the screen means a right eye prevalence persisting from the right side of the screen: fixations begin synchronously and the right eye stays fixating longer. We interpret this pattern as the right eye taking advantage of having first access to the previously unseen text on the left and of having a direct contralateral projection to the left hemisphere. Again, we attribute the right eye’s ability to do this to its being typically the dominant eye.

Embodied cognition

It is physically impossible for the muscular control of the eyes to guarantee 100% precise coordination of fixation onsets and offsets. The 50% level of Syn responses is itself impressive. However, the departures from complete control contribute adaptively to the high-level division of labor between the eyes, represented by ocular prevalence. The concept of “embodied cognition” has been used in a variety of ways (Wilson, 2002). The relationship we have described between the physical movement of the eyeballs and the conscious perception of the text is one such example. Conscious perception depends on ocular prevalence and the latter is an integral part of how the two eyes are necessarily co-ordinated.

Orthography and the hemispheres

We have described some very clear patterns of eye-movement data in reading. In this section we attempt to situate these patterns within a wider theoretical picture. That picture is based on our claim that the hemispheric division of the brain is the most powerful perspective within this field of theorizing – it is a “universal”, that is, it speaks to virtually all aspects of cognition and perception. That division is closely associated with the two eyes. The patterns of data we have presented sit within this claim and a coherent set of assumptions and predictions emerges.

Consider the issue of the asymmetry between lexical processing in the two hemispheres and the direction of reading. In a crossed fixation in a left-to-right orthography, the farthest lookahead into the upcoming text is from the left eye and is directly projected contralaterally into the lexically privileged left hemisphere, albeit from an ipsilateral projection from the left eye. But in a right-to-left orthography, the farthest lookahead is from the right eye, directly projecting to the right hemisphere; only the “backwards looking” right visual field of the right eye projects to the left hemisphere. Note that even in a rarer uncrossed fixation reading right-to-left, the projection of the then-leading left eye’s left visual field is still to the right hemisphere. Is this early reliance on the right hemisphere a problem for readers of a right-to-left orthography?

If reading simply involved accessing the lexical representations of the left hemisphere, then right-to-left orthographies would seem maladaptive in terms of the farthest lookahead. However, a number of hemispheric differences are potentially relevant to reading. Some may favor early left hemisphere involvement, like categorical lexical processing. There is also a left hemisphere preference for the horizontal (Pettigrew, 2001), a capacity for focal attention (e.g. Weissman & Woldorff, 2005), a preeminent network of phonological, semantic and syntactic processing (Vigneau et al., 2006), a propensity for categorical perception (Holmes & Wolff, 2012), inferencing based on coherence (Beeman et al., 2000), speed of executing routinized processing (Goldberg & Costa, 1981), processing of categorial spatial relations (Kosslyn, 1987), frequency-based elimination of lexical ambiguity (Burgess & Simpson, 1988), and the capacity for “fine-coding” (M. Beeman et al., 1994). All these appear to favor the privileged access to the left hemisphere in left-to-right reading.

In contrast, there are hemispheric differences that might favor right-to-left orthographies and direct initial projection to the right hemisphere. They include the ability to span the visual field more widely than the left hemisphere, attending to both extremes of the perceptual window (an example of the more general ability of the right hemisphere to incorporate aspects of left hemisphere processing) (Ellis et al., 2006; Somers & Sheremata, 2013), the capacity for “coarse-coding” (M. Beeman et al., 1994; see also, e.g., Grainger et al., 2012; Grainger & Ziegler, 2011, for a more abstract use of the coarse/fine distinction in visual word recognition), predictive inferencing (M. J. Beeman et al., 2000), processing of coordinate spatial relations (Kosslyn, 1987), the capacity to deal with blurred images (Cowin & Hellige, 1994), processing of fleeting letter images (Hellige & Webster, 1979), multiple activation in cases of lexical ambiguity (Burgess & Simpson, 1988), and the processing of novelty (Goldberg & Costa, 1981). (Although note that many of the experiments demonstrating hemispheric differences will have been performed on participants with a history of solely reading left-to-right orthographies; it is conceivable that extended exposure to right-to-left orthographies may affect some differences in some tasks.)

An important suggestion by Peleg et al. (2010) concerns the claim that in the RH there is a disconnection between orthographic and phonological processing, as opposed to an absence of phonological processing (although see e.g., Vigneau et al., 2011). Such a hemispheric asymmetry facilitates processing of non-dominant meanings of homographs in the right hemisphere, particularly when there is disambiguating context, because there is no frequency-trained phonological processing to create powerful attractors within orthographic-phonological processing to select the dominant meaning. Such a hemispheric asymmetry is advantageous for the processing of words with no explicit vowel information and therefore with a high proportion of homographs in the lexicon, both homophonic (e.g. “bank”) and heterophonic (e.g. “bow”).

Thus, there are pros and cons regarding reading direction. Historically, the orthographies of the world have been written and read in a variety of directions – boustrophedon (“ox-turning”) alternated right-to-left and left-to-right with each line. Direction has been determined partly by the material technology of making marks and by the physicality of handedness. The phonological form of words in a language may favor a direction of writing; further, with a direction of reading in place, we can expect the exigencies of foveal splitting and hemispheric differences to assert themselves over cultural time, to produce appropriate orthographic features. We conclude that over cultural evolution different languages develop orthographic conventions and a reading direction in which there is a stable equilibrium partly based on hemispheric differences. Specifically, we might expect the farthest lookahead going directly to the right hemisphere to elicit orthographic conventions that match the known processing propensities of the right hemisphere.

A right-to-left orthography may exploit the RH’s capacity for coarse coding, for taking context of all kinds into account (cf. Beeman et al., 1994). We see this feature in Arabic and Hebrew. First, both have “unpointed” forms for skilled readers; the whole word is compressed into a shorter span, the vowels being inferred from the word’s triconsonantal root and from lexical and sentential context. Unpointed versions of Arabic and Hebrew texts are typically shorter than English by some one-third. Thus, for the same physical fixation disparity in reading English, a left-to-right orthography, the reader can access more information in the same size of parafoveal lookahead. Second, Arabic words tend to be semantically dense, containing complex meanings (AlJassmi et al., 2021), as is also the case for Hebrew. (Chinese certainly shares this feature of semantic density; note that Chinese has an earlier history of right-to-left reading.) Third, both Arabic and Hebrew weave their morphology non-concatenatively between the consonants of the root. Fourth, both Arabic and Hebrew consonants have position-specific forms, such that position in the word informs the identity of the letter. Fifth, (although only in ornate calligraphy) Arabic letters may be vertically superimposed on each other. Sixth, a cursive form predominates in Arabic. These six instances of the role of context are all relevant to the right hemisphere’s association with coarse coding. Finally, cortical processing is plastic in the face of repeated stimulation of a particular type. Ibrahim and Eviatar (2009) show a sensitivity to morphological complexity in both hemispheres in Arabic and Hebrew lexical processing, in contrast to English readers (see Schiff et al., 2012, for details of the represention of Hebrew morphology). This sensitivity may reflect the fact that complex morphology is constantly being projected parafoveally to the right hemisphere in reading Arabic and Hebrew (cf. Deutsch et al., 2003, 2005; Rima et al., 2020), or it may be that any such propensity in left-to-right readers is simply lost because it is needed much less.

The converse arguments apply with respect to left-to-right orthographies and the strengths of the left hemisphere. Word recognition takes advantage of the full orthographic specification of words. The phonology of vowels is processed parafoveally (Schotter et al., 2012). As stated above, there is the caveat that the comparison of reading behaviors with respect to reading direction is currently far from comprehensive.

Dyslexia

The research we report here and in Zhu et al. (2021) may have implications for understanding dyslexia: note that monocular occlusion can ameliorate dyslexia (Stein et al., 2000). Some dyslexics may have less well-controlled eye-movements (Raghuram et al., 2018) and/or a less well-controlled hemispheric division of labor relevant to reading (e.g. Fabbro et al., 2001; Markee et al., 1996; Monaghan & Shillcock, 2008), resulting in suboptimal management of ocular prevalence in fusing the text images presented to the two eyes.

Conclusions

We compared multiline binocular reading analyses from Zhu et al. (2021) for the left-to-right orthographies English and Chinese with comparable data from the right-to-left orthographies Arabic and Hebrew. The data were small timing mismatches between the two eyes at the start and end of binocular fixations. We have provided critical further evidence that such asynchronies are consonant with ocular prevalence moving back and forth between the two eyes, such that input to the left eye tends to be prioritized in conscious, fused visual processing of textual targets in the left visual field and right-eye input tends to be prioritized for fixations in the right visual field. These asynchronies are best understood in terms of the greater speed and acceleration of the lateral rectus muscle in performing abductive eye movements. We suggest that if one eye’s fixation begins earlier or ends later it may elicit or maintain prevalence for that eye. Crucially, this theory of ocular prevalence holds for both directions of reading. Further, there are departures from the symmetry of the patterns of asynchronies, which we suggest reflecting the tendency for the right eye to be the dominant eye, with greater muscular control resulting from its role when fusion is impossible or undesirable, as in sighting tasks. The furthest text lookahead in reading right-to-left orthographies is projected directly to the right hemisphere. We suggest that various aspects of right-to-left orthographies match the right hemisphere’s processing propensities and may owe their cultural evolution to this fact: a language tends to settle into orthographic conventions that reflect the direction of reading and the processing predispositions of the cerebral hemispheres.

Acknowledgments

We thank Scott McDonald for the construction of the stimulus materials, experiment infrastructure and data format. The original data collection was funded in part by ESRC (UK) Grants R39195 and R39942.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

All materials and data are openly available: (see https://osf.io/wk89e/?view_only=77337de874ad4189a428b201e2d5095c)

Additional information

Funding

The work was supported by the Economic and Social Research Council [R39942]; Economic and Social Research Council [R39195].