Vivian Chang Yu-Tse Chien Hsin-Yu Huang WILLIAM ADDISON DWIGGINS.
researchers.mq.edu.au · Web viewMacquarie University PURE Research Management System. This is the...
Transcript of researchers.mq.edu.au · Web viewMacquarie University PURE Research Management System. This is the...
Macquarie University PURE Research Management System
This is the accepted author manuscript version of an article published as:
Liu, Y., Yu, L., & Reichle, E. D. (2019). The dynamic adjustment of saccades during Chinese reading: Evidence from eye movements and simulations. Journal of Experimental Psychology: Learning, Memory, and Cognition, 45(3), 535–543.
Access to the published version: https://doi.org/10.1037/xlm0000595
© American Psychological Association, 2018. This paper is not the copy of record
and may not exactly replicate the authoritative document published in the APA
journal. Please do not copy or cite without author's permission. The final article is
available, upon publication, at https://doi.org/10.1037/xlm0000595
1
The Dynamic Adjustment of Saccades During Chinese Reading:
Evidence from Eye Movements and Simulations
Yanping Liu, Sun Yat-sen University, China
Lei Yu, Sun Yat-sen University, China
Erik D. Reichle, Macquarie University, Australia
Author Note
Address correspondence to Yanping Liu, Department of Psychology, Guangdong
Provincial Key Laboratory of Social Cognitive Neuroscience and Mental Health, Sun
Yat-sen University, 135 Xingang Xi Rd, Guangzhou, China, 510275; e-mail:
[email protected]. This research was supported by the National Natural
Science Foundation of China (31500890), the Fundamental Research Funds for the
Central Universities (17wkpy64), and the U.S. National Institute of Health
(RO1HD075800).
Word count = 5,000
2
Abstract
This article reports an eye-movement experiment in which participants scanned
continuous sequences of Landolt-Cs for target circles to examine the visual and
oculomotor constraints that might jointly determine where the eyes move in a task that
engages many of the perceptual and motor processes involved in Chinese reading but
without lexical or linguistic processing. The lengths of the saccades entering the
Landolt-C clusters were modulated by the processing difficulty (i.e., gap sizes) of
those clusters. Simulations using implemented versions of default-targeting (Yan,
Kliegl, Richter, Nuthmann, & Shu, 2010) versus dynamic-adjustment (Liu, Reichle, &
Li, 2016) models of saccadic targeting indicated that the latter provided a better
account of our participants’ eye movements, further supporting the hypothesis that
Chinese readers “decide” where to move their eyes by adjusting saccade length in
response to processing difficulty rather than by selecting default saccade targets. We
discuss this hypothesis in relation to both what is known about saccadic targeting
during the reading of English versus Chinese and current models of eye-movement
control in reading.
Keywords: Chinese reading; eye-movement control; Landolt-C paradigm; visual
search.
Running Head: Reading Landolt-C sequences.
3
Humans are remarkably adept at exploiting the environmental regularities to
optimize behavior (Anderson, 1990). It should therefore come as little surprise that, in
complex tasks like reading, we exploit regularity to allow (near) optimal performance
(e.g., see Liu, Reichle, & Gao, 2013; Reichle & Laurent, 2006). For example, because
common words are identified more rapidly than uncommon words, readers spend less
time fixating high- than low-frequency words (e.g., Inhoff & Rayner, 1986; Schilling,
Rayner, & Chumbley, 1998). And similarly, because words can be identified most
rapidly when fixated near their centers (e.g., Rayner & Morrison, 1981), readers (of
most alphabetic languages) tend to move their eyes to locations near the centers of
words (e.g., Rayner, 1979). Such eye-movement behaviors exploit the inherent
regularities of text to support rapid reading while maintaining some overall level of
comprehension. What is less clear, however, is how well such behaviors generalize
across languages and writing systems that exhibit less—or perhaps different—patterns
of regularity. One prime example that has been the focus of much recent research is
Chinese (for a review, see Yu & Reichle, 2017).
Chinese differs significantly from the European languages and writing systems
that have most often been used in experiments to understand reading. For example,
Chinese words are not comprised from letters or clearly demarcated by boundaries,
but instead consist of 1-4 equally sized, box-shaped characters comprised of 1-36
strokes (see Fig. 1) arranged into continuous sequences without spaces or other
identifiers to separate the individual words. Chinese readers must therefore use their
knowledge of the language to somehow segment character strings into their
corresponding words (e.g., Li, Rayner, & Cave, 2009) for their identification and for
deciding when and where to move their eyes during reading.
Because word segmentation/identification is a computationally difficult process,
our understanding of their relation to eye-movement control is incomplete. For
example, although early studies of Chinese reading purportedly showed that character
processing is more important (e.g., has a larger influence on fixation durations) than
word processing during Chinese reading (e.g., Hoosain, 1992), more recent studies
suggest that word processing plays a larger role, with the modulating effects of word
frequency on fixation durations, for example, often overshadowing the smaller
character-frequency effects (e.g., Li, Bicknell, Liu, Wei, & Rayner, 2014; Yan, Tian,
Bai, & Rayner, 2006).
4
Figure 1
And in a similar manner, early studies of Chinese reading suggested that saccade
targeting is possibly character-based, as suggested by the means of fixation-location
distributions being approximately uniform over words (Tsai & McConkie, 2003;
Yang & McConkie, 1999). This hypothesis was partially refuted by a corpus analysis
reported by Yan et al. (2010; see also Yan & Kliegl, 2016) which indicated that the
distributions of fixations on words that were the recipients of only one fixation tended
to be centered near the centers of the words, whereas the distributions of the first
fixations on words that received multiple fixations tended to be centered near the
beginnings of the words. Yan et al. (2010) provided a default-targeting account of this
finding: Chinese readers direct their eyes towards the middle of a segmented word
because it will likely be identified from that viewing location, but direct their eyes
towards the beginning of an unsegmented word because it is less likely to be
identified and thus more likely to require a second fixation. Unfortunately, the
relationship between whether a word-segmentation success and the number of times it
will be fixated is correlational. As demonstrated using simulations in which saccade
length was fixed in length but with some random variability, fixations near the
beginning of a word are more likely than fixations near the center of a word to be
followed by a subsequent fixation on the same word (see also, Li, Liu, & Rayner,
2011).
More recent empirical evidence suggests that, instead of directing their eyes
towards specific target locations, Chinese readers dynamically adjust their saccade
lengths to reflect local difficulty with lexical processing (Liu, Reichle, & Li, 2015,
2016; Liu, Huang, Li, & Gao, 2017). A computational implementation of this
alternative account simulated a number of findings related to saccade targeting,
including how the frequency of and fixation position on wordN influence the length of
the saccade to wordN+1 (Liu, Huang, Gao, & Reichle, 2017; Wei, Li, & Pollatsek,
2013), as well as how the length of this saccade is influenced by both frequency and
predictability of wordN+1 (Liu et al., 2016; Liu, Guo, Yu, & Reichle, 2017).
Importantly, this dynamic-adjustment model provides a better quantitative account of
these findings than an implementation of Yan et al.’s (2010) default-targeting account.
5
The success of the dynamic-adjustment account suggests that, in the absence of
word boundary information, readers switch from moving their eyes towards default
saccade targets (i.e., the centers of parafoveal words) to another mode of maximizing
their reading efficiency—the dynamic adjustment of saccade lengths as a function of
on-going lexical processing difficulty. But is this shift towards dynamic-adjustment of
saccades specific to reading (or perhaps specific to the reading of Chinese), or does it
instead reflect task demands that might be shared by other visual-cognitive tasks?
One paradigm that has been used to dissociate the visual and oculomotor
processes of reading from those that are language specific is the Landolt-C “reading”
task (Williams & Pollatsek, 2007; Williams, Pollatsek, & Reichle, 2014; Vanyukov,
Warren, Wheeler, & Reichle, 2012). In this task, participants are instructed to scan
through linear arrays of Landolt-Cs (i.e., ring-shaped stimuli having missing segments
or gaps of variable size and/or orientations) to search for target stimuli—rings with no
missing gaps. Because these Landolt-C stimuli can be arranged into cluster sequences
resembling “words,” the task has been used to study saccadic targeting under
conditions that resemble the reading of alphabetic languages, and that engages vision,
attention, and memory, but without the various lexical and linguistic processes that are
normally engaged during reading. Experiments using this paradigm have
demonstrated that participants direct their eyes towards the centers of Landolt-C
clusters and spend less time fixating on those that are easy to process (e.g., contain
larger gaps and/or occur more often during the course of the experiment). Such
findings again demonstrate that humans exploit regularities to optimize task
performance, and that aspects of reading-like eye-movement behavior (e.g., targeting
the centers of clusters) reflect general rather than reading-specific task demands.
What remains unclear, however, is whether the dynamic adjustment of saccade length
during Chinese reading (e.g., see Liu, Reichle, & Li, 2016) also reflects general rather
than reading-specific task demands.
This article will report the results of an eye-movement experiment in which a
variant of the basic Landolt-C paradigm (adapted to resemble Chinese reading; see
Fig. 2) was used to examine how the absence of clear word boundaries affects
saccade targeting. The main objective was to determine if, as is the case with the
actual reading of Chinese, the absence of “word” boundaries causes participants to
dynamically adjust their saccade lengths rather than moving their eyes to default
6
targets (e.g., the centers of upcoming Landolt-C clusters). To discriminate between
these two possibilities, we also report the results of two simulations using
implementations of the dynamic-adjustment versus default-targeting models.
Experimental Method
Participants
Thirty native Chinese-speaking students recruited from Sun Yat-sen University
were paid for their participation. All participants had normal or corrected-to-
normal vision and were naïve about the purpose of the experiment.
Apparatus
Stimuli were displayed on a 27-inch LED monitor (ASUS, PG27AQ) with a
resolution of 2,560 × 1,440 pixels and a 144-Hz refresh rate. Stimulus presentation
was controlled using an OpenGL-based Psychophysics Toolbox 3 with EyeLink
Toolbox extensions in Matlab. Eye movements were recorded using a SR-Research
Ltd. Eyelink1000 eye tracker (1,000-Hz sampling rate) using a tower setup with
forehead and chin rests to minimize noise due to head movements. Viewing was
binocular but only the right eye was recorded.
Stimuli and Design
The stimuli were configured to resemble Chinese sentences (cf., Figs. 1 vs. 2),
with each “character” being a Landolt-C (36 pixels × 36 pixels, with a 1°, 2°, 3°, or 4°
gap in the left, right, top, or bottom of the character) and each “word” being a cluster
of 1, 2, 3, or 4 characters. The space between any two successive characters was four
pixels. Sixteen unique 2-character exemplar words were generated, with each having a
unique combination of gap size and orientation, and each being repeated 50 times
across the experiment. 1,904 other words that appeared only once across the
experiment were generated using unique combinations of character number per cluster,
gap sizes, and gap orientations, and by allowing small random permutations of gap
angle. Eight words were randomly assigned to each “sentence” string to generate 330
trials, with each sentence containing 0, 1, or 2 targets (i.e., characters without gaps)
7
that could appear with equal probability within any word except the exemplars or the
first or last word within a sentence.
Figure 2
Procedure
Upon arriving, participants were given task instructions, gave informed consent,
and then seated 58 cm from the monitor (so that one character subtended ∼1° of the
visual angle). The eye-tracker was calibrated and validated using a 3-dot procedure at
the beginning of the experiment, with additional calibrations and validations being
conducted as necessary. A drift-check procedure was performed before each trial; a
sentence was displayed after participants successfully fixated a white box (1° × 1°)
located at the position of the first character in the sentence. Participants were
instructed to scan through each sentence, indicating the number of targets in each
using response buttons on a Microsoft SideWinder Game Pad. Participants completed
eight practice trials (not included in our analyses) and then completed the remaining
experimental trials in a random order.
Experimental Results
Accuracy
The mean overall target-number response accuracy was 0.86 (SD = 0.106).
Eye-Movement Results
Our primary analysis focuses on the two-character exemplars because
approximately 72% of Chinese words are composed of two characters. Fixations on
exemplars in which blinks occurred and fixations immediately preceding/following
clusters containing targets were first removed from our analyses, leaving 82.62% of
the total fixations on exemplars. (This precaution was necessary because fixations
near/on targets are not representative of fixations elsewhere; e.g., because of the
requirement to keep track of the number of targets.) To determine how cluster
processing affected fixation durations, our analyses examined how properties of the
fixated exemplar clusters and the spatially adjacent clusters influenced: (1)
8
first-fixation duration (FFD), or the duration of the first fixation on the exemplar
during first-pass scanning; (2) gaze duration (GD), or the sum of all first-pass fixation
durations on the exemplar; (3) total-viewing time (TT), or the sum of all fixation
durations on the exemplar. To determine how parafovea cluster processing affected
saccade targeting, our analyses also examined how exemplar properties influenced
saccade length into the exemplars using: (4) progressive-saccade length (PSL), or the
length of the forward saccade launched from a 2-character pre-exemplar region during
first-pass scanning; and (5) incoming-saccade length (ISL), or the length of the
forward saccade launched from a 2-character pre-exemplar region during first-pass
scanning conditional upon resulting fixation being on the exemplar. These two
analyses were restricted to saccades launched from a 2-character pre-exemplar region
to ensure that the exemplars received some amount of parafoveal processing.
Linear mixed-effects models were built for each measure, using parsimonious
random structure by iterative reduction of insignificant variance and covariance
components from maximal models (see Bates, Kliegl, Vasishth, & Baayen, 2015).
For fixation-duration measures on exemplars, exemplar gap size and repetition
number, as well as the gap size and length of spatially adjacent clusters were
included as predictors. For saccade-length into exemplars, to simplify, only exemplar
gap size and repetition number were included as predictors. The models were then
fitted using the lme4 package (ver. 1.1-13) and p-values were estimated using
lmerTest package (ver. 2.0-33) in R (ver. 3.4.1).
Fixation Durations. Tables 1-3 indicate that fixation-duration measures decreased
with increasing gap size of the fixated exemplar (all ps < 0.001), the increasing gap
size of subsequent cluster (all ps ≤ 0.001), and—with one measure—the increasing
gap size of preceding cluster (GD: p = 0.031). Gaze duration and total-viewing time
decreased with exemplar repetition (GD: p = 0.048; TT: p < 0.001), but first-fixation
duration was not influenced by exemplar repetition (p = 0.574). These results are
broadly consistent with Chinese-reading experiments showing that fixation durations
on words decrease as their processing ease increases (Liversedge, Zang, Zhang, Bai,
Yan, & Drieghe, 2014; Yan et al., 2006; Yang & McConkie, 1999; see Li et al., 2014
for a corpus analysis), and with results showing that fixation durations on words can
be influenced by the orthographic processing ease of parafoveal words (Kennedy,
1998, 2000; White, 2008).
9
Tables 1-3
Saccade Length. Table 4 indicates that both saccade-length measures increased
with the increasing gap size of the exemplar cluster (PSL: b = 0.12, SE = 0.02, t =
7.00, p < 0.001; ISL: b = 0.09, SE = 0.01, t = 7.92, p < 0.001), although the effect of
exemplar repetition did not reach significance (ps > 0.119). These results are broadly
consistent with Chinese-reading experiments showing that forward saccade length
increases with parafoveal processing ease (Liversedge et al., 2014; see also White &
Liversedge, 2006). Our results therefore also suggest that the dynamic adjustment of
saccade length reflects general demands (i.e., demands associated with parafoveal
processing difficulty) rather than factors specific to reading Chinese.
Table 4
Computational Modeling
The simulations reported below shed light on how the saccades moving the eyes
into the exemplar clusters were influenced by their gap size, and whether this
behavior is better explained via the default targeting of saccades or the
dynamic-adjustment of saccade length. The simulations using each model were based
on 10,000 Monte Carlo runs, each of which involved first sampling a saccade launch
site from a uniform distribution covering the pre-exemplar region so that a precise
saccade target (Simulation 1) or length (Simulation 2) could then be calculated using
Equations 1-4. (The method for finding best-fitting model parameters is described in
the Appendix.) To account for any gap-size effect, the simulations were fit separately
for each gap-size condition, compensating for the scarcity of data in each condition by
collapsing across the smaller (1-2°) and larger (3-4°) gap sizes.
Simulation 1: Default-Targeting Model
The fundamental assumption of this model is that, if an exemplar cluster is
segmented, a saccade is directed towards its center; otherwise, a saccade is directed
10
towards its beginning, allowing the cluster to be refixated. The model does not specify
precisely how this happens, but instead uses Yan et al.’s (2010) assumptions about the
relationship between segmentation and fixation number to estimate the probability of
exemplar’s segmentation from how often it was fixated, using four mutually exclusive
and exhaustive saccade behaviors: (1) refixating the 2-character pre-exemplar region;
(2) fixating and then moving the eyes from the exemplar (presumably because it was
segmented in the parafovea); (3) fixating and then refixating the exemplar
(presumably because it was not segmented in the parafovea); and (4) skipping the
exemplar. Polynomial regression functions (Equation 1) were fit to each possible
saccade launch site (binned to the nearest 0.5 character), with the constraint that the
probabilities of the four types of eye-movement behaviors from each saccade site
summed to 1. To account for exemplar gap size, these functions were fit separately for
smaller and larger gap sizes. In Equation 1, x represents the distance (in character
spaces) between the pre-exemplar launch site and the leftmost edge of the exemplar,
and k0, k1, and k2 respectively represent the intercept, linear, and quadratic polynomial
coefficients.
(1) p(x) = k2 x2 + k1 x + k0
The estimated probabilities were used to specify saccade targets as follows: (1)
saccades to refixate the pre-exemplar region were directed towards this region’s center;
(2) parafoveal exemplar segmentation caused the eyes to move towards its center; (3)
unsuccessful exemplar segmentation caused the eyes to move towards its beginning
(i.e., the center of its first character); and (4) exemplar skips were directed towards the
beginning of post- exemplar cluster (i.e., the center of its first character). Because
visual acuity and the perceptual span are limited (i.e., the latter extends 2–3 characters
to the right of fixation; see Chen & Tang, 1998; Inhoff & Liu, 1998), those rare
instances where the eyes might move past (i.e., to the right of) the beginning of
post-exemplar were not simulated. Variance was added to the saccade target to
simulate saccadic error, which was sampled from a Gaussian distribution with μ = 0,
and σ values selected to fit the empirical fixation-position distributions of incoming
saccades on exemplars.
Simulation 2: Dynamic-Adjustment Model
11
The fundamental assumption of this model is that readers modulate saccade
length as a function of the amount of parafoveal processing that has been completed.
The assumption was instantiated using a simplifying assumption: Saccade length is a
linear function of parafoveal preprocessing. To do this, the amount of exemplar
preprocessing completed from the pre-exemplar region was sampled from a gamma
distribution having shape, α, and scale, β, parameters as described by Equation 2.
(2) preprocessing = gamma (α, β)
Using this equation, the amount of exemplar preprocessing (as determined by the
value of α) was modulated by exemplar gap size as specified by Equation 3. In this
equation, η0 is a constant representing the minimal value of α, and η1 is a parameter
that modulates the influence of gap size on α.
(3) α = η0 + η1 condition
Saccade length (in character spaces) is then linearly related to preprocessing
using Equation 4, where λ is a free parameter scaling this relationship, and intrinsic
variability in saccade length being determined by the parameter β.
(4) length = λ preprocessing
= λ gamma (α, β)
= gamma (η0 + η1 condition, λβ)
Figure 3
Simulation Results
Figure 3 shows that, relative to the default-targeting model, the
dynamic-adjustment model provides a better quantitative fit of the observed
relationships between the saccade launch sites and the subsequent fixation landing
sites, for both progressive saccades (i.e., all saccades launched from the pre-exemplar
region, irrespective of whether they resulted in fixations on an exemplar;
default-targeting: MSE = 6.71 × 10-1; dynamic-adjustment: MSE = 6.30 × 10-3) or
12
incoming saccades (i.e., only saccades launched from the pre-exemplar region that
resulted in exemplar fixations; default-targeting: MSE = 2.75×10-2;
dynamic-adjustment: MSE = 8.70×10-3). Also note that the default-targeting model
exhibited poorer performance across its full range of parameter values (see the shaded
region of Figure 3A).
Discussion
The present article examined whether the dynamic-adjustment account of
saccadic targeting during reading is specific to reading (or perhaps even the reading
of Chinese), or whether it might instead reflect the more general visual and/or
oculomotor constraints of tasks that require the rapid identification of visual patterns
embedded in continuous arrays (e.g., our Landolt-C paradigm). The results of our
experiment support the latter interpretation by demonstrating that the length of the
saccade entering Landolt-C exemplar clusters is modulated by the processing
difficulty (i.e., gap sizes) of those clusters, consistent with prior evidence that
on-going processing difficulty modulates saccade length during the reading of
Chinese (e.g., Liu et al., 2015, 2016; Liu, Huang, Gao et al., 2017; Liu, Huang, Li et
al., 2017; Liu, Guo et al., 2017). Although one might argue against this conclusion on
the grounds that our Landolt-C task does not entail two key components of reading
(i.e., word identification and linguistic processing), the fact that our task does not
engage language processing but does engage other perceptual, cognitive, and motor
processes is precisely why the observed dynamic adjustment of saccade length must
reflect general task demands rather than demands specific to reading.
Additionally, a direct comparison of the performance of explicit computational
versions of the dynamic-adjustment versus default-targeting models suggests that the
former provides a more accurate description of participants’ performance in our
experiment, lending further general support to the hypothesis that, during the reading
of Chinese text, readers modulate saccade lengths in a manner that is sensitive to the
moment-to-moment processing difficulty, rather than selecting one (of a small
possible number) or pre-defined saccade targets (e.g., the beginning or middle of the
upcoming word). Of course, it is important to consider this conclusion in relation to
the full range of possible accounts of saccade targeting during reading.
13
At one extreme, readers might adopt some simple heuristic (e.g., the “fixed
saccade length” assumption used in the simulations reported by Li et al., 2011; see
also Yan et al., 2010) in deciding where to move their eyes. At the other extreme,
readers might use specific saccade targets, as posited by current models of
eye-movement control during the reading of alphabetic languages like both English
(e.g., E-Z Reader: Reichle, Pollatsek, Fisher, & Rayner, 1998; Reichle, Warren, &
McConnell, 2009; Reichle, 2011) and German (e.g., SWIFT: Engbert, Nuthmann,
Richter, & Kliegl, 2005; Schad & Engbert, 2012). Neither of these extremes provides
a plausible account of saccadic targeting in Chinese or our experiment; whereas the
former fails to predict that saccade lengths are modulated by processing difficulty
(e.g., Li et al., 2014; Liu et al., 2015; Wei et al., 2013), the latter erroneously predicts
preferred-viewing locations (which are generally absent; e.g., Li et al., 2011; Liu et al.,
2016; Liu, Huang, Gao, et al., 2017; Liu, Guo, et al., 2017). Therefore, through the
process of elimination, one is left with our preferred account: Chinese readers’
decisions about where to move their eyes are based on information other than word
boundaries. By our account, this information is the relative level of processing
difficulty being experienced at any given point in time. In the context of actual
reading, this processing is related to the segmentation and/or identification of the
upcoming words, while in the context of our Landolt-C paradigm, the processing is
related to the discrimination required to know whether an upcoming cluster is likely to
be a target.
Of course, we acknowledge that our account is incomplete, and that the
“boundary conditions” that determine when participants or readers shift from using
default saccade targets to dynamically adjusting the lengths of their saccades has not
been established. For example, although we have argued that the latter strategy
provides the best account of saccadic targeting in Chinese reader, the two leading
models of eye-movement control in reading (i.e., E-Z Reader and SWIFT) were
developed to explain the patterns of eye movements observed during the reading of
alphabetic languages, and consequently, incorporate the default-targeting assumption
to good approximation. However, evidence that this assumption is only an
approximation is provided by demonstrations that, even in the reading of English, for
example, the length of a saccade exiting a word can be modulated by its frequency
(e.g., Rayner, Ashby, Pollatsek, & Reichle, 2004; White & Liversedge, 2006). We
14
therefore suspect that, in the context of reading, the decisions about where to move
the eyes is jointly determined by the quality (i.e., regularity) of possible saccade
targets, as well as whatever difficulty the reader might currently be experiencing in
lexical processing. This perspective suggests that both factors contributed to the
decisions, with some sort of weighting being assigned to each factor (e.g., more
weight to default targets in English vs. more weight to saccade-length adjustment in
Chinese), rather than being a strict dichotomy. Future empirical work is required to
show this conclusively; future modeling work is required to demonstrate how the two
types of strategies might be integrated and how this integration might be modulated
by languages, writing systems, and possibly other variables.
15
References
Anderson, J. R. (1990). The adaptive character of thought. Hillsdale, NJ: Erlbaum.
Bates, D., Kliegl, R., Vasishth, S., & Baayen, H. (2015). Parsimonious mixed models.
arXiv:1506.04967.
Chen, H. & Tang, C. (1998). The effective visual field in reading Chinese. Reading
and Writing, 10, 245–254.
Engbert, R., Nuthmann, A., Richter, E., & Kliegl, R. (2005). SWIFT: A dynamical
model of saccade generation during reading. Psychological Review, 112,
777-813.
Hoosain, R. (1992). Psychological reality of the word in Chinese. In H.-C. Chen &
O.J.L. Tzeng (Eds), Language processing in Chinese (pp.111–130).
Amsterdam: Elsevier.
Inhoff, A. W. & Liu, W. (1998). The perceptual span and oculomotor activity during
the reading of Chinese sentences. Journal of Experimental Psychology:
Human Perception and Performance, 24, 20–34.
Inhoff, A. W. & Rayner, K. (1986). Parafoveal word processing during eye fixations
in reading: Effects of word frequency. Perception & Psychophysics, 40,
431-439.
Kennedy, A. (1998). The influence of parafoveal words on foveal inspection time:
Evidence for a processing trade-off. In G. Underwood (Ed.), Eye guidance in
reading and scene perception (pp. 149–179). Oxford, UK: Elsevier.
Kennedy, A. (2000). Parafoveal processing in word recognition. Quarterly Journal of
Experimental Psychology A: Human Experimental Psychology, 53, 429–455.
Li, X., Bicknell, K., Liu, P., Wei, W., & Rayner, K. (2014). Reading is fundamentally
similar across disparate writing systems: A systematic characterization of
how words and characters influence eye movements in Chinese reading.
Journal of Experimental Psychology: General, 143, 895-913.
Li, X., Liu, P., & Rayner, K. (2011). Eye movement guidance in Chinese reading: Is
there a preferred viewing location? Vision Research, 51, 1146-1156.
Li, X., Rayner, K., & Cave, K. R. (2009). On the segmentation of Chinese words
during reading. Cognitive Psychology, 58, 525-552.
16
Liu, Y., Guo, S., Yu, L., & Reichle, E.D. (2017). Word predictability affects saccade
length in Chinese reading: An evaluation of the dynamic-adjustment model.
Psychonomic Bellutin & Review. https://doi.org/10.3758/s13423-017-1357-x
Liu, Y., Huang, R., Gao, D-G., & Reichle, E.D. (2017). Further tests of a
dynamic-adjustment account of saccade targeting during the reading of
Chinese. Cognitive Science, 41, 1264–1287.
Liu, Y., Huang, R., Li, Y., & Gao, D-G. (2017). The word frequency effect on
saccade targeting during Chinese reading: Evidence from a survival analysis
of saccade length. Frontiers in Psychology, 8:116.
Liu, Y., Reichle, E. D., & Gao, D.-G. (2013). Using reinforcement learning to
examine dynamic attention allocation during reading. Cognitive Science, 37,
1507-1540.
Liu, Y., Reichle, E. D., & Li, X. (2015). Parafoveal processing affects outgoing
saccade length during the reading of Chinese. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 41, 1229-1236.
Liu, Y., Reichle, E. D., & Li, X. (2016). The effect of word frequency and parafoveal
preview on saccade length during the reading of Chinese. Journal of
Experimental Psychology: Human Perception and Performance, 42,
1008-1025.
Liversedge, S. P., Zang, C., Zhang, M., Bai, X., Yan, G., & Drieghe, D. (2014). The
effect of visual complexity and word frequency on eye movements during
Chinese reading. Visual Cognition, 22, 441–457.
Rayner, K. (1979). Eye guidance in reading: Fixation locations within words.
Perception, 8, 21-30.
Rayner, K., Ashby, J., Pollatsek, A., & Reichle, E. D. (2004). The effects of
frequency and predictability on eye fixations in reading: Implications for the
E-Z Reader model. Journal of Experimental Psychology: Human Perception
and Performance, 30, 720-730.
Rayner, K., & Morrison, R. E. (1981). Eye movements and identifying words in
parafoveal vision. Bulletin of the Psychonomic Society, 17, 135-138.
17
Reichle, E. D. (2011). Serial attention models of reading. In S. P. Liversedge, I. D.
Gilchrist, & S. Everling (Eds.), Oxford handbook on eye movements (pp.
767-786). Oxford, U.K.: Oxford University Press.
Reichle, E. D. & Laurent, P. (2006). Using reinforcement learning to understand the
emergence of “intelligent” eye-movement behavior during reading.
Psychological Review, 113, 390-408.
Reichle, E. D., Pollatsek, A., Fisher, D. L., & Rayner, K. (1998). Toward a model of
eye movement control in reading. Psychological Review, 105, 125-157.
Reichle, E. D., Warren, T., & McConnell, K. (2009). Using E-Z Reader to model the
effects of higher-level language processing on eye movements during reading.
Psychonomic Bulletin & Review, 16, 1-21.
Schad, D.J. & Engbert, R. (2012). The zoom lens of attention: Simulating shuffled
versus normal text reading using the SWIFT model. Visual Cognition, 20,
391-421.
Schilling, H. E. H., Rayner, K., & Chumbley, J. I. (1998). Comparing naming, lexical
decision, and eye fixation times: Word frequency effects and individual
differences. Memory and Cognition, 26, 1270-1281.
Tsai, J. L. & McConkie, G. W. (2003). Where do Chinese readers send their eyes? In
R. R. J. Hyona & H. Deubel (Eds.), The mind’s eye: Cognitive and applied
aspects of eye movement research (pp. 159-176). Amsterdam, the Netherlands:
Elsevier.
Vanyukov, P. M., Warren, T., Wheeler, M. E., & Reichle, E. D. (2012). The
emergence of frequency effects in eye movements. Cognition, 123, 185-189.
Wei, W., Li, X., & Pollatsek, A. (2013). Word properties of a fixated region affect
outgoing saccade length in Chinese reading. Vision Research, 80, 1-6.
White, S. J. (2008). Eye movement control during reading: Effects of word frequency
and orthographic familiarity. Journal of Experimental Psychology: Human
Perception and Performance, 34, 205–223.
White, S. J. & Liversedge, S. P. (2006). Foveal processing difficulty does not
modulate non-foveal orthographic influences on fixation positions. Vision
Research, 46, 426–437.
18
Williams, C. C. & Pollatsek, A. (2007). Searching for an O in an array of Cs: Eye
movements track moment-to-moment processing in visual search. Perception
and Psychophysics, 69, 372-381.
Williams, C. C., Pollatsek, A., & Reichle, E. D. (2014). Examing eye movements in
visual search through clusters of objects in a circular array. Journal of
Cognitive Psychology, 26, 1-14.
Yan, G., Tian, H., Bai, X., & Rayner, K. (2006). The effect of word and character
frequency on the eye movements of Chinese readers. British Journal of
Psychology, 97, 259-268.
Yan, M. & Kliegl, R. (2016). CarPrice versus CarpRice: word boundary ambiguity
influences saccade target selection during the reading of Chinese sentences.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 42,
1832–1838.
Yan, M., Kliegl, R., Richter, E., Nuthmann, A., & Shu, H. (2010). Flexible
saccade target selection in Chinese reading. The Quarterly Journal of
Experimental Psychology, 63, 705-725.
Yang, H. & McConkie, G. W. (1999). Reading Chinese: Some basic eye-movement
characteristics. In J. Wang, A. W. Inhoff, & H.-C. Chen (Eds.), Reading
Chinese script (pp. 207-222). Mahwah, NJ: Erlbaum.
Yu, L. & Reichle, E. D. (2017). Chinese vs. English: Insights on cognition during
reading. Trends in Cognitive Sciences, 21, 721-724.
19
Table 1. LMM analyses of first-fixation durations (ms).
PredictorsModel Values4 (ms)
b SE t p Min. Max.
ClusterPosition Intercept 328.58 9.80 33.54 < 0.001 - -
Previous# Characters1 0.11 1.40 0.08 0.940 301.35 301.67
Gap Size2 -1.12 1.22 -0.92 0.363 303.19 299.83
CurrentRepetitions3 0.04 0.07 0.56 0.574 300.55 302.47
Gap Size2 -6.16 1.29 -4.77 < 0.001 310.75 292.27
Subsequent# Characters1 0.15 1.35 0.11 0.913 292.05 292.49
Gap Size2 -4.20 1.20 -3.49 0.001 298.56 285.98
Notes:1. Number of characters: min. = 1 character, max. = 4 characters.2. Gap size: min. = 1°, max. = 4°.3. Repetitions: min. = 1, max. = 50.4. Estimates of predicted variable values were calculated while fixing the values of the other variables equal to their means.
20
Table 2. LMM analyses of gaze durations (ms).
PredictorsModel Values4 (ms)
b SE t p Min. Max.
ClusterPosition Intercept 432.42 14.82 29.19 < 0.001 - -
Previous# Characters1 -0.90 1.87 -0.48 0.629 348.53 345.82
Gap Size2 -3.52 1.61 -2.19 0.031 352.46 341.90
CurrentRepetitions3 -0.21 0.11 -1.98 0.048 352.29 342.06
Gap Size2 -18.29 2.25 -8.14 < 0.001 374.61 319.74
Subsequent# Characters1 -2.39 1.90 -1.26 0.209 323.33 316.16
Gap Size2 -6.87 1.79 -3.85 < 0.001 330.05 309.43
Notes:1. Number of characters: min. = 1 character, max. = 4 characters.2. Gap size: min. = 1°, max. = 4°.3. Repetitions: min. = 1, max. = 50.4. Estimates of predicted variable values were calculated while fixing the values of the other variables equal to their means.
21
Table 3. LMM analyses of total-viewing times (ms).
PredictorsModel Values4 (ms)
b SE t p Min. Max.
ClusterPosition Intercept 495.33 20.10 24.65 < 0.001 - -
Previous# Characters1 -1.83 2.80 -0.65 0.518 381.57 376.09
Gap Size2 -2.26 2.18 -1.04 0.307 382.21 375.45
CurrentRepetitions3 -0.56 0.12 -4.60 < 0.001 392.52 365.13
Gap Size2 -24.15 2.97 -8.13 < 0.001 415.05 342.61
Subsequent# Characters1 -2.31 2.11 -1.09 0.275 346.08 339.14
Gap Size2 -10.36 2.17 -4.78 < 0.001 358.15 327.07
Notes:1. Number of characters: min. = 1 character, max. = 4 characters.2. Gap size: min. = 1°, max. = 4°.3. Repetitions: min. = 1, max. = 50.4. Estimates of predicted variable values were calculated while fixing the values of the other variables equal to their means.
22
Table 4. LMM analyses of progressive and incoming saccade length (in characters).
Saccade Type VariablesModel Values3
(characters)
b SE t p Min. Max.
Progressive Saccade
Intercept 3.02 0.22 13.70 < 0.001 - -
Repetitions1 -0.01 0.003 -1.61 0.119 3.33 3.07
Gap Size2 0.12 0.02 7.00 < 0.001 3.01 3.38
Incoming Saccade
Intercept 1.86 0.04 45.93 < 0.001 - -
Repetitions1 0.001 0.001 0.96 0.344 2.08 2.11
Gap Size2 0.09 0.01 7.92 < 0.001 1.97 2.22
Notes:1. Repetitions: min. = 1, max. = 50.2. Gap size: min. = 1°, max. = 4°.3. Estimates of predicted variable values were calculated while fixing the values of the other variables equal to their means.
23
Figure Caption
Figure 1. Examples of two Chinese sentences and their translations. In the first,
the sequence of four underlined characters correspond to two words. In the second,
the same characters correspond to a single word.
Figure 2. Examples of experimental materials, with two targets (i.e., circles) and one
exemplar cluster being rendered in gray for illustrative purposes.
Figure 3. The predicted relationship (in character spaces) between the saccade launch
site (from the pre-exemplar region) and the subsequent fixation landing site (on the
exemplar) generated by the: (A) default-targeting model (Simulation 1); and (B)
dynamic-adjustment model (Simulation 2). The symbols show the observed means
averaged within each launch-distance bin, the black and gray lines respectively
represent the simulated progressive and incoming saccades, and both launch sites and
landing sites are aligned to the beginning of the exemplar. The shaded region in panel
(A) demarcates the default-targeting model’s performance across its full parameter
domain (i.e., between the two extreme cases in which exemplars are never vs. always
segmented in the parafovea). (Notes: LE = larger gap-size exemplars; SE = smaller
gap-size exemplars; PS = progressive saccade; IS = incoming saccade.)
24
Figure 1.
25
Figure 2.
26
Figure 3.
27
Appendix
Simulation 1 parameters: As Equation 1 shows, polynomial regression functions
were used to estimate the probabilities of observing the four different types of
saccades using the method of least squares. Because these probabilities summed to 1
for each saccade launch site, only the probabilities associated with three saccades
types were estimated. Finally, the values of σ, which control saccadic-error variability,
were chosen to maximize the goodness-of-fit to the empirical fixation-position
distributions of incoming saccades in the smaller and larger gap-size conditions
separately (MSE = 0.327 and 0.847, respectively). Table A1 lists the best-fitting
parameters. Figure A1 shows that these parameters accurately describe the empirical
data (i.e., probability of refixating pre-exemplar region: MSE = 1.783 × 10-5;
probability of fixating exemplar center: MSE = 3.479 × 10-4; probability of fixating
exemplar beginning: MSE = 1.951 × 10-5). Simulation 1 thus required 20 free
parameters.
Simulation 2 parameters: The expected value of Equation 4 is λβ (η0 + η1condition),
or the predicted value using the mean first progressive saccade length from the
pre-exemplar region. Thus, two groups of parameters, λβη1 and λβη0, are coefficients
for a regression equation for progressive saccade length using gap-size condition as a
predictor variable (i.e., smaller condition: 1-2°; larger condition: 3-4°). And because
the variance associated with Equation 4 (i.e., the variance associated with saccadic
error) is given by the quantityλ2β2(η0 + η1condition), the parameter pair λβ can also be
estimated using the empirical distribution of fixations on exemplars. The final
parameter values used to simulate the results were: η0 = 3.178; η1 = 0.254; and λβ =
1.166 and 0.875 for the smaller and larger gap-size conditions, respectively.
Simulation 2 thus required four (groups of) free parameters.
28
Table A1. Best-fitting parameters for default-targeting model (Simulation 1).
Exemplar Gap Size Saccade Type k2 k 1 k 0 σ
Larger
Refixate Pre-Exemplar Region 0.139 0.121 0.024
0.847Fixate Exemplar Center -0.115 -0.224 0.354
Fixate Exemplar Beginning -0.060 -0.155 -0.026
Smaller
Refixate Pre-Exemplar Region 0.137 0.093 0.018
0.327Fixate Exemplar Center -0.158 -0.289 0.372
Fixate Exemplar Beginning -0.064 -0.159 -0.004
29
Figure A1. Observed (symbols) and estimated (lines) probabilities of refixating the
pre-exemplar region, fixating the exemplar center (i.e., single fixation), fixating the
exemplar beginning (i.e., first-of-multiple fixations), and skipping the exemplar, as a
function of exemplar gap size. (Notes: SE = smaller gap-size exemplars; LE = larger
gap-size exemplars.)