Reading direction and the central perceptual span: Evidence from Arabic and English

In English and other alphabetic languages read from left to right, useful information acquired during each fixational pause is generally reported to extend much further to the right of each fixation than to the left. However, the asymmetry of the perceptual span for alphabetic languages read in the opposite direction (i.e., from right to left) has received very little attention in empirical research. Accordingly, we investigated the perceptual span for Arabic, which is one of the world’s most widely read languages and is read from right to left, using a gaze-contingent window paradigm in which a region of text was displayed normally around each point of fixation, while text outside this region was obscured. Skilled Arabic readers who were bilingual in Arabic and English read Arabic and English sentences while a window of normal text extended symmetrically 0.5o to the left and right of fixation or asymmetrically, by increasing this window to 1.5o or 2.5o to either the left or the right. When English was read, performance across window conditions was superior when windows extended rightward. However, when Arabic was read, performance was superior when windows extended leftward and was essentially the reverse of that observed for English. These findings show for the first time that a leftward asymmetry in the central perceptual span occurs when Arabic is read and, for the first time in over 30 years, provide a new indication that the perceptual span for alphabetic languages is modified by the overall direction of reading.

Normal reading relies on making saccadic eye movements along lines of text and ending each movement with a brief fixational pause during which information is acquired. However, the extent of the area at each fixational pause within which information influences reading (the perceptual span) is limited. In particular, estimates of the perceptual span have often been reported using an eye-tracking technique in which an area of text extending outward from each fixation is displayed normally during reading but text beyond this area is obscured (e.g., by replacing the original letters with xs). Using this technique, several studies have shown that skilled readers of English and other alphabetic systems read from left to right obtain useful information from an area extending 14-15 character spaces to the right of fixation but no more than 3-4 character spaces to the left, and certainly no further than the beginning of the fixated word (e.g., McConkie & Rayner, 1975, 1976Rayner, Well, & Pollatsek, 1980;Underwood & McConkie, 1985). Consequently, it is generally accepted that, when reading from left to right, the perceptual span extends much further to the right of fixation than to the left.
The nature of the perceptual span is viewed widely as reflecting the written characteristics of the language being read (e.g., Ikeda & Saida, 1978;Inhoff & Liu, 1998). However, while the parameters of the perceptual span in English and other alphabetic languages read from left to right have been investigated often (for a review, see Rayner, 2009), the perceptual span for languages read from right to left has received little attention. Indeed, only one study has specifically addressed this issue (in Hebrew), and here the focus was the direction of the span's overall extent (Pollatsek, Bolozky, Well, & Rayner, 1981). In this study, native Israeli readers who were bilingual in Hebrew and English read sentences in which a window of normal text extended asymmetrically either 14 characters to the left of fixation and 4 characters to the right or 4 characters to the left of fixation and 14 characters to the right. Reading performance for Hebrew was superior when windows extended more to the left, whereas performance for English was superior when windows extended more to the right, suggesting that the overall direction of reading modified the asymmetry of the perceptual span. But this seminal study stands alone in its investigation of the perceptual span in languages read from right to left. Consequently, given the importance of the perceptual span for reading and for guiding the implementation of computational models (e.g., Engbert, Nuthmann, Richter, & Kliegl, 2005;Reichle, Rayner, & Pollatsek, 2003), examining the nature of the perceptual span in other languages read from right to left is crucial for developing a full understanding of the processes underlying reading.
Arabic has a global written usage second only to writing systems using the Latinate alphabet and so represents an ideal language for developing a full understanding of the perceptual span that exists for languages read from right to left. However, an examination of the literature shows that no information about the perceptual span for Arabic has ever been published. Accordingly, and following the approach of Pollatsek et al. (1981), we investigated the perceptual span of bilingual readers of Arabic and English, using the window technique to control the amount of information that was available to the left and right of fixation during each fixational pause when Arabic and English text was read.
Rather than investigating the overall extent of the perceptual span, the focus of the present study was the influence of information within an area extending 2.5 o either side of fixation. The reason for this approach is that the perceptual span encompasses a range of different types of information (e.g., interword spaces, word shape, letter identities; see Rayner, 2009), broadly reflecting retinal eccentricity. Indeed, previous research using English suggests that letter identification during reading extends to only eight or nine characters to the right of fixation (equal to approximately 2.5 o of visual angle under normal reading conditions) and just four characters (equal to approximately 1 o ) to the left (e.g., Häikiö, Bertram, Hyönä, & Niemi, 2009;Underwood & McConkie, 1985). But asymmetry in this area (which we shall call the central perceptual span) appears to be particularly influential for reading because it provides information required to identify the fixated word and, crucially, important information about the next word along that aids parafoveal preprocessing. Consequently, given its importance, the central perceptual span is well suited to reveal contrasting directional asymmetries when Arabic and English are read, and so the present study investigated the influence of this area, extending 2.5 o either side of fixation.
If asymmetry in the central perceptual span is determined primarily by reading direction, Arabic should show a directional asymmetry that is essentially the opposite to that for English. However, the situation may be more complex than this. In particular, like English words, Arabic words presented to the left and right of fixation project directly to each contralateral hemisphere and, as with English words, rightsided presentations produce perceptual superiority due to the left-hemisphere dominance for language that is typical for humans (e.g., Almabruk, Paterson, McGowan, & Jordan, 2011;Ibrahim & Eviatar, 2009. Consequently, while an asymmetrical perceptual span to the right of fixation when English is read may facilitate reading because of greater use of information projected to the dominant left hemisphere, a reversal of this asymmetry when Arabic is read would involve greater use of information projected to the right (nondominant) hemisphere, and this may be highly disadvantageous (for discussions, see Almabruk et al., 2011). This possibility is strengthened by studies of hemispheric asymmetry that suggest that the visual and linguistic characteristics of Arabic make it highly unsuited to right-hemisphere processing and even that the right hemisphere plays no part at all in Arabic word recognition (Eviatar, Ibrahim, & Ganayim, 2004;Ibrahim & Eviatar, 2009; see also Jordan, Paterson, & Almabruk, 2010). Moreover, some researchers have argued that the processing of words falling within central vision is influenced greatly by hemispheric asymmetries (for a critique, see Jordan & Paterson, 2009) and, indeed, that this influence is closely related to reading direction because of disruptive righthemispheric projections (e.g., Brysbaert & Nazir, 2005;Brysbaert, Vitu, & Schroyens, 1996). Consequently, it may be the case that reading Arabic is especially disadvantaged by leftward asymmetry in the central perceptual span, and so a leftward asymmetry may be entirely absent for this language.
As a result, the influence of hemispheric dominance and reading environment provide different predictions for asymmetry in the central perceptual span in Arabic. Specifically, if the central perceptual span is determined primarily by hemispheric projections, both Arabic and English should produce a rightward asymmetry. In contrast, if the central perceptual span is determined primarily by reading environment, the central perceptual span should show a rightward asymmetry for English but a leftward asymmetry for Arabic. If this shift in asymmetry were obtained, it would show for the first time that a leftward asymmetry occurs for Arabic and so provide a novel indication that the asymmetry of the central perceptual span is modified by the overall direction of reading.
To investigate this issue, we used a version of the moving window technique that differed from that used in previous research (e.g., McConkie & Rayner, 1975, 1976Pollatsek et al., 1981). Since its inception, the general principle of the technique has been that text within a specified window around the point of fixation is shown normally and text outside this window is obscured. In previous research, obscuring text has typically been achieved using letter replacements to change letter identities while preserving other characteristics of the text. However, while this approach is suitable for manipulating the perceptual span in printed English, where each individual letter is spatially separate and distinct from its neighbors and has a constant width within a typeface, the same approach is not suitable for printed Arabic, which is a cursive script and where the same letters have variable widths and shapes depending on their location within words (e.g., Boudelaa & Marslen-Wilson, 2001). As a result, letter replacements would perturb the normal format of Arabic text and produce changes that were artifactually different from those produced for English. Consequently, the present study avoided these problems by using visual filters to obscure the visibility of letters outside each specified window, while leaving the original letter content of both languages unchanged (see Fig. 1). Previously, this approach has been used successfully for investigating the overall perceptual span in English (Paterson, McGowan, & Jordan, 2013), and preliminary investigations have shown that it is ideal for revealing differences, should they occur, between the central perceptual span used for reading Arabic and English.

Participants
Participants were 12 native Arabic speakers from the University of Leicester who had English as a second language at the level of proficiency required for postgraduate study. All participants were right-handed, demonstrated right-hemifield dominance for Arabic and English words, and had normal or corrected-to-normal visual acuity, as determined by a Bailey-Lovie Eye Chart (Bailey & Lovie, 1980).

Design and materials
Stimuli were 120 Arabic sentences and 120 matched English sentences that were close paraphrases of the Arabic sentences. Across languages, sentences were matched for numbers of words (Arabic, mean = 11; English, mean = 11) and characters (Arabic, mean = 57, English, mean = 59). Each sentence was shown entirely as normal or with normal text only within a window around the point of fixation. Five types of window were used: one symmetrical, two extending leftward, and two extending rightward. The symmetrical window was used to provide a baseline condition and extended just 0.5 o each side of fixation (.5_.5). The two leftward windows extended 1.5 o to the left and 0.5 o to the right (1.5_.5) and 2.5 o to the left and 0.5 o to the right (2.5_.5). In a similar vein, the two rightward windows extended 0.5 o to the left and 1.5 o to the right (.5_1.5) and 0.5 o to the left and 2.5 o to the right (.5_2.5). Text outside each window was obscured by using MATLAB to leave only spatial frequency content with a peak frequency of 2.20 cycles per degree (cpd) and low-pass and high-pass cutoff frequencies of 1.65-3.30 cpd (see Patching & Jordan, 2005a, 2005b. Custom software ensured that each window moved in close synchrony with eye movements, and display changes were made within 10-12 ms. The phenomenological experience was that each window moved in perfect synchrony with the eyes during reading. Within each language, 2.5°of text encompassed approximately nine letters.
All 240 sentences were randomized and sampled using a Latin square design so that each participant saw 20 different sentences in each of the six display conditions in each language. This ensured that all sentences were shown equally often in each condition in the experiment and that sentences in each language were shown only once to each participant. Arabic and English sentences were presented in separate sessions in a randomized order, and order of sessions was counterbalanced across participants. An additional 12 sentences were used as practice items at the beginning of each session.

Apparatus and procedure
Eye movements were recorded using an Eyelink 2 K eyetracker with a spatial resolution of 0.01°. Sentences were displayed on a 19-in. monitor, and eye position was sampled at 1000 Hz using corneal reflection and pupil tracking. On each trial, participants fixated a location on the right (for Arabic displays) or the left (for English displays) of the screen, and a sentence was then presented with its first letter at the  Fig. 1 Examples of Arabic and English sentences displayed entirely as normal and with text falling within central vision shown as normal while text outside is filtered (see the "Method" section). In these examples, the location of the fixation is denoted by the vertical dashed line. Note that the visual appearance of the filtered text in the figure is approximate, due to restrictions in resolution and print medium fixation location. Participants were instructed to read normally and for comprehension and answered a comprehension question after each sentence. Calibration was checked between trials, and the eyetracker was recalibrated as necessary.

Results
Participants showed high levels of comprehension for Arabic and English (response accuracy was over 90 % for each language). Reading rates (calculated as words per minute) are shown in Fig. 2, and fixation duration (the average length of fixational pauses during reading), number of fixations (the number of these fixational pauses), number of regressive saccades (backward movements in the text), and progressive saccade length are reported in Table 1. For each measure, effects of language (Arabic, English) and display condition (normal plus five window types) were analyzed using a repeated measures analysis of variance, computing error variance over participants (F 1 ) and stimuli (F 2 ). Following each analysis, post hoc comparisons (Bonferroni-corrected t-tests) examined the effects more closely.

Discussion
The findings of this study provide clear evidence that the central perceptual span extends further to the left when Arabic is read and further to the right when English is read and so indicate that the asymmetry for each language is determined primarily by reading direction. Indeed, evidence of directional asymmetry in both languages was apparent not only in reading rates, but also in the duration and number of fixations, indicating that each asymmetry facilitated the acquisition of information from text during each fixational pause. In both languages, therefore, it seems that reading benefits from forward-directed asymmetries that allow upcoming words in the central perceptual span to be viewed during fixations and, when these views are not available (as in the appropriate window conditions of the present study), reading becomes more difficult. Comparable differences in the influence of asymmetries across each language were also observed for regressions, indicating that forward-directed asymmetries in both languages helped avoid the need to recheck text that had already been passed. However, progressive saccade length was not affected by the particular type of windows used, suggesting that changes in the asymmetrical availability of information within the central perceptual span did not influence the targeting of progressive saccades. This distinction between when to move the eyes from fixation and where to move them next when progressing through text is consistent with other research (Rayner, 2009). The directional asymmetry observed for English is consistent with previous indications concerning the perceptual span, but the leftward asymmetry observed for Arabic is especially striking because of the processing disadvantages that right -hemisphere projections are known to provide for this language. In particular, not only is the right hemisphere generally nondominant for language, but evidence also suggests that this hemisphere is particularly unsuitable for processing Arabic words (Eviatar et al., 2004;Ibrahim & Eviatar, 2009. Consequently, since the directional asymmetry observed for Arabic text was clearly beneficial for reading and, presumably, develops from experience with this reading environment, it is reasonable to conclude that this leftward asymmetry provides advantages for reading Arabic that are not outweighed by disadvantages of right-hemisphere projections. Indeed, the existence of a leftward perceptual span for Arabic suggests that disruption produced by right -hemisphere projections may be offset by other factors, and two sources of influence seem likely. The first is a region of central vision around fixation that contains an intermingling of ganglion cells that project bilaterally to both hemispheres (see Jordan & Paterson, 2009). The size of this region is unlikely to extend much more than 1°either side of fixation, but this may nevertheless help ameliorate disadvantageous effects of unilateral right-hemisphere projections. Indeed, even areas away from fixation where information projects unilaterally to just the right hemisphere may produce little or no functional division in processing between the two hemispheres because of interhemispheric interactions (Almabruk et al., 2011;Jordan, Almabruk, McGowan, & Paterson, 2011;Jordan, Fuggetta, Paterson, Kurtev, & Xu, 2011). Accordingly, the claims made by some researchers that a precise split in hemispheric projections occurs at the point of each fixation and exerts substantial effects on reading performance (e.g., Brysbaert & Nazir, 2005;Brysbaert et al., 1996) is not consistent with the beneficial effects of a leftward perceptual span observed for Arabic (it should be noted that many other studies have shown that this "split-fovea theory" lacks empirical support; for detailed reviews, see Jordan & Paterson, 2009. The second source of influence comes from the nature of textual reading. In particular, although lateralized presentations of individual stimuli indicate that the right hemisphere is remarkably poor at processing Arabic words, textual reading provides a very different environment in which forward-directed attentional processes and contextual and semantic cues can help enhance the processing of upcoming words (Rayner, 2009). Consequently, although most of the visual field projects unilaterally to each contralateral hemisphere, the evidence of an influential leftward asymmetry when Arabic is read indicates that the beneficial effects of a leftward perceptual span are not wholly disadvantaged by right-hemisphere projections. 1 In sum, the nature of the perceptual span for alphabetic languages read from right to left has received very little attention and, for many years, the findings of only one study on this topic (Pollatsek et al., 1981) and for only one language (Hebrew) have been available in the literature. The present study now extends this earlier work to show that a leftward asymmetry in the central perceptual span occurs for Arabic and, for the first time in over 30 years, provides a new indication that the perceptual span for alphabetic languages is modified by the overall direction of reading. Consequently, although major writing systems (like Arabic, Hebrew, and English) differ fundamentally in their visual and linguistic complexity, it now seems more likely that a component of reading common to alphabetic languages is the acquisition of information during each fixational pause from an area that extends asymmetrically in the direction of reading. It remains to be seen whether further investigations of the relationship between the reading direction of an alphabetic language and its perceptual span now proceed more quickly than before.
1 It is also worth noting that, for Arabic and English, performance advantages were sometimes observed when windows of normal text extended asymmetrically in the direction opposite to reading (i.e., to the right in Arabic and to the left in English). One possibility is that this effect reflected the bilingual abilities of the participants taking part in our study. However, recent evidence (using English text) from nonbilingual populations suggests that the leftward component of the perceptual span for English is actually substantially larger than is generally assumed (Jordan, McGowan, & Paterson, 2012; see also Apel, Henderson, & Ferreira, 2012;Binder, Pollatsek, & Rayner, 1999;Rayner, Castelhano, & Yang, 2009). Accordingly, given the reciprocal nature of the central perceptual span across Arabic and English indicated by the present study, it may well be the case that reading generally benefits from information acquired from an area of text around the point of fixation that is more symmetrical than previously thought. This may also explain why greatly obscuring text in the direction opposite to reading (1.5_.5 and 2.5_.5 for Arabic and .5_1.5 and .5_2.5 for English) produced fewer regressions than did normal text. Here, when sufficient information was available in the direction of reading to achieve good comprehension, readers appear to have avoided making regressions because the spatial extent of information in that direction normally used for reading was greatly reduced.