Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2020 Mar 1.
Published in final edited form as: J Exp Psychol Learn Mem Cogn. 2018 Jul 9;45(3):544–551. doi: 10.1037/xlm0000598

Eye-Movement Evidence for the Mental Representation of Strokes in Chinese Characters

Lili Yu 1, Jianping Xiong 2, Qiaoming Zhang 3, Denis Drieghe 4, Erik D Reichle 5
PMCID: PMC6326902  NIHMSID: NIHMS948584  PMID: 29985038

Abstract

Although strokes are the smallest identifiable units in Chinese words, the fact that they are often embedded within larger units (i.e., radicals and/or characters that comprise Chinese words) raises questions about how and even if strokes are separately represented in lexical memory. The present experiment examined these questions using a gaze-contingent boundary paradigm (Rayner, 1975) to manipulate the parafoveal preview of the first of two-character target words. Relative to a normal preview, the removal of whole strokes was more disruptive (i.e., resulting in longer looking times on targets) than the removal of an equivalent amount of visual information (i.e., number of pixels) from strokes located either in similar locations or throughout the entire character. These findings suggest that strokes are represented as discrete functional units rather than visual features or integral parts of the radicals/characters in which they are embedded. We discuss the theoretical implications of this conclusion for models of Chinese word identification.

Keywords: Chinese reading, Strokes, Chinese characters, Eye movements

The compositional role of individual letters in the mental representation of printed words in English and other alphabetic writing systems is widely accepted, as demonstrated by the fact that all existing computational models of word identification make explicit theoretical assumptions about how letters are combined to access other lexical information from memory (e.g., Davis, 2010; Gomez, Ratcliff, & Perea, 2008; Whitney, 2001). What remains less clear, however, is whether or not, in written Chinese, the individual strokes (so named because Chinese was historically written using a brush and ink) that comprise words play a similar role and thus similarly represented in the lexicon. This question arises because, in contrast to alphabetic writing systems, where words are composed from linear arrays of letters, Chinese words are composed of 1–4 characters (see Fig. 1) that are themselves usually composed of 1–36 strokes. These strokes are often also arranged into clusters called radicals that can convey semantic or phonological information, and it is because of this 2-dimensional, hierarchical structuring of Chinese words that empirical efforts to understand how they are represented in memory have yielded mixed results about both the relative importance of strokes (e.g., as compared to radicals) and how they might be represented in the lexicon (for empirical and theoretical reviews, see respectively Yu & Reichle, 2017, and Reichle & Yu, 2017).

Figure 1.

Figure 1

Open in a new tab

An example of the boundary paradigm, the experimental sentences, and the regions of interest. Panel A shows the four possible previews of the first character in the target words prior to a reader’s eyes crossing an invisible boundary. Panel B shows display change resulting in the target being made visible after a reader’s eyes have crossed the boundary. (The English translation of the example sentence is: His favorite way to relax is to go into the study and look at his stamp collection, with the underlined word being the two-character target word in the example sentence.)

For example, early studies examined the role of strokes by manipulating their number, with the rationale being that, if the number of strokes influences how characters are recognized, then strokes are likely represented in the mental lexicon. Although such effects of stroke number have been reliably found in character-identification tasks, the number of strokes is often confounded with both the number of radicals (i.e., characters with many strokes often contain multiple radicals; Liu, Shu, & Li, 2007) and other factors (e.g., degree of visual crowding of strokes within characters; Yu, Zhang, Priest, Reichle, & Sheridan, 2018).

Another method that has been used to demonstrate the mental representation of strokes comes from an eye-movement experiment reported by Yan et al. (2012; see also Tseng, Chang, & Wang, 1965). In this experiment, 15%, 30%, or 50% of the strokes were removed from target characters, with the removed strokes being those that: (1) are normally written early when a character is written1; (2) are normally written later in a character; and (3) do not contribute to the overall “envelop” or shape of a character. The key finding was that, with 30% or more of the strokes being removed, the removal of the early strokes was more disruptive than the removal of the later strokes, and with the removal of shape-preserving strokes being the least disruptive. (Flores d’Arcais, 1994, observed a similar processing advantage for early compared to later strokes in character-identification tasks.) This suggests that strokes are represented in the lexicon, but perhaps to varying degrees, with earlier-written strokes being more important than internal or later-written strokes—similar to what has been shown with English, where the initial letters of a word are more important for word identification than internal or ending letters (Rayner, White, Johnson & Liversedge, 2006; White, Johnson, Liversedge, & Rayner, 2008).

This conclusion, however, has been questioned on the basis of counter evidence. For example, three studies have examined the possible differential weighting of information in the representation of Chinese characters by removing fragments from characters rather than individual strokes (e.g., removing “ Inline graphic” or “ Inline graphic” from the character “ Inline graphic”; Liu, 1983; Peng, 1982; Tsao & Wang, 1983). These studies collectively show that removing fragments from the left and/or upper parts of characters is more disruptive to their identification than removing fragments from the right and/or lower parts. Expanding upon this finding using an eye-movement study, Wang et al. (2013) used a visual-redundancy metric to remove character segments of varying levels of informativeness. Perhaps not surprisingly, reading was more disrupted by the removal of informative (non-redundant) segments, and these informative segments tended to be located on the left side of the characters. Possible explanations for these findings include the fact that, when reading from left to right, the left sides of characters are closer to the high-acuity center of vision and subject to less lateral interference from proximal strokes, and/or the upper-left quadrants/left sides of characters tend to have greater composability (i.e., are consistent with more possible characters; Peng, 1982).

Given these incongruent findings, the main objective of the current experiment is to demonstrate the psychological reality of strokes as functional units in the identification of Chinese characters by directly comparing the effects of removing strokes versus informative fragments/segments from within the same characters. In other words, the current experiment is intended to show that strokes are processed and represented as discrete, separable units rather than as sets of visual features or integral components of the radicals or characters in which they are embedded.

To do this, we used a gaze-contingent boundary paradigm (Rayner, 1975) where some type of preview of a target word is immediately replaced by the target as reader’s eyes move across an invisible “boundary” located to the left of the target. In our variant of this paradigm, we manipulated the preview of the first character in 2-character target words (see Fig. 1). As Figure 2 shows, the preview was one of four types: (1) identical preview, the actual target character, providing a baseline against which to measure the possible effects of the other three previews; (2) stroke-removal preview, the target character with some number of the earliest-written strokes removed; (3) fragment-removal preview, the target character with same proportion of pixels removed from strokes in the left side or upper-left quadrant of the character; or (4) segment-removal preview, the target character with the same proportion of pixels removed, but from across all of the strokes in the character. These previews were intended to test two hypotheses. First, as Figure 3 shows, our primary interest was to know if removing strokes per se is more disruptive to reading than removing equivalent amounts of visual information from equally informative parts of the characters; we therefore predicted (H1) longer fixations in the stroke-removal than fragment-removal condition, with the null prediction (H0) being no difference between these two conditions. Of secondary interest was the hypothesis that removing visual information from informative locations would be more disruptive than removing equivalent amounts of visual information from whole characters; previous studies (e.g., Liu, 1983; Peng, 1982; Tsao & Wang, 1983) suggest that the fragment-removal condition should produce longer fixations than the segment-removal condition.

Figure 2.

Figure 2

Open in a new tab

Examples of the four preview conditions: (1) identical preview; (2) stroke-removal preview; (3) fragment-removal preview; and (4) segment-removal preview. The gray portions of the character are for illustrative purposes and correspond to the pixels that were removed to create each of the previews.

Figure 3.

Figure 3

Open in a new tab

The predictions for the four preview conditions.

Method

Participants

Sixty-eight undergraduates with normal or corrected-to-normal vision from Henan Normal University took part in the experiment, which was approved by Institute of Education, Henan Normal University (Data from an additional 11 participants were collected but excluded from our analyses because they noticed the boundary changes four or more times.)

Materials and Design

Sixty-eight single-radical target characters were selected to avoid any potential difficulty with interpreting our results. As illustrated in Figure 1, there were four preview conditions: (1) identical preview; (2) stroke-removal preview, wherein early-written strokes were removed from a character; (3) fragment-removal preview, wherein pixels from the left side or upper-left quadrant were removed; and (4) segment-removal preview, wherein the initial segment of each stroke in a character were removed. The same number of pixels were removed from conditions 2–4 (assessed by Adobe Photoshop CS6) and the overall proportion of pixels removed from each character was ~30% to maximize the disruption caused by the manipulation (as per Yan et al., 2012). To render the stroke- and fragment-removal conditions maximally dissimilar, the stroke fragments that were removed in the latter condition were selected from the left side or upper-left quadrant of the character, subject to the constraint that they were not the same (early written) strokes removed from the former condition. (This was verified using the RadicalLocator software2; see Yu, Reichle, Jones, & Liversedge, 2015).

Sixty-four extra participants who did not participate in the eye-tracking experiment indicated how many characters could be generated from each of the previews in conditions 2–4, with each participant experiencing all conditions but a given character only once using a Latin-square design; on average, the original character was accurately reported 90% of the time, and 1.1 characters being generated across conditions [F (2, 203) = 0.30, p = 0.74].

All target characters were the initial characters of 2-character target words, which were embedded near the middle of their sentence frames. Target plausibility and sentence naturalness ratings were collected from 15 subjects who did not participate in the eye-tracking experiment using a 5-point scale (5 = plausible/natural); targets were rated as plausible (M = 4.1 out of 5) and sentences as natural (M = 4.0 out of 5). 15 additional subjects completed cloze norms on the target words; targets were unpredictable, only being guessed on average 1.4% of the time.

The preview conditions were counterbalanced across participants and items using a Latin-square design so that each participant encountered all conditions as often but read each sentence only once.

Apparatus

Participants’ eye movements were recorded by an SR Research EyeLink-1000 desktop eye-tracker with a sampling rate of 1,000 Hz. A chin rest was used to minimize participants’ head movements. The display monitor was a NESO FS210A CRT monitor with a refresh rate of 120 Hz and a screen resolution of 1,024 × 768 pixels. Sentences were displayed as Song font in black (RGB: 0, 0, 0) on a grey background (RGB: 150, 150, 150), with 36 pixels per character and a 1-pixel gap between characters. The distance between the monitor and participant was approximately 75 cm, so that each character occupied ~1° of visual angle. Although viewing was binocular, only the participants’ right eyes were tracked.

Procedure

Participants gave informed consent prior to their participation and were instructed to read sentences for comprehension. A three-point horizontal calibration was used at the beginning of the experiment to ensure eye-tracking accuracy, with a maximum acceptable error of 0.4° (Mean < 0.2°). Prior to each trial, a drift-calibration dot appearing at the location of first character in the sentences, which was followed by a gaze-contingent trigger that participants needed to fixate for 20 ms to display the sentence. The participant then read the sentence at his/her own pace and pressed the space-key when done.

Each participant read 138 sentences consisting of 10 practice sentences, followed by 68 experimental sentences randomly interspersed with 60 filler sentences. 35% of the sentences were followed by yes/no comprehensive questions. Participants answered 89% of these questions correctly, indicating that they comprehended the sentences.

Results

Linear Mixed Models (LMMs) using the lme4 package (version 1.1-12, Bates et al., 2015) in R (R Core Team 2016) were used to analyze the data. Two sets of contrasts were used in the analysis: (1) a treatment contrast with the identical-preview being treated as a baseline and three planned contrasts comparing it with each of the other preview conditions; (2) a sliding contrast comparing (a) segment- versus fragment-removal previews and (b) fragment- versus stroke-removal previews to respectively assess the disruption caused by (a) removing strokes versus pixels and (b) removing left-side/upper-left quadrant pixels versus pixels from across the characters. Both participants and items were entered in the models as random effects, with the maximum random effects structure specifying intercepts and slopes for preview effects across participants and items (Barr, Levy, Scheepers, & Tily, 2013). Absolute t- and z-values equal to or greater than 1.96 indicates significance using α = 0.05.

Four regions of interest were included in our analyses to fully assess the disruptions caused by removing strokes/pixels from characters: the pre-target word, the target character (i.e., the first character of the target word), the target word, and the post-target word. Five dependent measures were calculated for each region: (1) first-fixation duration (FFD), or the duration of the initial first-pass fixation in a region; (2) single-fixation duration (SFD), or the duration of the first-pass fixation in a region that is fixated exactly once; (3) gaze duration (GD), or the sum of all first-pass fixations in a region; (4) go-past time (GP), or the sum of all fixations from the first fixation into a region until the eyes exit the region to the right (i.e., this measure includes any regressive fixations that exit a region to the left); and (5) skipping probability (PrS), the probability of a region being skipped. Although measures 1–3 are posited to reflect the early processing of a region, measure 4 may include fixations that drift back to prior characters due to the small visual angle (~1°) of each character, and measure 5 likely reflects the parafoveal pre-processing of up-coming characters (Rayner, 2009).

Prior to completing the analyses, all fixations shorter than 60 ms or longer than 600 ms were removed (4.0% of the data), as were fixations more than three standard deviations above the mean per participant for each measure, resulting in the loss of an additional 0.1%-2.1% of the data across measures. We also removed trials in which participants blinked while fixating a critical region (4.0% of trials), the display change was triggered prematurely (5.6% of trials) or required more than 10 ms to complete (5.3% of trials), or a saccade “hooked back” to the pre-target word after triggering the display change (3.4% trials). These exclusions left 3,779 trails.

Because the log-transformed data yielded a similar pattern of results as the untransformed data, the latter are reported for transparency. The means and standard deviations are shown in Table 1, and the statistical models for target characters and words are shown in Table 2.

Table 1.

Mean fixation times and skipping probabilities, as a function of preview types for all regions of interest. (Standard deviations are in parentheses.)

Region Dependent Measure Identical Preview Stroke-Removal Preview Fragment-Removal Preview Segment-Removal Preview
Pre-Target Word FFD 241 (80) 237 (74) 243 (78) 233 (76)
SFD 238 (79) 235 (74) 240 (77) 231 (77)
GD 273 (119) 267 (110) 271 (117) 256 (107)
GP 309 (175) 306 (175) 305 (174) 295 (167)
PrS 0.23 (0.42) 0.22 (0.42) 0.22 (0.42) 0.22 (0.41)
Target Character FFD 244 (82) 254 (87) 242 (82) 247 (81)
SFD 243 (82) 255 (89) 243 (83) 250 (81)
GD 249 (91) 264 (96) 256 (98) 258 (91)
GP 292 (180) 348 (220) 329 (221) 326 (224)
PrS 0.59 (0.49) 0.59 (0.49) 0.59 (0.49) 0.55 (0.50)
Target Word FFD 249 (86) 254 (87) 247 (85) 249 (84)
SFD 250 (87) 256 (89) 247 (83) 251 (84)
GD 295 (142) 308 (143) 302 (145) 299 (141)
GP 356 (235) 386 (240) 376 (252) 371 (254)
PrS 0.23 (0.42) 0.18 (0.38) 0.20 (0.40) 0.18 (0.38)
Post-Target Word FFD 235 (82) 239 (81) 238 (82) 234 (78)
SFD 234 (81) 238 (79) 235 (81) 232 (76)
GD 273 (132) 273 (124) 275 (133) 265 (131)
GP 353 (276) 344 (315) 369 (308) 365 (331)
PrS 0.26 (0.44) 0.26 (0.44) 0.26 (0.44) 0.30 (0.46)
Open in a new tab

Table 2.

LMM fixed-effect estimates for all measures and preview conditions on target characters and words.

Dependent Measure Contrast Target Character Target Word
b SE |t| b SE |t|
FFD Intercept 243.13 5.89 41.26 246.97 5.59 44.15
Stroke - Identical 7.55 5.60 1.35 3.68 4.47 0.82
Fragment - Identical −4.49 5.58 0.80 −3.77 4.63 0.81
Segment - Identical 0.74 5.51 0.13 −0.50 4.56 0.11
Intercept 244.50 5.20 47.02 246.84 4.92 50.15
Stroke - Fragment 10.91 5.94 1.84 6.68 3.93 1.70
Fragment - Segment −5.75 5.98 0.96 −2.65 3.94 0.67
SFD Intercept 242.27 6.07 39.94 248.67 5.89 42.25
Stroke - Identical 9.77 5.76 1.70 5.57 5.06 1.10
Fragment - Identical −2.36 5.74 0.41 −2.91 4.70 0.62
Segment - Identical 4.25 5.64 0.75 0.44 4.65 0.09
Intercept 246.10 5.43 45.31 249.87 5.33 46.90
Stroke - Fragment 10.06 5.75 1.75 8.51 4.52 1.88
Fragment - Segment −6.88 5.58 1.23 −3.86 4.56 0.85
GD Intercept 246.67 6.69 36.85 288.99 9.71 29.75
Stroke - Identical 12.08 6.31 1.91 11.00 8.44 1.30
Fragment - Identical 3.59 6.28 0.57 4.24 7.30 0.58
Segment - Identical 6.14 6.19 0.99 3.82 7.85 0.49
Intercept 253.65 5.98 42.40 295.14 9.48 31.13
Stroke – Fragment 8.37 6.73 1.24 7.28 7.78 0.94
Fragment - Segment −1.59 6.45 0.25 1.06 6.88 0.15
GP Intercept 293.16 14.04 20.88 353.75 14.32 24.70
Stroke - Identical 49.82 14.80 3.37 26.53 11.91 2.23
Fragment - Identical 26.89 14.95 1.80 11.80 12.06 0.98
Segment - Identical 22.90 14.82 1.55 11.57 13.68 0.85
Intercept 330.49 13.68 24.17 376.52 15.34 24.55
Stroke – Fragment 39.08 17.64 2.22 15.39 12.34 1.25
Fragment - Segment 4.27 15.82 0.27 2.53 12.34 0.21
SKP Intercept 0.39 0.11 3.73 −1.53 0.16 9.40
Stroke - Identical 0.01 0.10 0.06 −0.38 0.13 3.05
Fragment - Identical 0.02 0.10 0.18 −0.22 0.12 1.81
Segment - Identical −0.17 0.10 1.75 −0.40 0.13 3.22
Intercept 0.35 0.09 3.73 −1.87 0.15 12.51
Stroke - Fragment −0.02 0.11 0.15 −0.16 0.13 1.27
Fragment - Segment 0.19 0.10 1.97 0.17 0.13 1.35
Open in a new tab

Note: Significant or marginally significant terms are presented in bold.

Pre-target Words

On the pre-target words, first fixations (b = −8.29, SE = 4.00, t = 2.08) and gaze durations (b = −16.29, SE = 5.72, t = 2.85) were shorter with the segment-removal than identical preview (Fig. 4A), producing a parafovea-on-fovea effect in which orthographic properties of a parafoveal word modulated fixations on a foveal (i.e., fixated) word. This pattern was not observed with the stroke- or fragment-removal previews, suggesting that the finding with the segment-removal condition was anomalous; although not predicted, this result is consistent with the hypothesis that irregular orthographic patterns “pop out,” rapidly drawing the eyes away from the pre-target region (Hyönä, 1995; Plummer & Rayner, 2012)3.

Figure 4.

Figure 4

Open in a new tab

Mean first-fixation and gaze durations (with standard errors of the means) as a function of preview for (A) pre-target and (B) target words.

Target Character

Consistent with our “pop out” account of the segment-removal preview, target characters in this condition were skipped less often than characters in the identical or fragment-removal conditions (marginally significant: b = −0.17, SE = 0.10, t = 1.75; b = 0.19, SE = 0.10, t = 1.97, respectively).

None of the three preview conditions differed from the identical preview for first-fixation or single-fixation durations (all ts < 1.70). However, the stroke-removal preview was more disruptive than the fragment-removal preview, resulting in marginally longer first- (b = 10.91, SE = 5.94, t = 1.84) and single-fixation durations (b = 10.06, SE = 5.75, t = 1.75). This pattern was not significant with gaze durations (t = 1.24) but was evident in go-past times (b = 39.08, SE = 17.64, t = 2.22), with the overall pattern thus being consistent with the hypothesis that strokes are important functional units, and that their removal is more problematic for lexical processing than the removal of comparable amounts of visual information from strokes in similar locations (i.e., in the left side and/or upper-left quadrant of characters). Also consistent with this hypothesis is that both gaze durations (marginally significant, b = 12.08, SE = 6.31, t = 1.91) and go-past times (b = 49.38, SE = 14.82, t = 3.33) were longer for the stroke-removal than identical preview.

Finally, go-past times were also marginally longer in the fragment-removal than identical preview (b = 26.89, SE = 14.95, t = 1.80), providing evidence for removing fragments from left side and/or upper-left quadrant of characters hinders their identification. No other differences were evident between the segment-removal and either the identical preview or fragment-removal preview for any measure (all ts < 1.55).

Target Word

The fixation-duration measures on this two-character region exhibited a similar pattern as on the target characters (Fig.4B), with marginally inflated first and single-fixations durations in the stroke-removal relative to fragment-removal previews (FFD: b = 6.68, SE = 3.93, t = 1.70; SFD: b = 8.51, SE = 4.52, t = 1.88), and longer go-past times in the stroke-removal relative to identical previews (b = 26.53, SE = 11.91, t = 2.23). However, the skipping rate was higher for identical previews than for stroke-removal previews (b = −0.38, SE = 0.13, t = 3.05), segment-removal previews (b = −0.40, SE = 0.13, t = 3.22), and fragment-removal previews (marginally significant: b = −0.22, SE = 0.12, t = 1.81).4

Post-Target Word

No effects were observed on the post-target words expect an elevated skipping rate following the segment-removal than identical preview (b = 0.22, SE = 0.11, t = 2.01). We suspect that this finding may reflect the slight reduction of target-character skipping in this condition—that it may have afforded more parafoveal preview and thus more skipping of the post-target word.

Discussion

In the present experiment, a boundary paradigm was used to examine the processing costs associated with removing individual strokes from characters, fragments of strokes from specific within-character locations, and segments of all within-character strokes. The first hypothesis addressed by this experiment concerns the basic representation of strokes in lexical memory: If strokes are represented in the lexicon, then the processing costs associated with their removal from preview should be greater than the processing costs associated with the removal of the same amount of information from fragments in similar locations (see Fig. 3). Consistent with the hypothesis, the stroke-removal preview caused more disruption to target characters/words processing than did the fragment-removal preview, as indexed by both longer first- and single-fixations durations in the former than latter condition, and by longer gaze durations and go-past times for the stroke-removal preview compared to the identical-preview condition.

However, prior analyses (see Wang et al., 2013; Yan et al., 2012) have suggested that the contour of the incomplete characters is highly correlated to how readily they can be recognized because characters with missing informative segments also have the lowest proportion of overlapping perimeters and vertices to their original characters. Therefore, to ensure that the observed disruption in the stroke-removal condition was actually due to the cost associated with the missing strokes per se, and not differences in the contours of the stroke-removal versus fragment-removal previews, we compared the convex perimeters (i.e., the perimeter connecting all of the vertices) of the stroke-removal versus fragment-removal previews; a paired t-test indicated there was no difference between the two conditions [t (134) = 1.19, p = 0.23], consistent with our claim that the observed disruption was caused by removing representational units (i.e., of strokes) that are important for lexical processing rather than by, for example, reducing the informativeness of the remaining information.

Our experiment also examined the possible processing costs associated with different within-character locations by comparing the disruption caused by removing information from the left sides and upper-left quadrants of characters (i.e., fragment-removal preview) with the disruption caused by removing information from across whole characters (i.e., segment-removal preview). Unfortunately, our conclusions here are more equivocal because of the anomaly (i.e., reduced fixation durations) observed on the pre-target words in the segment-removal condition. In other words, the observed null effect between the fragment- versus segment-removal previews on target characters/words may reflect a “trade-off” in which shorter pre-target fixations in the segment-removal condition afford less target-word preview, thereby inflating fixations on the target words in this condition.

Finally, it is worth noting that the current findings have theoretical implications for existing and future accounts of Chinese word identification (see Reichle & Yu, 2017). Foremost, our results are more consistent with models that postulate the representation of strokes as discrete functional units (e.g., Li, Rayner, & Cave, 2009) than models that do not (e.g., Perfetti, Liu, & Tan, 2005). And this is not to say that strokes are represented in the lexicon as perceptual features; they instead function—at least in the context of single-radical characters—as orthographic units, similar to letters in alphabetic writing systems. Similarly, our results suggest that models require assumptions about the strokes being weighted differentially during processing, although whether this is due to visual-acuity limits (e.g., Li et al., 2009), the serial allocation of attention (e.g., Taft, Zhu, & Peng, 1999), and/or other factors (e.g., the order in which strokes are normally written) remains unclear. Finally, considerable effort has been directed towards explaining how, in alphabetic language, the order of letters is encoded and represented in the lexicon (e.g., Davis, 2010). We suspect that written Chinese will provide an informative arena for evaluating these competing hypotheses because of the inherent complexity of Chinese words (e.g., the fact that strokes are arranged along two spatial dimensions as compared to one for letters in alphabetic writing systems).

Acknowledgments

This work was supported by a National Institute of Health grant RO1HD075800 awarded to the last author and a National Social Science Fund of China grant NSSFC16BYY071 awarded to the second author.

Footnotes

1

This is possible because the strokes in Chinese characters are typically written in a fixed order.

2

Radical Locator was used to calculate visual-similarity scores between the stroke-removal previews and two possible types of fragment-removal previews: previews generated by removing information from the left side vs. upper-left quadrant of target characters. The fragment-removal preview that was most dissimilar to the stroke-removal preview was then selected for use in the experiment.

3

The sliding contrast also revealed shorter first-fixation durations (b = 7.63, SE = 3.80, t = 2.01) and gaze durations (b = 11.60, SE = 5.17, t = 2.25) for the segment- than fragment-removal previews, again likely because of the “pop out” effect associated with the segment-removal preview.

4

We also examined whether the removal of information from two character locations (i.e., left side vs. upper-left quadrant) in the fragment-removal condition differentially affected our findings. Analyses based on the target-word data (~1,000 trails for each location) suggested stroke removal was more disruptive to overall processing than the removal of fragments from either character location.

References

  1. Barr DJ, Levy R, Scheepers C, Tily HJ. Random effects structure for confirmatory hypothesis testing: Keep it maximal. Journal of memory and language. 2013;68:255–278. doi: 10.1016/j.jml.2012.11.001. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Bates D, Maechler M, Bolker B, Walker S. Fitting linear mixed-effects models using lme4. Journal of Statistical Software. 2015;67:1–48. doi: 10.18637/jss.v067.i01. [DOI] [Google Scholar]
  3. Davis CJ. The spatial coding model of visual word identification. Psychological review. 2010;117:713–758. doi: 10.1037/a0019738. [DOI] [PubMed] [Google Scholar]
  4. Flores d’Arcais GB. Order of strokes writing as a cue for retrieval in reading Chinese characters. European Journal of Cognitive Psychology. 1994;6:337–355. doi: 10.1080/09541449408406519. [DOI] [Google Scholar]
  5. Gomez P, Ratcliff R, Perea M. The overlap model: a model of letter position coding. Psychological review. 2008;115:577–601. doi: 10.1037/a0012667. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Hyönä J. Do irregular letter combinations attract readers’ attention? Evidence from fixation locations in words. Journal of Experimental Psychology: Human Perception and Performance. 1995;21:1365–1373. doi: 10.1037/0096-1523.21.1.68. [DOI] [Google Scholar]
  7. Li X, Rayner K, Cave KR. On the segmentation of Chinese words during reading. Cognitive Psychology. 2009;58:525–552. doi: 10.1016/j.cogpsych.2009.02.003. [DOI] [PubMed] [Google Scholar]
  8. Liu IM. Cueing function of fragments of Chinese characters in reading. Acta Psychologica Taiwainica. 1983;25:85–90. in Chinese. [Google Scholar]
  9. Liu Y, Shu H, Li P. Word naming and psycholinguistic norms: Chinese. Behavior research methods. 2007;39:192–198. doi: 10.3758/BF03193147. [DOI] [PubMed] [Google Scholar]
  10. Peng RX. A preliminary report on statistical analysis of the structure of Chinese characters. Acta Psychologica Sinica. 1982;14:385–390. in Chinese. [Google Scholar]
  11. Perfetti CA, Liu Y, Tan LH. The Lexical Constituency model: Some implications of research on Chinese for general theories of reading. Psychological Review. 2005;112:43–59. doi: 10.1037/0033-295X.112.1.43. [DOI] [PubMed] [Google Scholar]
  12. Plummer P, Rayner K. Effects of parafoveal word length and orthographic features on initial fixation landing positions in reading. Attention, Perception, & Psychophysics. 2012;74:950–963. doi: 10.3758/s13414-012-0286-z. [DOI] [PubMed] [Google Scholar]
  13. R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing; Vienna, Austria: 2016. https://www.R-project.org/ [Google Scholar]
  14. Rayner K. The perceptual span and peripheral cues in reading. Cognitive Psychology. 1975;7:65–81. doi: 10.1016/0010-0285(75)90005-5. [DOI] [Google Scholar]
  15. Rayner K. The 35th Sir Frederick Bartlett Lecture: Eye movements and attention in reading, scene perception, and visual search. Quarterly journal of experimental psychology. 2009;62:1457–1506. doi: 10.1080/17470210902816461. [DOI] [PubMed] [Google Scholar]
  16. Rayner K, White SJ, Johnson RL, Liversedge SP. Raeding wrods with jubmled lettres there is a cost. Psychological science. 2006;17:192–193. doi: 10.1111/j.1467-9280.2006.01684.x. [DOI] [PubMed] [Google Scholar]
  17. Reichle ED, Yu L. Cognitive Science. Manuscript in Press; 2017. Models of Chinese Reading: Review and Analysis. [DOI] [PubMed] [Google Scholar]
  18. Taft M, Zhu X, Peng D. Positional specificity of radicals in Chinese character recognition. Journal of Memory and Language. 1999;40:498–519. doi: 10.1006/jmla.1998.2625. [DOI] [Google Scholar]
  19. Tsao YC, Wang TG. Information distribution in Chinese characters. Visible Language. 1983;17:357–364. [Google Scholar]
  20. Tseng SC, Chang LH, Wang CC. An informational analysis of the Chinese language: I. The reconstruction of the removed strokes of the ideograms in printed sentence-texts. Acta Psychologica Sinica. 1965;10:299–306. in Chinese. [Google Scholar]
  21. Wang HC, Schotter ER, Angele B, Yang J, Simovici D, Pomplun M, Rayner K. Using singular value decomposition to investigate degraded Chinese character recognition: evidence from eye movements during reading. Journal of research in reading. 2013;36:35–50. doi: 10.1111/j.1467-9817.2013.01558.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. White SJ, Johnson RL, Liversedge SP, Rayner K. Eye movements when reading transposed text: The importance of word beginning letters. Journal of Experimental Psychology: Human Perception and Performance. 2008;34:1261–1276. doi: 10.1037/0096-1523.34.5.1261. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Whitney C. How the brain encodes the order of letters in a printed word: The SERIOL model and selective literature review. Psychonomic Bulletin & Review. 2001;8:221–243. doi: 10.3758/BF03196158. [DOI] [PubMed] [Google Scholar]
  24. Yan G, Bai X, Zang C, Bian Q, Cui L, Qi W, Rayner K, Liversedge SP. Using stroke removal to investigate Chinese character identification during reading: Evidence from eye movements. Reading and Writing. 2012;25:951–979. doi: 10.1007/s11145-011-9295-x. [DOI] [Google Scholar]
  25. Yu L, Reichle ED. Chinese versus English: Insights on Cognition during Reading. Trends in Cognitive Sciences. 2017;21:721–724. doi: 10.1016/j.tics.2017.06.004. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Yu L, Reichle ED, Jones M, Liversedge SP. Radicallocator: A software tool for identifying the radicals in Chinese characters. Behavior Research Methods. 2015;47:826–836. doi: 10.3758/s13428-014-0505-8. [DOI] [PubMed] [Google Scholar]
  27. Yu L, Zhang Q, Priest C, Reichle ED, Sheridan H. Character-complexity effects in Chinese reading and visual search: A comparison and theoretical implications. Quarterly Journal of Experimental Psychology. 2018;71:140–151. doi: 10.1080/17470218.2016.1272616. [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES