This is paper has been accepted for publication in The Journal of experimental psychology: human perception
and performance, 2016.
This article may not exactly replicate the final version published in the APA journal. It is not the copy of record.
Running head: SPONTANEOUS RE-READING WITHIN SENTENCES
Spontaneous re-reading within sentences: Eye movement control and visual sampling
Sarah J. White
Laura M.T. Lantz
Kevin B. Paterson
Department of Neuroscience, Psychology and Behaviour, University of Leicester, UK
Corresponding Author:
Sarah J. White
Department of Neuroscience, Psychology and Behaviour,
University of Leicester, University Road, Leicester, LE1 7RH, UK
[email protected]
Spontaneous re-reading within sentences
1
Abstract
Three experiments examine the role of previously read text in sentence comprehension
and the control of eye movements during spontaneous re-reading. Spontaneous re-reading
begins with a regressive saccade and involves re-inspection of previously read text. All
three experiments employed the gaze contingent change technique to modulate the
availability of previously read text. In Experiment 1 previously read text was permanently
masked either immediately to the left of the fixated word (beyond wordn) or more than one
word to the left (beyond wordn-1). The results of Experiment 1 indicate that the availability
of the word immediately to the left (wordn-1) is important for comprehension. Experiments
2 and 3 further explored the role of previously read text beyond wordn-1. In these studies
text beyond wordn-1 was replaced, retaining only word length information, or word length
and shape information. Following a regression back within a sentence, meaningful text
either re-appeared or remained unavailable during re-reading. The experiments show that
the visual format of text beyond wordn-1 (the parafoveal postview) is important for
triggering regressions. The results also indicate that, as least for more complex sentences,
the availability of meaningful text is important in driving eye movement control during rereading.
Keywords: re-reading, eye movements, regressions
Spontaneous re-reading within sentences
2
Considerable research has been undertaken into the nature of word recognition and
eye movement control as the eyes initially move forward through text (Rayner, 2009).
Models of eye movement control now provide good accounts of such “first-pass” reading
behavior (Engbert, Longtin, & Kliegl, 2002, Reichle, Pollatsek, Fisher, & Rayner, 1998).
Some research has also been undertaken into the processes underlying the re-reading that
occurs for repeated readings of text (Hyönä, & Niemi, 1990; Raney, & Rayner, 1995).
However compared to studies of reading during first-pass, relatively few studies have
examined eye movement control for spontaneous re-reading. That is, re-reading of
previously read text that occurs following a spontaneous regression back in the text. Ten to
fifteen percent of fixations during reading are regressive (Rayner, 2009), hence
spontaneous re-reading is a key aspect of the reading process.
Re-reading may occur as a result of processing difficulty associated with text
comprehension (e.g. Just & Carpenter, 1980; for a review see Clifton, Staub, & Rayner,
2007) or falling confidence in the identity of previously read words (Bicknell & Levy,
2010). Models of eye movement control predict that post-lexical processing modulates the
likelihood of regressing back in the text (Engelmann, Vasishth, Engbert, & Kliegl, 2013;
Reichle, Warren, & McConnell, 2009). However it is notoriously difficult to study the
mechanisms underlying spontaneous re-reading as it does not always occur, and when it
does occur, the timing and location of the re-reading is variable (von der Malsburg, Kliegl,
& Vasishth, 2015; von der Malsburg & Vasishth, 2011). Two recent studies manipulated
how previously read words were displayed within a sentence using the gaze contingent
change technique (Booth & Weger, 2013; Schotter, Tran & Rayner, 2014). These studies
enabled naturalistic reading behavior such that regressions were triggered spontaneously,
Spontaneous re-reading within sentences
3
rather than due to another task such as responding to a target word (Weger & Inhoff,
2007). The present study builds on this work, employing variants of the trailing mask
paradigm developed by Schotter et al.
Three key issues are addressed: The first issue is the importance of the availability
of previously read text for sentence comprehension. In particular, the role of the word
immediately to the left of the fixated word (wordn-1) is examined in Experiment 1. The
second key issue is the role of previously read text in triggering re-reading (regressions).
Experiments 2 and 3 specifically examine how the nature of the parafoveal postview of
previously read text affects the likelihood of moving the eyes back in the text. The third
key issue is the role of meaningful text in driving eye movement control during re-reading.
Experiments 2 and 3 examine how the availability of text during re-reading modulates
reading behavior. These three key issues are set out in more detail below.
Sentence comprehension
Visual re-sampling of previously read text may be essential for comprehension if
the text is used as an “external memory” (Kennedy, 1983, 1992; O’Regan, 1992). In
contrast, visual re-sampling of previously read text may provide minimal benefit for
comprehension if re-reading serves to confirm the initial interpretation of the text
(Christiansen, Luke, Hussey, & Wochna, 2016), if it provides “time out” for continued
text processing (Mitchell, Shen, Green, & Hodgson, 2008), or if simply fixating at the
location of previously encoded information facilitates re-processing (Ballard, Hayhoe,
Pook, & Rao, 1997; Ferreira, Apel, & Henderson, 2008; Kennedy, 1992).
Spontaneous re-reading within sentences
4
Booth and Weger (2013) (Experiment 3) examined whether words are visually resampled during re-reading. In their study a target word in a sentence was changed to a
word with a different meaning following a spontaneous regression. Subsequently, when
asked to select the meaning of the sentence, participants who regressed and fixated on the
changed word were more likely to select the changed meaning, indicating that words are
visually re-sampled during re-reading. However, word recognition for fixated words is
known to occur automatically (MacLeod, 1991; Neely, 1977). As Booth and Weger note,
the changed words may have been automatically re-sampled during re-reading, overwriting memory for the originally processed word. Therefore, although Booth and
Weger’s study shows that words are visually re-sampled during re-reading, it is not clear if
such visual re-sampling is actually necessary for comprehension.
Nevertheless, recent work by Schotter et al. (2014) indicates that visual resampling of previously read text is important for comprehension (see also Benedetto et al.,
2015; Harvey & Walker, 2014). Schotter et al. introduced the trailing mask paradigm,
such that all words to the left of the fixated word were permanently masked, preserving
only word length information. Comprehension accuracy was lower in the trailing mask
condition compared to the control, indicating that comprehension is facilitated by the
opportunity to re-sample text during re-reading. However it could be that the opportunity
to re-sample the word immediately to the left of the fixated word (wordn-1) was especially
important in this study. Previous research indicates that the orthographic characteristics of
the word to the left of the fixated word (wordn-1) can be processed in the parafovea
(Binder, Pollatsek, & Rayner, 1999; Jordan, McGowan, Kurtev & Paterson, 2016; RoyCharland et al., 2012; Wang, Tsai, Inhoff, & Tzeng, 2009). Other studies indicate that re-
Spontaneous re-reading within sentences
5
reading of wordn-1 is associated with continued lexical processing of that word (Engbert et
al., 2002; Reichle et al., 1998), falling confidence in the identification of the word
(Bicknell & Levy, 2011), or due to a corrective saccade following a mislocated fixation
(see Vitu, 2005, for a review). Given that the availability of wordn-1 is important for
orthographic and lexical processing of words, it may also be important for sentence
comprehension. Experiment 1 builds on Schotter et al.’s study by employing the trailing
mask paradigm to further examine whether the availability of previously read text is
important for comprehension, especially the word to the left of the fixated word (wordn-1).
Comprehension is also assessed in Experiments 2 and 3. Importantly, Experiments
2 and 3 directly manipulate text replacement permanence for text beyond wordn-1, that is,
whether previously read text beyond wordn-1 remains unavailable, or re-appears, once rereading commences. If visual re-sampling of text during re-reading is necessary for
comprehension then we would expect to see an effect of text replacement permanence,
with lower comprehension scores when previously read text remains permanently
unavailable. However note that the present study employs standard procedures for
assessing comprehension question response accuracy in single sentence reading studies.
That is, sentences are followed by questions with multiple choice answers. These
procedures are not designed to be sensitive to subtle differences in comprehension (for
example, integration with prior knowledge), which are beyond the scope of this article.
The parafoveal postview
Classic perceptual span studies indicate that the parafoveal postview of previously
read text is processed no more than four characters to the left of fixation (McConkie &
Spontaneous re-reading within sentences
6
Rayner, 1976; Rayner, Well, & Pollatsek, 1980; but see: Jordan et al., 2016; Jordan,
McGowan, & Paterson, 2013; Rayner, Castelhano, & Yang, 2009; Rayner, Yang, Schuett,
& Slattery, 2014). Though the parafoveal postview has been shown to be processed at
least at a visual or orthographic level, prior to a regression (Apel, Henderson & Ferreira,
2012; McGowan, White, & Paterson, 2013). The parafoveal postview, as well as stored
representations of previously read text, may contribute to the programming of regressive
saccades. Regressions can be targeted to words within sentences (Kennedy & Murray,
1987) and within sentences to regions of difficulty (Frazier & Rayner, 1982), though
corrective saccades may be needed to accurately locate the target (Inhoff & Weger, 2005).
Kennedy and colleagues proposed that previously read text is spatially coded (Kennedy,
1983, 1992; Kennedy, Brooks, Flynn, & Prophet, 2003; Kennedy & Murray, 1987; see
also Fischer, 2000), though spatial coding may be much coarser further from fixation
(Inhoff & Weger, 2005). Other studies have indicated that regression targeting can be
modulated by linguistic knowledge or processing demand, for example, modulated by the
number of intervening words between the start and end of the regressive saccade (Weger
& Inhoff, 2007). Recent work also indicates that verbal memory may be important in
guiding regressions, as regression targeting is affected by articulatory suppression
(Guérard, Saint-Aubin, & Maltais, 2013; Guérard, Saint-Aubin, Maltais, & Lavoie, 2014).
Other work indicates that spatial layout, in addition to linguistic guidance, can also
modulate targeting of regressions (Mitchell et al., 2008). To summarize, a range of factors
are likely to contribute to the processes underlying programming of the metrics of
regressive saccades. The present study focuses on how the nature of the parafoveal
postview modulates the likelihood of regressions being triggered.
Spontaneous re-reading within sentences
7
Several studies indicate that the availability of a previously read sentence can
modulate the likelihood of regressing back (Booth & Weger, 2013; Inhoff & Weger, 2005;
Kennedy, 1982, 1983). The availability of text or word shape cues in the parafoveal
postview of previously read text may be important in the triggering of regressive saccades
within sentences. In Schotter et al.’s (2014) study there were very few regressions in the
trailing mask condition, which could be due to word shape cues being removed from the
parafoveal postview. Nevertheless, results from non-reading tasks (Altmann, 2004;
Richardson & Spivey, 2000) indicate that attention to previously read text could perhaps
result in regressions back even in the absence of visual/spatial cues.
In the present study, Experiments 2 and 3 manipulate the nature of the parafoveal
postview of previously read text such that the postview was correct (control condition),
provided word length cues (“X” masks) or provided word shape cues (visually similar
letters). As noted above, text replacement permanence was also manipulated, that is,
whether previously read text remains unavailable, or re-appears, once re-reading
commences. For the temporary text replacement conditions these manipulations are
similar to studies that employed the moving window technique (McConkie & Rayner,
1975, 1976; Rayner, 2014). The temporary word length replacement condition is the same
as the “two word” condition employed in Rayner et al.’s (1980, Experiment 3) study, and
the temporary word shape replacement condition is the same as the “n-2” visually similar
condition employed by Jordan et al. (2016). Rayner et al. did not report regression
measures, but Jordan et al. showed a slight increase in the number of regressions when
letters in the parafoveal postview were replaced. Importantly, Experiments 2 and 3 provide
an opportunity to compare how word length compared to word shape parafoveal postviews
Spontaneous re-reading within sentences
8
modulate the likelihood of regressions. If the likelihood of making a regression is reduced
in the word length replacement conditions then this will indicate that the visual format of
the parafoveal postview is important in triggering regressions.
Eye movement control during re-reading
The mechanisms underlying eye movement control once re-reading has
commenced are assumed to be similar to those that occur during first-pass reading
(Reichle et al., 2009). However studies have shown differences between model predictions
and observed data (Inhoff, Greenberg, Solomon, & Wang, 2009; Warren, White, &
Reichle, 2009). For example, Warren et al. investigated the causes of wrap-up effects and
compared the findings with simulations using E-Z Reader 10. The model predictions for
the likelihood of making a regression provided a good fit with the observed data. However
there were differences between the observed data and model predictions for go-past time,
a measure that includes re-reading. Clearly there is a need for much more empirical and
theoretical work in this area. Ultimately detailed studies will be needed to examine the
precise characteristics of eye movement control during re-reading; for example, how word
characteristics and parafoveal preview modulate re-reading behavior.
The present study focuses on whether text availability modulates re-reading
behavior. The manipulation of text replacement permanence in Experiments 2 and 3
provides a test of whether meaningful text helps drive eye movement behavior during rereading. If eye movement control during re-reading is driven by visual re-sampling of the
words, then there should be an effect of text replacement permanence such that re-reading
time should be reduced when meaningful text is unavailable (permanent replacement)
Spontaneous re-reading within sentences
9
compared to when it is available (temporary replacement). However if re-reading behavior
is driven by the re-allocation of attention to the location of previously encoded text, then
eye movement behavior may not necessarily be driven by visual re-sampling of words,
and re-reading behavior may be similar regardless of text availability (no effect of text
replacement permanence). Another possibility, perhaps especially if re-reading behavior
ordinarily provides “time out” for continued text processing (Mitchell et al., 2008), is that
there may be sufficient flexibility in the eye movement control systems that different eye
movement behaviors may arise, even if comprehension is unchanged. For example, longer
first-pass reading times might compensate for shorter re-reading times.
Experiment 1
Experiment 1 focuses on the first key issue, the importance of the availability of
previously read text for sentence comprehension. As outlined in the Introduction, the
availability of wordn-1 during first-pass may be especially important for comprehension. In
contrast, re-reading beyond wordn-1 may function to confirm the initial interpretation of the
text (Christiansen et al., 2016) or provide “time out” for continued processing (Mitchell et
al., 2008). Therefore visual re-sampling for text beyond wordn-1 may be less important for
comprehension. Experiment 1 tests this by examining whether sentence comprehension is
only affected when all text to the left of the fixated word is masked (“Beyond wordn”) or
whether sentence comprehension is also impaired when wordn-1 is preserved during firstpass and all text to the left of wordn-1 is masked (“Beyond wordn-1”). That is, in the
Beyond wordn condition masking starts one word before the fixated word, whereas in the
Beyond wordn-1 condition masking starts two words before the fixated word. These
Spontaneous re-reading within sentences 10
manipulations are illustrated in Figure 1. If the availability of wordn-1 during first-pass is
especially critical for comprehension then question accuracy should be lower in the
Beyond wordn condition compared to the control. If comprehension question accuracy can
remain high, with little difference in response accuracy between the control and the
Beyond wordn-1 condition, then the visual re-sampling demonstrated by Booth and Weger
(2013) may not always be necessary for comprehension.
Insert Figure 1 here
The Beyond wordn condition employs the trailing mask paradigm introduced by
Schotter et al. (2014). However in Schotter et al.’s study there was a cue before each trial
indicating if the mask would be displayed, and the mask did not retain word shape
information (only word length). There were few regressions back in the trailing mask
condition, but it is unclear if this was due to the removal of previously read text, or due to
the cue or the type of mask. Therefore in Experiment 1 there were no cues before each
trial to indicate whether the text would be masked, and the masks retained word shape
information as well as word length (visually similar letters). Also, in order to examine
whether lexical processing was modulated by the trailing mask manipulation, Experiment
1 also included a manipulation of word frequency.
Method
Participants. Thirty members of the University of Leicester community
participated in the study. Participants either received course credit for their participation or
Spontaneous re-reading within sentences 11
monetary compensation. Participants were native English speakers with normal vision and
no history of reading disorders. All were naïve to the purpose of the experiment.
Apparatus. Eye movements were monitored using an EyeLink 1000 eye tracker
(SR Research Ltd.). Pupil location was sampled at a rate of 1000Hz. Viewing was
binocular though only movements of the right eye were recorded. The sentences were
presented on a ViewSonic P227fb monitor with a refresh rate of 7ms (150Hz). The
viewing distance was 80cm and one character subtended approximately 0.3 degrees of
visual angle. Sentences were presented as a single line of text (maximum 83 characters), in
Courier New bold font with text presented in black on a light grey background.
Materials and design. There were six conditions: two variables, display format and
critical word frequency, were manipulated in a 3 (display: control, Beyond wordn, Beyond
wordn-1) X 2 (word frequency: high, low) design. The Beyond wordn condition employed a
trailing mask manipulation such that all words to the left of the fixated word were
permanently replaced. The Beyond wordn-1 condition employed a similar manipulation,
except that all words to the left of wordn-1 were permanently replaced. In both conditions
text was replaced with visually similar letters (see Figure 1).
There were 60 high and 60 low frequency critical words, all were between four and
six letters long (M = 5.3, SD = 0.7) (see White, Drieghe, Liversedge, & Staub, under
revision). Word frequencies were calculated as Zipf values, based on the SUBTLEX-UK
corpus (Van Heuven, Mandera, Keuleers, & Brysbaert, 2014). The high frequency words
had significantly higher Zipf values (M = 5.24, SD = 0.30) than the low frequency words
(M = 3.36, SD = 0.32) (t(59)=30.48, p < 0.001). Each pair of critical words was embedded
in the same neutral sentence frame up to and including the word after the critical word (see
Spontaneous re-reading within sentences 12
Figure 1 caption for an example). A different set of twelve participants completed a cloze
task: they were provided with each of the 60 experimental sentences up to, but not
including, the critical word and were asked to provide a word that could fit as the next
word in the sentence. None of the completions included either of the critical words,
demonstrating that the critical words were not predictable from the initial sentence
context. Each sentence display was always followed by a three option multiple choice
question to test for comprehension. A separate set of 18 participants completed a prescreen
task in which they were presented with the same multiple choice questions and answers
(not the sentences) as used in the main experiment, and were asked to guess the correct
answer. 43% of guesses were correct. The prescreen results demonstrate that it was not
possible to achieve a high comprehension score without reading the sentences.
The six conditions were manipulated within participants and items following a
Latin square design. Each participant was presented with two of these counterbalanced
sets in separate blocks, such that they were presented with both the high and low
frequency versions of each item in separate blocks. This was possible because the initial
sentence frames were very neutral and the sentence endings distinct across the two word
frequency conditions. Therefore participants were presented with all 120 experimental
sentences and an additional 84 filler sentences. The order of the lists was counterbalanced
and the order of items randomized for each participant within each block. The first block
was preceded by six practice trials.
Procedure. Participants first completed a visual acuity test, reading letters from an
ETDRS visual acuity chart at the same viewing distance as employed during the
experiment (Ferris, & Bailey, 1996). Participants were instructed to read the sentences for
Spontaneous re-reading within sentences 13
comprehension and to respond to questions by pressing buttons on a game controller. They
were advised that some of the sentence displays may appear strange but they should aim to
read the sentences for comprehension. A chinrest and forehead rest minimized head
movements. The eye tracker was calibrated using a three point horizontal calibration.
Calibration accuracy was checked at all three positions every third trial and centrally prior
to every trial, recalibrations were undertaken when necessary (ensuring spatial accuracy
<0.3º). Participants fixated on a fixation cross at the position of the start of the line of text
before each item was presented.
In the Beyond wordn and Beyond wordn-1 conditions, once a saccade moved into a
word on first-pass the words beyond wordn or wordn-1 were replaced. Word boundaries
were located at the center of the space between words. Given the time for data transfer,
processing time, and display change refresh, the display change occurred within ~15ms of
the eye crossing the boundary, such that the display change occurred during the saccade
into the word, or a few milliseconds after the boundary was crossed.
Analyses. Fixations less than 80ms and within one character width of the previous
or next fixation were merged. Following this, fixations less than 80 and more than 1200ms
were excluded. Trials were excluded if there was more than one blink during reading of
the sentence. Three participants were replaced due to more than 15% of trials being
excluded. Overall 98% of trials were included in the analyses. For 80% of the trials
included in the analyses there were no blinks. The effects of word frequency were
examined using eye movement measures for the critical word. First-pass refers to eye
movement behavior for words before moving out of them, either to the right (following a
fixation or a skip of that word) or to the left (regression out). First fixation durations, gaze
Spontaneous re-reading within sentences 14
durations (the sum of first-pass fixations on the word) and total time (sum of all fixations
on the word) are reported. The effects of the three display conditions were also examined
for global measures. Sentence reading time is calculated from when the sentence was
presented on the screen until the participant pressed a button to continue. Question
response time is calculated from when the question was presented on the screen until the
participant made a button response. The “Number of first-pass fixations” and average
“First-pass fixation duration” are reported based on first-pass fixations per sentence. Firstpass fixations are defined as above, that is, fixations that occur before moving to the right
of a word or before a regression out of it. These first-pass measures provide an insight into
the nature of word processing as the eyes first move through the text. Any effects of
display condition for these first-pass measures must arise either due to the postview of text
to the left of fixation, or following re-reading of words earlier in the sentence. Additional
measures examined the effect of display condition on re-reading behavior. “Regression to
wordn-1” is the proportion of trials including at least one leftward saccade to the previous
word in the sentence. “Long-range regression” is the proportion of trials including at least
one leftward saccade beyond the previous word in the sentence, that is, beyond wordn-1.
“Re-reading time” is the sum of fixations on words after first-pass for trials for which rereading occurred. “Re-reading fixation duration” is the average duration of fixations on
words after first-pass, per sentence.
Analyses were undertaken using linear mixed effects models (Baayen, Davidson,
& Bates, 2008) using R (R Core Team, 2016) and the lme4 package (Bates, Maechler &
Bolker, 2011). Participants and items were treated as crossed random effects. A maximal
random effects structure was employed with random subject and item intercepts, and
Spontaneous re-reading within sentences 15
random subject and item slopes (Barr, Levy, Scheepers, & Tily, 2013). Trial order was
initially included in the random effects structure, but removed from item and then also the
subject random effects structure when the model failed to converge. None of the models
converged with trial order included in the random effects structure. After removing trial
order, if a model failed to converge, the random effects structure of the model was
progressively trimmed, first items and then subjects, first removing correlations between
factors, then interactions, and then random factors (any models that failed to converge are
detailed in footnotes or Table notes). Analyses were undertaken on both the raw data and
log-transformed data, although only the analyses for the raw data are reported. Unless
stated in the text or Tables, results that were significant for the raw data were also
significant for the log-transformed data. Comprehension question accuracy and the
measures for the proportion of trials including regressions (binary data) were analyzed
using logistic models. t/z values greater than 1.96 were considered significant.
For Experiment 1, baseline contrasts were used to examine the effects of display
condition: Control vs. Beyond wordn and Control vs. Beyond wordn-1. For analyses of the
critical word, the two word frequency conditions were examined using a sliding contrast,
in addition to the two display condition contrasts and interactions with word frequency.
For the measures across the sentence, in order to examine if there were any differences in
behavior as the experiment progressed, analyses were undertaken as follows: without trial
order, with centered trial order (additive), and with centered trial order as an interaction.
Analyses are reported that include the interaction with trial order only if model fit was
improved by including this interaction. In addition, analyses are reported that include the
additive effect of trial order only if model fit was improved compared to the base model.
Spontaneous re-reading within sentences 16
Otherwise, only results for the base model are reported. Note that negative estimates for
effects of trial order reflect shorter reading times or fewer fixations as the experiment
progressed.
Results
Means are shown in Table 1 and the results for the linear mixed effects models are
reported in Table 2. For question response accuracy the models including trial order did
not provide better fits than the base model (χ2s < 1.8, ps > 0.6). For question response
time, the first-pass measures, long-range regressions and re-reading fixation duration, the
models including an additive effect of trial order provided better fits than the base models
(χ2s > 9, ps < 0.01), but the models with the interaction did not provide better fits than the
additive models (χ2s < 4, ps > 0.15). For sentence reading time, regressions to wordn-1 and
re-reading time the models including interaction with trial order provided significantly
better fits than the models including an additive effect of trial order (χ2s > 7, ps < 0.05).
The interactions with trial order are detailed below.
Insert Tables 1 and 2 here
Question response time and accuracy. Question response time was significantly
longer and accuracy significantly lower in the Beyond wordn condition compared to the
control. There was no difference in question response time or accuracy between the
control and the Beyond wordn-1 condition. The results indicate that sentence
Spontaneous re-reading within sentences 17
comprehension was more difficult in the Beyond wordn condition due to the removal of
wordn-1.
Sentence reading time and first-pass measures. For sentence reading time there
was a significant interaction between trial order and the control vs. Beyond wordn contrast
(b = -2.23, se = 0.54, t = -4.17) but no interaction between trial order and the control vs.
Beyond wordn-1 contrast (b = -0.78, se = 0.53, t = -1.46). The interaction is illustrated in
Figure 2, which shows Loess curves for sentence reading time for each of the display
conditions as a function of trial order. Figure 2 indicates that for all three conditions
sentence reading times became shorter as the experiment progressed, and the interaction is
characterized by especially long sentence reading times in the Beyond wordn condition
compared to the control at the start of the experiment.
Insert Figure 2 here
First-pass fixation durations were significantly longer in the Beyond wordn
condition compared to the control, but there was no difference between the control and
Beyond wordn-1 condition. There were no significant effects for number of first-pass
fixations.
Re-reading. For re-reading time and regressions to wordn-1 there was a significant
interaction between trial order and the control vs. Beyond wordn contrast (Re-reading
time: b = -1.53, se = 0.50, t = -3.04; Regressions to wordn-1: b = -0.004, se = 0.002, t = 2.62) but no interactions between trial order and the control vs. Beyond wordn-1 contrast
(t/zs < 1). The Loess curves for re-reading time shown in Figure 2 indicate that once
Spontaneous re-reading within sentences 18
participants learnt that wordn-1 was not available after first-pass, they spent less time rereading (also characterized by fewer regressions to wordn-1) compared to when the text
always remained available (control).
There were significantly more trials including long-range regressions in the control
condition compared to the Beyond wordn and Beyond wordn-1 conditions. There was no
difference in re-reading time or regressions to wordn-1 between the control and the Beyond
wordn-1 condition. Re-reading fixation durations were significantly longer in the Beyond
wordn-1 condition compared to the control, but the control vs. Beyond wordn contrast was
not significant.
Critical word. Means for first-fixation duration (FFD), gaze duration (GD) and
total time (TT) on the critical word are shown in Table 3. All three measures1 produced
significant effects of word frequency (FFD: b = 31.71, se = 6.36, t = 4.99; GD: b = 45.79,
se = 6.89, t = 6.65; TT: b = 99.56, se = 14.86, t = 6.70). Neither first fixation duration or
gaze duration produced any significant effects of the display condition contrasts or any
interactions with word frequency (ts < 1.5).
Insert Table 3 here
In contrast to the first-pass measures for the critical word, total times were
significantly longer in the Beyond wordn-1 condition compared to the control (b = -23.74,
se = 9.26, t = -2.56) and there was a significant interaction with word frequency (b = 58.54, se = 20.27, t = -2.89) such that the effect of word frequency is larger in the control
condition compared to when text was replaced beyond wordn-1. There was a similar pattern
Spontaneous re-reading within sentences 19
for control vs. Beyond wordn, however although the contrast (b = -26.64, se = 12.28, t = 2.17) and the interaction with word frequency (b = -34.75, se = 15.91, t = -2.18) were
significant for the raw data, the interaction was not reliable for the model based on the log
transformed data. It is possible that the effects based on raw data were driven especially by
long total times (skewed distributions) such that the effect was attenuated for the models
based on log transformed data (see Balota, Aschenbrenner, & Yap, 2013). Reduced rereading in the Beyond wordn and Beyond wordn-1 conditions likely limited the effects of
word frequency on re-reading behavior, as reflected in the smaller effects of word
frequency in total time in these conditions compared to the control.
Discussion
Experiment 1 has important implication for the first key issue outlined in the
Introduction, the importance of the availability of previously read text for sentence
comprehension. Crucially, the results of Experiment 1 indicate that the availability of
wordn-1 during first-pass is important for comprehension. The significantly lower question
response accuracy for the Beyond wordn condition compared to the control is consistent
with the findings of Schotter et al. (2014). However there was no difference in question
response accuracy between the control and the Beyond wordn-1 condition, indicating that
the lower comprehension in the Beyond wordn condition is due to wordn-1 being
unavailable. The inability to visually re-sample wordn-1 after first-pass may have resulted
in some words not being accurately recognized in the Beyond wordn condition, leading to
a small but significant detriment in comprehension. However there was no difference in
question response accuracy between the Beyond wordn-1 and control condition. Therefore,
although words beyond wordn-1 are visually re-sampled if they are available to re-read
Spontaneous re-reading within sentences 20
(Booth & Weger, 2013), the results of Experiment 1 indicate that such visual re-sampling
may not always be essential to achieve a reasonable level of comprehension, at least as
measured by comprehension questions for quite simple sentences.
The permanent replacement of previously read text in Experiment 1 modulated rereading behavior. There were fewer trials including long-range regressions in the
replacement compared to control conditions. Also, the interactions with trial order and the
control vs. Beyond wordn contrast indicate that the effect of this manipulation on eye
movement behavior developed as the experiment progressed. Perhaps as participants learnt
that previously read text was replaced and would not re-appear (especially wordn-1 in the
Beyond wordn condition) they were less likely to make a regression to wordn-1 and reduced
their re-reading time, resulting in shorter sentence reading times. Experiments 2 and 3
manipulate the type and permanence of text replacement, and therefore provide more
direct tests of how parafoveal postview affects triggering of regressions, and how text
availability modulates eye movement control during re-reading.
Experiment 1 also shows that the availability of previously read text can modulate
other aspects of reading behavior. Although initial lexical processing of words appears to
be broadly similar across the display conditions (similar effects of word frequency on firstpass), overall there were longer first-pass fixation durations in the Beyond wordn
condition. Longer first-pass fixation durations could be due to the incorrect parafoveal
postview (Jordan et al., 2013, 2016), or they may compensate for shorter re-reading times.
Longer question response times may also compensate for reduced re-reading. These
differences may also reflect flexibility in eye movement control in response to display
format.
Spontaneous re-reading within sentences 21
Experiment 2
Experiments 2 and 3 further examine the role of previously read text beyond
wordn-1. As outlined in the Introduction, these experiments included manipulations of text
replacement type (word length vs. word shape) and permanence (temporary vs.
permanent), as illustrated in Figure 3. Experiment 1 examined how the availability of
previously read text affects sentence comprehension. The permanent text replacement
conditions in Experiments 2 and 3 are similar to the Beyond wordn-1 condition in
Experiment 1, which showed no difference in question response accuracy compared to the
control. Therefore the manipulations in Experiment 2 were not anticipated to have any
effect on question response accuracy. Nevertheless, if comprehension does depend on the
availability of meaningful text during re-reading, then an effect of text replacement
permanence might be predicted. That is, higher question response accuracy when
meaningful text is available (temporary replacement) compared to when it is permanently
replaced.
Insert Figure 3 here
Experiments 2 and 3 address the second and third key issues outlined in the
Introduction, that is, the effect of the availability of previously read text on triggering of
regressions and eye movement control during re-reading. If word shape cues in the
parafoveal postview modulate the likelihood of regressions then there should be a smaller
proportion of trials including regressions when only word length cues are available (word
Spontaneous re-reading within sentences 22
length replacement condition) compared to the word shape replacement condition. If eye
movement control during re-reading is driven by visual re-sampling of words then there
should be an effect of text replacement permanence on re-reading behavior, with longer
re-reading times in the temporary, compared to the permanent, text replacement
conditions.
Method
Participants. There were twenty-five participants in Experiment 2, other details are
the same as for Experiment 1.
Apparatus. The apparatus was the same as for Experiment 1.
Materials and design. There was a control condition, with no contingent changes,
and four conditions for which text beyond wordn-1 was replaced. Two variables were
manipulated in a 2 (text replacement type: word length, word shape) X 2 (text replacement
permanence: permanent, temporary) design. In the word length conditions letters were
replaced by “X”s. In the word shape conditions letters were replaced with other visually
similar letters (ascender replaces ascender etc.) (see Figure 3). There were 120
experimental items. The stimuli were straightforward single sentences, including items
adapted from Juhasz, Liversedge, White, and Rayner (2006) along with new items.
Participants pressed buttons to respond “yes” or “no” to comprehension questions. A
separate set of 18 participants were presented with only the questions, and were asked to
guess the correct answer. 54% of guesses were correct. These results demonstrate that it
was not possible to achieve a high comprehension score without reading the sentences.
The five conditions were manipulated within participants and items following a Latin
square design. Five lists of 144 sentences were constructed and five participants were
Spontaneous re-reading within sentences 23
randomly allocated to each list. Twenty-four of the sentences were filler items (with no
gaze contingent changes). The lists were presented in two separate blocks with the order
randomized within each block for each participant.
Procedure. A Bailey Lovie chart (Bailey & Lovie, 1976) was used to screen for
normal visual acuity. There were nine practice trials at the start of the experiment. In the
four text replacement conditions, once a saccade moved into a word on first-pass the
words beyond wordn-1 were replaced2. In the temporary replacement conditions text reappeared if there was a regression to a previous word. Otherwise the procedure was the
same as for Experiment 1.
Analyses. The initial analysis procedures were the same as for Experiment 1.
Overall 97% of trials were included in the analyses and one participant was replaced due
to >30% of trials being removed. For 80% of the trials included in the analyses there were
no blinks. The analysis procedures were similar to Experiment 1, employing linear mixed
effect models with maximal random effects structures. Sliding contrasts for the type and
permanence of text replacement were defined to examine the effects of text replacement
type and permanence, and the interaction, for the four text replacement conditions. In
addition, baseline models compared the control condition with each of the four text
replacement conditions. The hypotheses are primarily tested by examining the contrasts
for text replacement type and permanence. In addition, baseline models provide an
indication of whether replacement of previously read text affects reading behavior
compared to a normal text presentation. However, especially given the number of extra
contrasts generated, the baseline results are interpreted cautiously, with emphasis on the
results that are in line with the pattern shown by the 2X2 models.
Spontaneous re-reading within sentences 24
Results
Table 4 shows the means for each condition. The results for the baseline models
are shown in Table 5 and the results for the 2X2 models are shown in Table 6. For
question response accuracy, long-range regressions and re-reading fixation durations the
models including trial order did not provide better fits than the base model (χ2s < 5.6, ps >
0.06). For all other measures the models including an additive effect of trial order
provided better fits than the base model (χ2s > 5.1, ps < 0.05), but the models with the
interaction did not provide better fits than the additive models (χ2s < 7, ps > 0.13).
Insert Tables 4, 5 and 6 here
Question response time and accuracy. Question response times were significantly
shorter and question response accuracy was significantly higher in the control condition
compared to all four of the text replacement conditions but the 2X2 analyses showed no
effects of text replacement type, permanence, or an interaction. The accuracy results
contrast with Experiment 1 which showed no difference in response accuracy between the
control and the Beyond wordn-1 condition. Importantly, despite the small differences,
question response accuracy was very high across all the conditions.
Sentence reading time and first-pass measures. For sentence reading time the 2X2
analysis showed that sentence reading times were shorter when text was replaced with
word length compared to word shape information, there was no effect of permanence and
no interaction. Sentence reading times were also significantly shorter in the word length
Spontaneous re-reading within sentences 25
replacement conditions compared to the control. The results indicate that the presence of
word shape information beyond wordn-1 has a key influence in determining reading
behavior. The number of first-pass fixation durations was unaffected by display condition.
For first-pass fixation durations there was an effect of text replacement permanence, with
slightly longer first-pass fixations in the temporary compared to permanent replacement
conditions. First-pass fixation durations were also significantly longer in the temporary
replacement conditions compared to the control.
Re-reading. For the 2X2 models for all four of the regression and re-reading
measures (proportion of trials including a regression to wordn-1, proportion of trials
including a long range regression, re-reading time, re-reading fixation duration) there were
significant effects of text replacement type such that there were a higher proportion of
trials including a regression and longer re-reading time and fixation durations when the
text was replaced with word shape cues, compared to just word length cues. For all four
of these measures there was no effect of permanence and no interaction. In line with these
results, there were also significantly fewer long-range regressions and significantly shorter
re-reading times in the word length replacement conditions compared to the control. Also
note that for the proportion of trials including a regression to wordn-1, re-reading time, and
re-reading fixation duration, the contrast for control vs. permanent word shape
replacement was significant. The foveal presentation of nonwords (or orthographically
unfamiliar letter sequences) during re-reading could have inhibited processing of the text,
resulting in more re-reading behavior.
Discussion
Spontaneous re-reading within sentences 26
The question response accuracy scores in Experiment 2 relate to the first key issue
set out in the introduction, the importance of the availability of previously read text for
sentence comprehension. In Experiment 2 question response accuracy was very high
across all conditions, even for conditions for which re-reading time and sentence reading
times were reduced. However in contrast to the results for the Beyond wordn-1 condition in
Experiment 1, in Experiment 2 question response accuracy was slightly higher in the
control compared to the text replacement conditions, though there was no indication that
response accuracy was lower when meaningful text was unavailable during re-reading
(permanent replacement condition) compared to when it re-appeared (temporary
replacement condition). Nevertheless, the small differences in question response accuracy
raise the possibility that the availability of previously read text beyond wordn-1 (even just
the parafoveal postview) may affect sentence comprehension. The effect of text
replacement on question response accuracy is further explored in Experiment 3 for more
difficult sentences and questions.
Experiment 2 has important implications for the second key issue outlined in the
Introduction, that is, the effect of the parafoveal postview on triggering of regressions. In
Experiment 2 there was a clear effect of text replacement type, with a smaller proportion
of trials including regressions when the parafoveal postview beyond wordn-1 included only
word length cues compared to both word length and shape cues. In line with these results,
in Schotter et al.’s (2014) study there were also few regressions when only word length
cues were available to the left. It could be that word shape information beyond wordn-1
plays a key role in triggering or programming regressions. Another possibility is that the
word length text replacement cues ("X"s) were sufficiently salient that they suppressed
Spontaneous re-reading within sentences 27
triggering of regressions. Either way, the reduced likelihood of triggering a regression
resulted in reduced opportunity for re-reading, hence the shorter re-reading and sentence
reading times in the word length replacement conditions.
The third key issue, the effect of the availability of previously read text on eye
movement control during re-reading, is explored in Experiments 2 and 3 by examining
effects of text replacement permanence. However in Experiment 2 there were no effects of
text replacement permanence on re-reading behavior. The null effects could be due to
limited potential for re-reading in the word length replacement postview conditions, and
due to re-reading fixations being driven by fixation on orthographically unfamiliar
nonwords in the permanent word shape replacement condition. Although the percentage of
regressive fixations in Experiment 2 (control: 13%) was similar to normal (10-15%,
Rayner, 2009), the straightforward nature of the sentences could have resulted in relatively
rare incidences of integration failure. It could be that the availability of meaningful text is
particularly important in driving eye movement control during re-reading when text
integration fails. These issues are explored further in Experiment 3.
Experiment 3
As in Experiment 2, Experiment 3 included manipulations of the type and
permanence of text replacement. The sentences in Experiment 3 were more complex, but
sentence difficulty was not directly manipulated in order to minimize the number of
experimental conditions. The sentences included object relative clauses, as these are likely
to generate more regressions compared to simpler sentences. For example, Staub (2010)
showed higher rates of regressions for sentences with object, compared to subject, relative
Spontaneous re-reading within sentences 28
clauses (e.g. from the noun, proportion regressions: 0.4 vs. 0.16 respectively). Examples of
the Experiment 3 stimuli are shown in Table 7.
Insert Table 7 here
Experiment 3 addresses the three key issues as set out in the Introduction. First,
Experiment 3 provides a further test of whether the availability of previously read text is
important for comprehension. Visual re-sampling during re-reading may be more
important for more complex sentences, due to these sentences producing an increased
working memory load (Booth & Weger, 2013). If the availability of previously read text is
important for complex sentence comprehension then there should be an effect of text
replacement permanence on question response accuracy, with lower accuracy in the
permanent text replacement conditions. Question difficulty is also manipulated in
Experiment 3. One possibility is that shallower levels of comprehension (tested with easy
questions) are unaffected by text replacement, whereas deeper levels of comprehension
(tested with difficult questions) are impaired when meaningful text is unavailable during
re-reading.
Experiment 3 also provides a further test of the second key issue, the effect of the
parafoveal postview on triggering of regressions. One possibility is that regressions may
be triggered in more complex sentences regardless of the nature of the parafoveal
postview. Experiment 3 therefore examines whether the likelihood of regressions is
modulated by the availability of word shape cues in the parafoveal postview, even for
more difficult sentences.
Spontaneous re-reading within sentences 29
Finally, Experiment 3 provides a further test of the third key issue, whether the
availability of previously read text modulates eye movement control during re-reading. In
a self-paced reading study (combined with eye movement recordings) Kennedy and
Murray (1984) showed that there was greater sensitivity to syntactic ambiguity, and there
were more re-inspections, when previously read text remained available. Therefore the
presence of meaningful text may be especially important in driving eye movement
behavior when sentences are more difficult to process. That is, there may be an effect of
text replacement permanence, with longer re-reading times in the temporary, compared to
permanent, text replacement conditions.
Method
Participants. There were 30 participants in Experiment 3, the other details were the
same as for Experiment 1.
Apparatus. The apparatus was the same as for Experiment 1.
Materials and design. There was a control condition, with no contingent changes,
and four text replacement conditions for which two variables were manipulated in a 2 (text
replacement type: word length, word shape) X 2 (text replacement permanence:
permanent, temporary) design, as for Experiment 2. Participants pressed buttons to
respond “yes” or “no” to comprehension questions. Question difficulty was also
manipulated (2X2X2 design, plus two control conditions). The difficult questions assessed
comprehension of the referents within the object relative clauses whereas the easy
questions probed understanding of more straightforward aspects of the sentences (see
examples in Table 7). A separate set of 18 participants were presented with only the
questions, and were asked to guess the correct answer. 54% of guesses were correct for the
Spontaneous re-reading within sentences 30
easy questions and 56% of guesses were correct for the difficult questions. These results
demonstrate that it was not possible to achieve a high comprehension score without
reading the sentences.
There were 140 experimental items and the ten conditions were manipulated within
participants and items following a Latin square design. Ten lists of 309 sentences were
constructed and three participants were randomly allocated to each list. There were 169
filler items. 102 filler items included contingent changes similar to those for Experiment 3,
except that they took the form of blank, line or word length masks beyond wordn-1 (for
brevity these data are not reported here). 144 of the filler items had simple sentence
constructions. The lists were split randomly across three blocks of trials and randomized
for each participant within each block. 18 sentences were presented separately at the
beginning of the first session for practice.
Procedure. The procedure was the same as for Experiment 2.
Analyses. The analysis procedures were the same as for Experiment 2. Overall
95% of trials were included in the analyses. One participant was replaced due to >30% of
trials being removed. For 77% of the trials included in the analyses there were no blinks.
Results
Table 8 shows the means for each condition. The results for the baseline models
are shown in Table 9 and the results for the 2X2 models are shown in Table 10. Including
trial order did not improve model fit for question response accuracy, first-pass fixation
duration or re-reading fixation duration (χ2s < 9, ps > 0.14). For question response time,
number of first-pass fixations, proportion of trials with regressions to wordn-1, and
proportion of trials with long-range regressions, the models including an additive effect of
Spontaneous re-reading within sentences 31
trial order provided better fits than the base models (χ2s > 23, ps < 0.001), but the models
including the interaction did not provide better fits than the additive models (χ2s < 9, p >
0.08). For the 2X2 model for re-reading time the model including the interaction with trial
order did provide a significantly better fit than the additive model (χ2 = 10.42, p < 0.05).
The interactions with trial order are detailed in the text below. For other models (rereading time baseline, sentence reading time) including an additive effect of trial order
provided better fits than the base models (χ2s > 35, ps < 0.001). However although
including the interaction with trial order produced a marginally significant or significantly
better fit than the additive model for the models based on raw data (χ2s > 7, ps < 0.09),
including the interaction did not provide a significantly better fit for the log transformed
data. Therefore the additive model results are reported.
Insert Tables 8, 9 and 10 here
Question response time and accuracy. For both question response time and
accuracy, for both easy and difficult questions, there were no significant effects of text
replacement type or permanence or any significant differences compared to the control. In
line with Experiments 1 and 2, the results demonstrate that high comprehension question
response accuracy can be achieved even when previously read text beyond wordn-1 is
unavailable.
Sentence reading time and first-pass measures. For the 2X2 analyses, in line with
Experiment 2, sentence reading times were shorter when text was replaced with word
length compared to word shape information and there was no interaction with text
Spontaneous re-reading within sentences 32
replacement permanence. In line with these results, sentence reading times were
significantly shorter in the word length replacement conditions compared to the control. In
contrast to Experiment 2, for the analysis based on raw data, there was a significant main
effect of text permanence such that sentence reading times were longer when the text
replacement was temporary compared to permanent. This result is discussed further in
conjunction with re-reading times below. For the first-pass measures there were no
significant effects of text replacement type or permanence and no interactions. However
first-pass fixation durations were significantly longer for all four of the text replacement
conditions compared to the control.
Re-reading. For the 2X2 models for the proportion of trials including a regression
to wordn-1, the proportion of trials including a long range regression, and re-reading time
there were significant effects of text replacement type. There were a smaller proportion of
trials including a regression and shorter re-reading time when the text was replaced with
word length cues, compared to word shape cues (also compared to the control). These
effects of text replacement type are in line with the results of Experiment 2 (though in
Experiment 3 there was no effect of text replacement type for re-reading fixation
duration). There was also a significant interaction between trial order and the effect of text
replacement type for re-reading time (b = -1.22, se = 0.35, t = -3.50) but no interactions
between trial order and the effect of text replacement permanence or the interaction
between type and permanence of text replacement (ts < 1). The interaction with text
replacement type is illustrated by the Loess curves in Figure 4. The Figure indicates that
there were consistently short re-reading times throughout the experiment when there was a
word length replacement. In contrast, when there was a word shape replacement re-
Spontaneous re-reading within sentences 33
reading times were longer at the start of the experiment, and became progressively shorter
over the course of the experiment (similar to the pattern as indicated by the additive effects
of trial order for other measures).
Insert Figure 4 here
The measures for the proportion of trials including a regression produced no
effects of text replacement permanence and no interactions with text replacement type.
However, in line with the results for sentence reading time, the 2X2 analyses based on raw
data showed significant effects of text replacement permanence for re-reading time and rereading fixation duration, and no interactions with text replacement type. There were
longer re-reading times and shorter fixation durations in the temporary compared to the
permanent text replacement conditions. However, as for sentence reading time, these
effects were not reliable for the models based on log transformed data. Boxplots for all
three of these measures are shown in Figure 5. The boxplots indicate that the longer
sentence and re-reading times in the temporary replacement condition, and longer rereading fixation durations in the permanent replacement condition, are driven by a greater
degree of skew. Differences that arise from such skewed distributions are likely to be
attenuated for the models based on log transformed data (see Balota et al., 2013).
Interestingly, the pattern of results indicates that the effects of text replacement
permanence may hold only for a subset of cases for which fixations or reading times are
much longer. For example, it could be that the availability of meaningful text during rereading only drives re-reading behavior (increasing re-reading and sentence reading time)
Spontaneous re-reading within sentences 34
when integration fails. Similarly, the permanent text replacements may have only
lengthened re-reading fixation durations compared to the temporary text replacement
conditions for a subset of cases, perhaps if recovery from integration failure was hindered
by the absence of meaningful text.
Insert Figure 5 here
Discussion
In contrast to Experiment 2 there were no significant differences in question
response accuracy in the text replacement conditions compared to the control. In
Experiment 3, even when the sentences and questions are difficult, question response
accuracy is similar to normal when meaningful text is removed beyond wordn-1. Therefore,
in relation to the first key issue set out in the Introduction, the results of Experiment 3
provide no evidence that the availability of previously read text is important for sentence
comprehension.
In relation to the second key issue, consistent with the results of Experiment 2, the
nature of the parafoveal postview is clearly important in determining the likelihood of
regressions. There were more regressions when text beyond wordn-1 was replaced with
word shape compared to word length information. Word shape information may have a
key role in triggering or programming regressions, or it could be that the word length text
replacement cues (“X”s) suppressed triggering of regressions. Text replacement type also
affected re-reading times, similar to Experiment 2, the shorter re-reading times might at
least in part be due to the reduced likelihood of initiating re-reading.
Spontaneous re-reading within sentences 35
In contrast to Experiment 2, for the analyses based on raw data there were effects
of text replacement permanence for sentence reading time, re-reading time and the number
of re-reading fixation durations. These results are crucial for the third key issue set out in
the Introduction, they indicate that, at least for more complex sentences, the availability of
meaningful text during re-reading is important in driving eye movement behavior during
re-reading. Also note that first-pass fixation durations were longer in all four of the text
replacement conditions compared to the control. These differences may be due to the
parafoveal postview to the left (Jordan et al., 2013, 2016) perhaps in particular prior to
planned regressions (Apel et al., 2012). However these differences may also reflect
flexibility in eye movement behavior in response to the availability of previously read text.
General Discussion
Together the experiments yield three key findings: First, Experiment 1
demonstrates that the availability of wordn-1 is especially important for sentence
comprehension. Second, Experiments 2 and 3 demonstrate that the availability of word
shape cues in the parafoveal postview are important in triggering of regressions. Third, the
results of Experiment 3 indicate that, at least for more complex sentences, the availability
of meaningful text is important in driving eye movement behavior during re-reading.
Sentence comprehension
The results of Experiment 1 demonstrated, in line with Schotter et al. (2014), that
removing text permanently beyond wordn is detrimental for comprehension. However in
contrast, Experiment 1 showed that removing text permanently beyond wordn-1 had no
Spontaneous re-reading within sentences 36
effect on question response accuracy, indicating that the availability of wordn-1 is
important for comprehension. Continued orthographic and lexical processing of wordn-1,
both during parafoveal postview (Binder et al., 1999) and re-reading (Bicknell & Levy,
2011; Reichle et al., 1998; Vitu, 2005), may be particularly important for accurate word
recognition and therefore full sentence comprehension.
Experiments 2 and 3 showed no reduction in comprehension question response
accuracy when meaningful text remained unavailable during re-reading compared to when
it re-appeared. Re-reading beyond wordn-1 may be triggered by attention shifting back to
previously read text, rather than necessarily a need to visually re-sample previously read
text. Ordinarily, eye movement behavior during re-reading may be driven relatively
automatically, with visual re-sampling triggering lexical processing, saccade programming
and attention shifts to each word, similar to first-pass reading (Reichle et al., 2009). In
Booth and Weger’s (2013) study it is the re-sampled word that is integrated, participants
were generally unaware of the inconsistency between first-pass and second-pass text (see
also Sheridan & Reingold, 2012). Therefore it could be that the re-reading process “overwrites” the first-pass reading. That is, re-reading may involve re-sampling, re-recognition,
and re-integration of the text, even though these processes may not always be necessary to
further enhance comprehension (as long as first-pass recognition of words is accurate).
Nevertheless, the availability of previously read text beyond wordn-1 may still have
a role in sentence comprehension. Note that the removal of previously read text (even just
an incorrect parafoveal postview) for the simple sentences in Experiment 2 could have
contributed to slightly lower question response accuracy. Also, longer question response
times could be due to readers guessing the answers to comprehension questions, rather
Spontaneous re-reading within sentences 37
than making an informed decision (Logačev & Vasishth, 2016). Standard question
response accuracy scores may also fail to reflect differences in comprehension. The
norming studies demonstrated that the questions for all three experiments could not be
responded to accurately without presentation of the sentences. However it could be that
the level of comprehension probed by the questions, especially in Experiments 1 and 2,
may have required only a cursory reading of the sentences. Also in Experiment 3, repeated
reading of the same sentence structure might have facilitated comprehension of the object
relative clauses, or may have encouraged careful first-pass reading (see below). Future
studies may examine comprehension with stimuli and questions that are very carefully
designed to detect subtle differences, ideally generating a broad range of comprehension
levels that might better differentiate standards of comprehension. Further research may
also examine how re-reading and the availability of previously read text is linked to
comprehension specifically for cases where re-reading occurs (see Christiansen et al.,
2016; Schotter et al., 2014). Further studies may examine how re-reading of previously
read text may modulate confidence in text comprehension or integration of text with prior
knowledge. Visual re-sampling may also be key for integration with discourse context in
longer texts. It could be that working memory is sufficient for retaining the key content of
single sentences during first-pass, whereas the details for longer and more complex texts
may be more dependent on the “external memory” (O’Regan, 1992) of the text itself.
It could also be that the frequency and difficulty of the questions in the present
study encouraged a relatively cautious reading strategy (McConkie & Rayner, 1974;
McConkie, Rayner, & Wilson, 1973; Wotschack & Kliegl, 2013). For example, it could be
that a high threshold for confidence in recognition of each word (Bicknell & Levy, 2011)
Spontaneous re-reading within sentences 38
was required before leaving each word on first-pass. Visual re-sampling of previously read
text within a sentence may not be necessary when first-pass reading is very accurate. In
contrast, it could be that visual re-sampling during re-reading may be much more critical
for comprehension for less cautious reading for which recognition of words during firstpass may not always be accurate (see Bicknell & Levy, 2010). For example, the presence
of meaningful text during re-reading may be especially important when accurate encoding
in memory fails, for example, following mind wandering (Reichle, Reineberg, & Schooler,
2010) or when a word beyond wordn-1 in the text is initially misidentified, for example,
mistaken for a word neighbor (Bicknell & Levy, 2010, 2011; Levy, Bicknell, Slattery, &
Rayner, 2009). As Booth and Weger (2013) suggest, readers may simply learn that they
can depend on re-processing the visual input during re-reading.
The parafoveal postview
The results of Experiments 2 and 3 demonstrate that triggering of regressions can
be modulated by the parafoveal postview of previously read text. There were fewer
regressions and reduced re-reading behavior when text beyond wordn-1 was replaced with
only word length cues (“X”s). One possibility is that the visually salient word length text
replacements used here and in Schotter et al.’s study were sufficiently visually salient that
programming of regressive eye movements was suppressed. Jordan et al. (2016) showed a
slight increase in the number of regressions for a visually similar postview, however the
present study showed no significant increase in the likelihood of regressions for a visually
similar (word shape) postview. Nevertheless, in both studies the differences were small
and measures for the number or likelihood of regressions was broadly similar to normal.
Spontaneous re-reading within sentences 39
Future studies that manipulate text availability during re-reading should therefore aim to
retain word shape information prior to a regression, such that the frequency of regressions,
and therefore the opportunity for re-reading, is not substantially reduced.
Note that in the present study the visual word shape information always
corresponded to that of the words in the text. However Jordan et al.’s (2016) study showed
very little effect of the visual similarity of parafoveal postviews compared to the control.
Therefore it could be that any parafoveal postview that has a configuration similar to
standard text (that is, with varying word shape, but not necessarily matching to the shape
of words within a specific sentence) may enable triggering of regressions similar to
normally presented text. Also note that the gaze contingent change technique employed
here produces large and repeated changes to the text which are examined using global
dependent measures, averaging across reading behavior for the entire sentence. Future
studies that employ more subtle manipulations, such as a single boundary contingent
change within each sentence, may ultimately reveal more subtle effects of parafoveal
postview on the control of regressions and re-reading behavior.
Eye movement control during re-reading
The present study also has important implications for the mechanisms underlying
eye movement control during re-reading. For the more complex sentences in Experiment 3
(for analyses based on raw data) there were effects of text replacement permanence such
that there were longer sentence and re-reading times when the meaningful text re-appeared
during re-reading compared to when it remained unavailable. These results may hold
especially for a subset of cases associated with longer re-reading times, for example, due
Spontaneous re-reading within sentences 40
to integration failure. The results are perhaps consistent with a study by Kennedy and
Murray (1984) which showed greater sensitivity of eye movement behavior to syntactic
ambiguity when previously read text was available. Therefore the presence of meaningful
text is likely to be important in driving eye movement behavior during re-reading,
especially for more difficult sentences (see also Lantz, White, & Paterson, 2013).
Models of eye movement control during reading so far provide only a very basic
account of the mechanisms involved in re-reading beyond wordn-1 (Engelmann et al.,
2013; Reichle et al., 2009). Once a regression is triggered, previously read words may be
automatically visually re-sampled (Booth & Weger, 2013), and these re-samples may in
turn automatically trigger word recognition and integration processes similar to first-pass
reading. That is, the mechanisms underlying re-reading behavior may be triggered
automatically and driven by the visual re-samples and any processing difficulty caused by
those re-samples, rather than necessarily being required for comprehension. In contrast the
absence of visual re-samples in the permanent text replacement conditions (especially
Experiment 3) may have failed to activate the mechanisms for triggering eye movements
during re-reading, resulting in much shorter re-reading times. Future research will need to
examine whether the mechanisms underlying eye movement control during re-reading are
indeed very similar to those for eye movement behavior during first-pass reading, or
whether the mechanisms underlying eye movement control during re-reading operate
differently.
Interestingly in the present study, high comprehension question response accuracy
was maintained even when the time spent re-reading was significantly reduced. However
in these conditions, compared to the control condition, first-pass fixation durations and
Spontaneous re-reading within sentences 41
comprehension question response times were sometimes longer. It could be that when
fewer regressions are triggered (for example, due to previously read words replaced with
“X”s) extra processing time may just be taken in other forms, such as slower first-pass
reading. For the complex sentences in Experiment 3, it could be that readers may have
used the extended first-pass reading times to store a verbal representation of the sentence
in working memory, in order to reduce reliance on re-sampling of the words during rereading. These differences in behavior are indicative of flexibility in the mechanisms
underlying eye movement control during reading, such that a good level of comprehension
can be achieved despite differences in reading behavior.
To summarize, the present study demonstrates the importance of the availability of
wordn-1 for sentence comprehension, the importance of the nature of the parafoveal
postview for triggering regressions, and the importance of the availability of previously
read text in driving eye movement control during re-reading. Future work should examine
these issues for re-reading across multiple sentences. Furthermore, some groups of
readers, such as beginner readers (Blythe & Joseph, 2011), and older readers (Rayner,
Reichle, Stroud, Williams, & Pollatsek, 2006), produce more regressions than young adult
skilled readers. The nature of eye movement behavior during re-reading may also be
different for poorer readers (Murray & Kennedy, 1988). Examining the role of previously
read text for comprehension and eye movement control may therefore be crucial for
revealing the mechanisms underlying reading from a developmental perspective, as well
as for those with reading difficulties.
Spontaneous re-reading within sentences 42
References
Altmann, G. T. M. (2004). Language-mediated eye movements in the absence of a visual
world: The “blank screen paradigm. Cognition, 93, B79–B87.
http://dx.doi.org/10.1016/j.cognition.2004.02.005
Apel, J.K., Henderson, J.M., & Ferreira, F. (2012). Targeting regressions: Do readers pay
attention to the left? Psychonomic Bulletin & Review, 19, 1108-1113.
http://dx.doi.org/10.3758/s13423-012-0291-1
Baayen, R.H., Davidson, D.J., & Bates, D.M. (2008). Mixed-effects modeling with
crossed random effects for subjects and items. Journal of Memory and Language,
59, 390-412.
http://dx.doi.org/10.1016/j.jml.2007.12.005
Bailey, I.L., & Lovie, J.E. (1976). New design principles for visual acuity letter charts.
American Journal of Optometry and Physiological Optics, 53, 740-745.
Balota, D.A., Aschenbrenner, A.J., & Yap, M.J. (2013). Additive effects of word
frequency and stimulus quality: The influence of trial history and data
transformations. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 39, 1563-1571. http://dx.doi.org/10.1037/a0032186
Ballard, D. H., Hayhoe, M. M., Pook, P. K., & Rao, R. P. N. (1997). Deictic codes for the
embodiment of cognition. Behavioral and Brain Sciences, 20, 723–767.
http://dx.doi.org/10.1017/S0140525X97001611
Barr, D.J., Levy, R., Scheepers, C., & Tily, H.J. (2013). Random effects structure for
confirmatory hypothesis testing: Keep it maximal. Journal of Memory and
Language, 68, 255–278. http://dx.doi.org/10.1016/j.jml.2012.11.001
Spontaneous re-reading within sentences 43
Bates, D., Mächler, M., & Bolker, B. (2011). LME4: linear mixed-effects models using S4
classes. http://cran.R-project.org/package=lme4.
Benedetto, S., Carbone, A., Pedrotti, M., Le Fevre, K., Bey, L.A.Y., & Baccino, T. (2015).
Rapid serial visual presentation in reading: The case of Spritz. Computers in
Human Behavior, 45, 352-358. http://dx.doi.org/10.1016/j.chb.2014.12.043
Bicknell, K., & Levy, R. (2010). A rational model of eye movement control in reading.
In J. Hajič (Ed.), Proceedings of the 48th Annual Meeting of the Association for
Computational Linguistics. (pp. 1168–1178). Uppsala, Sweden: Association for
Computational Linguistics.
Bicknell, K., & Levy, R. (2011). Why readers regress to previous words: A statistical
analysis. Proceedings of the 33rd Annual Conference of the Cognitive Science
Society, pp. 931–936.
Binder, K.S., Pollatsek, A., & Rayner, K. (1999). Extraction of information to the left of
the fixated word in reading. Journal of Experimental Psychology: Human
Perception and Performance, 25, 1162-1172.
http://dx.doi.org/10.1037/0096-1523.25.4.1162
Blythe, H.I., & Joseph, S.S.L. (2011). Children’s eye movements during reading. In S. P.
Liversedge, I. D. Gilchrist, & S. Everling (Eds.), Oxford handbook on eye
movements (pp. 643-662). Oxford, England: Oxford University Press.
http://dx.doi.org/10.1093/oxfordhb/9780199539789.001.0001
Booth, R.W., & Weger, U. W. (2013). The function of regressions in reading: Backward
eye movements allow rereading. Memory & Cognition, 41, 82-97.
http://dx.doi.org/10.3758/s13421-012-0244-y
Spontaneous re-reading within sentences 44
Christiansen, K., Luke, S.G., Hussey, E.K., Wochna, K.L. (2016). Why reread? Evidence
from garden-path and local coherence structures, The Quarterly Journal of
Experimental Psychology, http://dx.doi.org/10.1080/17470218.2016.1186200
Clifton, C., Jr., Staub, A., & Rayner, K. (2007). Eye movements in reading words and
sentences. In: R. van Gompel (Ed.), Eye movements: A window on mind and brain
(pp. 341-372). Amsterdam: Elsevier.
Engbert, R., Longtin, A., & Kliegl, R. (2002). A dynamical model of saccade generation in
reading based on spatially distributed lexical processing. Vision Research, 42, 621636. http://dx.doi.org/10.1016/S0042-6989(01)00301-7
Engelmann, F., Vasishth, S., Engbert, R., & Kliegl, R. (2013). A framework for modeling
the interaction of syntactic processing and eye movement control. Topics in
Cognitive Science, 5, 452-474. http://dx.doi.org/10.1111/tops.12026
Ferreira, F., Apel, J., & Henderson, J. M. (2008). Taking a new look at looking at nothing.
Trends in Cognitive Sciences, 12, 405–410.
http://dx.doi.org/10.1016/j.tics.2008.07.007
Ferris, F. L., & Bailey, I. L. (1996). Standardizing the measurement of visual acuity for
clinical research studies. Ophthalmology, 103, 181–182.
http://dx.doi:10.1016/ S0161-6420(96)30742-2
Fischer, M. H. (2000). Perceiving spatial attributes of print. In: A. Kennedy, R. Radach, D.
Heller, & J. Pynte (Eds.), Reading as a perceptual process (pp. 89-117). North
Holland: Elsevier.
Frazier, L., & Rayner, K. (1982). Making and correcting errors during sentence
Spontaneous re-reading within sentences 45
comprehension: Eye movements in the analysis of structurally ambiguous
sentences. Cognitive Psychology, 14, 178–210.
http://dx.doi.org/10.1016/0010-0285(82)90008-1
Guérard, K., Saint-Aubin, J., & Maltais, M. (2013). The role of verbal memory in
regressions during reading. Memory & Cognition, 41, 122-136.
http://dx.doi.org/10.3758/s13421-014-0424-z
Guérard, K., Saint-Aubin, J., Maltais, M., & Lavoie (2014). The role of verbal memory in
regressions during reading is modulated by the target word’s recency in memory.
Memory & Cognition, 42, 1155-1170.
http://dx.doi.org/10.3758/s13421-014-0424-z
Harvey, H., & Walker, R. (2014). Reading with peripheral vision: A comparison of
reading dynamic scrolling and static text with a simulated central scotoma. Vision
Research, 98, 54-60. http://dx.doi.org/10.1016/j.visres.2014.03.009
Hyönä, J., & Niemi, P. (1990). Eye movements during repeated reading of text. Acta
Psychologica, 73, 259-280. http://dx.doi.org/10.1016/0001-6918(90)90026-C
Inhoff, A.W., Greenberg, S.N., Solomon, M., & Wang, C.-A. (2009). Word integration
and regression programming during reading: A test of the E-Z Reader 10 model.
Journal of Experimental Psychology: Human Perception and Performance, 35,
1571-1584. http://dx.doi.org/10.1037/a0014250
Inhoff, A. W., & Weger, U. (2005). Memory for words location during reading: Eye
movements to previously read words are spatially selective but not precise.
Memory & Cognition, 33, 447–461. http://dx.doi.org/10.3758/BF03193062
Jordan, T.R., McGowan, V.A., Kurtev, S., & Paterson, K.B. (2016). A further look at
Spontaneous re-reading within sentences 46
postview effects in reading: An eye-movements study of influences from the left of
fixation. Journal of Experimental Psychology: Learning, Memory, and Cognition,
42, 296-307. http://dx.doi.org/10.1037/xlm0000157
Jordan, T. R., McGowan, V. A., & Paterson, K. B. (2013). What’s left? An eye-movement
study of the influence of interword spaces to the left of fixation during reading.
Psychonomic Bulletin & Review, 20, 551-557.
http://dx.doi.org/10.3758/s13423-012-0372-1
Juhasz, B.J., Liversedge, S.P., White, S.J., & Rayner, K. (2006). Binocular coordination of
the eyes during reading: Word frequency and case alternation affect fixation
duration but not fixation disparity. Quarterly Journal of Experimental Psychology,
59, 1614-1625. http://dx.doi.org/10.1080/17470210500497722
Just, M.A., & Carpenter, P.A. (1980). A theory of reading: From eye fixations to
comprehension. Psychological Review, 87, 329-354.
http://dx.doi.org/10.1037/0033-295X.87.4.329
Kennedy, A. (1982). Eye movements and spatial coding in reading. Psychological
Research, 44, 313-322.
Kennedy, A. (1983). On looking into space. In: K. Rayner (Ed.), Eye movements in
reading: Perceptual and language processes (pp. 237-251.) New York: Academic
Press.
Kennedy, A. (1992). The spatial coding hypothesis. In: K. Rayner (Ed.). Eye movements
and visual cognition (pp. 379-396.). New York: Springer-Valag.
Kennedy, A., Brooks, R., Flynn, L.-A., & Prophet, C. (2003). The reader’s spatial code. In
Spontaneous re-reading within sentences 47
J. Hyönä, R. Radach, & H. Deubel (Eds). The mind's eye: Cognitive and applied
aspects of eye movement research (pp. 193-212). Amsterdam: Elsevier.
http://dx.doi.org/10.1016/B978-044451020-4/50012-8
Kennedy, A., & Murray, W. S. (1984). Inspection times for words in syntactically
ambiguous sentences under three presentation conditions. Journal of Experimental
Psychology: Human Perception and Performance, 10, 833-849.
http://dx.doi.org/10.1037/0096-1523.10.6.833
Kennedy, A., & Murray, W. S. (1987). Spatial coordinates and reading: Comments on
Monk (1985). Quarterly Journal of Experimental Psychology , 39A, 649-656.
http://dx.doi.org/10.1080/14640748708401807
Lantz, L.M.T., White, S.J., & Paterson, K.B. (2013). Spontaneous re-reading of
structurally ambiguous sentences: Eye movement behaviour and visual sampling.
(2013). In K. Holmqvist, F. Mulvey & R. Johansson (Eds.), Book of Abstracts of
the 17th European Conference on Eye Movements, 11-16 August 2013, in Lund,
Sweden. Journal of Eye Movement Research, 6(3), 322.
Levy, R., Bicknell, K., Slattery, T., & Rayner, K. (2009). Eye movement evidence that
readers maintain and act on uncertainty about past linguistic input. Proceedings of
the National Academy of Sciences of the United States of America, 106, 21086–
21090. http://dx.doi.org/10.1073/pnas.0907664106
Logačev, P., & Vasishth, S. (2016). Understanding underspecification: A comparison of
two computational implementations. The Quarterly Journal of Experimental
Psychology, 69, 996-1012. http://dx.doi.org/10.1080/17470218.2015.1134602
MacLeod, C. M. (1991). Half a century of research on the Stroop effect: An integrative
Spontaneous re-reading within sentences 48
review. Psychological Bulletin, 109, 163– 203.
http://dx.doi.org/10.1037/0033-2909.109.2.163
McConkie, G. W., & Rayner, K. (1974). Investigation of reading strategies: I.
Manipulating strategies through payoff conditions. Journal of Literacy Research,
6, 9–18. http://dx.doi.org/10.1080/10862967409547072
McConkie, G. W., & Rayner, K. (1975). The span of the effective stimulus during a
fixation in reading. Perception & Psychophysics, 17, 578-586.
McConkie, G. W., & Rayner, K. (1976). Asymmetry of perceptual span in reading.
Bulletin of the Psychonomic Society, 8, 365-368.
McConkie, G. W., Rayner, K., & Wilson, S. J. (1973). Experimental manipulation of
reading strategies. Journal of Educational Psychology, 65, 1–8.
http://dx.doi.org/10.1037/h0034822
McGowan, V.A., White, S.J., & Paterson, K.B. (2013). Spatial representations of
previously encountered text: Processing to the left of fixation during reading. In K.
Holmqvist, F. Mulvey & R. Johansson (Eds.), Book of Abstracts of the 17th
European Conference on Eye Movements, 11-16 August 2013, in Lund, Sweden.
Journal of Eye Movement Research, 6(3), 423.
Mitchell, D.C., Shen, X., Green, M.J., & Hodgson, T.L. (2008). Accounting for regressive
eye-movements in models of sentence processing: A reappraisal of the selective
reanalysis hypothesis. Journal of Memory and Language, 59, 266-293.
http://dx.doi.org/10.1016/j.jml.2008.06.002
Murray, W.S. & Kennedy, A. (1988). Spatial coding in the processing of anaphor by good
Spontaneous re-reading within sentences 49
and poor readers: Evidence from eye movement analyses. Quarterly Journal of
Experimental Psychology, 40A, 693- 718.
http://dx.doi.org/10.1080/14640748808402294
Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of
inhibitionless spreading activation and limited-capacity attention. Journal of
Experimental Psychology: General, 106, 226–254.
http://dx.doi.org/10.1037/0096-3445.106.3.226
O’Regan, J.K. (1992). Solving the “real” mysteries of visual perception: The world as an
outside memory. Canadian Journal of Psychology, 46, 461-488.
http://dx.doi.org/10.1037/h0084327
R Core Team (2016). R: A language and environment for statistical
computing. R Foundation for Statistical Computing, Vienna, Austria.
http://www.R-project.org
Raney, G.E., & Rayner, K. (1995). Word frequency effects and eye movements during
two readings of a text. Canadian Journal of Experimental Psychology, 49, 151173. http://dx.doi.org/10.1037/1196-1961.49.2.151
Rayner, K. (2009). Eye movements and attention in reading, scene perception, and visual
search. Quarterly Journal of Experimental Psychology, 62, 1457-1506.
http://dx.doi.org/10.1080/17470210902816461
Rayner, K. (2014). The gaze-contingent moving window in reading: Development and
review. Visual Cognition, 22, 242-258.
http://dx.doi.org/10.1080/13506285.2013.879084
Rayner, K., Castelhano, M. S., & Yang, J. (2009). Eye movements and the perceptual span
Spontaneous re-reading within sentences 50
in older and younger readers. Psychology and Aging, 24, 755–760.
http://dx.doi.org/10.1037/a0014300
Rayner, K., Reichle, E.D., Stroud, M.J., Williams, C.C., & Pollatsek, A. (2006). The effect
of word frequency, word predictability, and font difficulty on the eye movements
on young and older readers. Psychology and Aging, 21, 448-465.
http://dx.doi.org/10.1037/0882-7974.21.3.448
Rayner, K., Well, A. D., & Pollatsek, A. (1980). Asymmetry of the effective visual field in
reading. Perception & Psychophysics, 27, 537-544.
http://dx.doi.org/10.3758/BF03198682
Rayner, K., Yang, J., Schuett, S., Slattery, T.J. (2014). The effect of foveal and parafoveal
masks on the eye movements of older and younger readers. Psychology and Aging,
29, 205-212. http://dx.doi.org/10.1037/a0036015
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.
http://dx.doi.org/10.1037/0033-295X.105.1.125
Reichle, E.D., Reineberg, A.E., & Schooler, J.W. (2010). Eye movements during mindless
reading. Psychological Science, 21, 1300-1310.
http://dx.doi.org/10.1177/0956797610378686
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, 1–20. http://dx.doi.org/10.3758/PBR.16.1.1
Richardson, D. C., & Spivey, M. J. (2000). Representation, space and Hollywood Squares
Looking at things that aren’t there anymore. Cognition, 76, 269–295.
Spontaneous re-reading within sentences 51
http://dx.doi.org/10.1016/S0010-0277(00)00084-6
Roy-Charland, A., Saint-Aubin, J., Klein, R.M., MacLean, G.H., Lalande, A., & Belanger,
A. (2012). Eye movements when reading: The importance of the word to the left of
fixation. Visual Cognition, 20, 328-355.
http://dx.doi.org/10.1080/13506285.2012.667457
Schotter, E.R., Tran, R., & Rayner, K. (2014). Don’t believe what you read (only once):
Comprehension is supported by regressions during reading. Psychological Science,
25, 1218-1226. http://dx.doi.org/10.1177/0956797614531148
Sheridan, H., & Reingold, E. M. (2012). Perceptually specific and perceptually nonspecific influences on rereading benefits for spatially transformed text: Evidence
from eye movements. Consciousness and Cognition, 21, 1739-1747.
http://dx.doi.org/10.1016/j.concog.2012.10.002
Staub, A. (2010). Eye movements and processing difficulty in object relative clauses.
Cognition, 116, 71-86. http://dx.doi.org/10.1016/j.cognition.2010.04.002
Van Heuven, W.J.B., Mandera, P., Keuleers, E., & Brysbaert, M. (2014). Subtlex-UK: A
new and improved word frequency database for British English. Quarterly Journal
of Experimental Psychology, 67, 1176-1190.
http://dx.doi.org/10.1080/17470218.2013.850521
Vitu, F. (2005). Visual extraction processes and regressive saccades in reading. In G.
Underwood (Ed.), Cognitive processes in eye guidance (pp. 1-32). Oxford, NY:
Oxford University Press.
von der Marlsburg, T., Kliegl, R., & Vasishth, S. (2015). Determinants of scanpath
Spontaneous re-reading within sentences 52
regularity in reading. Cognitive Science, 39, 1675-1703.
http://dx.doi.org/10.1111/cogs.12208
von der Marlsburg, T., & Vasishth, S. (2011). What is the scanpath signature of syntactic
reanalysis? Journal of Memory and Language, 65, 109-127.
http://dx.doi.org/10.1016/j.jml.2011.02.004
Wang, C.-A., Tsai, J.-L., Inhoff, A.W., & Tzeng, O.J.L. (2009). Acquisition of linguistic
information to the left of fixation during the reading of Chinese text. Language and
Cognitive Processes, 24, 1097-1123.
http://dx.doi.org/10.1080/01690960802525392
Warren, T., White, S.J., & Reichle, E.D. (2009). Investigating the causes of wrap-up
effects: Evidence from eye movements and E-Z Reader. Cognition, 111, 132-137.
http://dx.doi.org/10.1016/j.cognition.2008.12.011
Weger, U.W., & Inhoff, A.W. (2007). Long-range regressions to previously read words
are guided by spatial and verbal memory. Memory & Cognition, 35, 1293-1306.
http://dx.doi.org/10.3758/BF03193602
White, S.J., Drieghe, D., Liversedge, S.P., & Staub, A. (under revision). The word
frequency effect during sentence reading: A linear or nonlinear effect of log
frequency? Quarterly Journal of Experimental Psychology
Wotschack, C., & Kliegl, R. (2013). Reading strategy modulates parafoveal-on-foveal
effects in sentence reading. The Quarterly Journal of Experimental Psychology,
66, 548–562. http://dx.doi.org/10.1080/17470218.2011.625094
Spontaneous re-reading within sentences 53
Footnotes
1
The maximal models for the critical word measures in Experiment 1 did not
converge (first fixation duration: correlation removed from items random effects structure;
gaze duration and total time: word frequency removed from items random effects
structure).
2
In Experiments 2 and 3 consecutive short (≤ 3 letter) words were considered as a
single word region, with the contingent change boundary located at the beginning of the
first word. Consequently in such cases more than one word to the left of the fixated word
was displayed during first-pass. For consistency, the same regions were employed for the
purpose of analysis, hence fixations within a region such as “so he” would be considered
first-pass before the eyes moved to the left or right of it.
Spontaneous re-reading within sentences 54
Acknowledgments
This research was supported by a Nuffield Foundation Undergraduate Research Bursary
and by a Small Grant from the Experimental Psychology Society (awarded to the first
author). The authors would like to thank SR Research for assistance in programming the
experiments. The authors would also like to thank Albrecht Inhoff and two anonymous
reviewers for their very helpful comments on a previous version of the manuscript.
Spontaneous re-reading within sentences
Table 1. Experiment 1: Means for Question Response Accuracy, Question Response Time, Sentence Reading Time, First-pass
and Re-reading Measures. Standard Deviations Shown in Parentheses.
Experiment 1
Control
Beyond wordn
Beyond wordn-1
Question response accuracy
0.94 (0.25)
0.88 (0.33)
0.94 (0.24)
Question response time
2812 (1413)
2984 (1564)
2847 (1413)
Sentence reading time
2828 (1098)
2779 (1030)
2743 (964)
Number of first-pass fixations
8.62 (1.77)
8.71 (1.80)
8.61 (1.75)
First-pass fixation duration
217 (37)
225 (38)
218 (38)
Regressiona to wordn-1
0.50 (0.50)
0.54 (0.50)
0.50 (0.50)
Long-range regressiona
0.70 (0.46)
0.58 (0.49)
0.67 (0.47)
Re-reading timeb
894 (828)
761 (750)
799 (666)
Re-reading fixation duration
243 (87)
250 (89)
260 (103)
Note. a Proportion of trials with at least one regression
b
Re-reading time for trials for which re-reading occurred
55
Spontaneous re-reading within sentences
56
Table 2. Experiment 1: Linear Mixed Effect Model Statistics for Question Response Accuracy, Question Response Time,
Sentence Reading Time, First-pass and Re-reading Measures.
Experiment 1
Intercept
Control vs.
Beyond
wordn
Control vs.
Beyond
wordn-1
Trial order
b
se
t/z
b
se
t/z
b
se
t/z
b
se
t/z
Question
response
accuracy
3.74
0.25
14.94 *
-0.94
0.17
-5.46 *
0.08
0.19
0.39
-
Question
response
time
2808.87
128.75
21.82 *
174.94
55.51
3.15 *
31.65
53.23
0.59
-2.71
0.33
-8.18 *
Sentence
reading
time
2826.94
134.48
21.02 *
-47.51
69.17
-0.69
-86.71
62.43
-1.39
-3.70
0.38
-9.65 *
Number of
first-pass
fixations
8.63
0.23
37.18 *
0.09
0.08
1.14
-0.01
0.06
-0.21
-0.002
0.0003
-7.22 *
First-pass
fixation
duration
217.16
4.63
48.94 *
7.81
2.07
3.77 *
1.13
1.36
0.83
-0.05
0.01
-6.63 *
Regression
to wordn-1
0.02
0.16
0.13
0.15
0.09
1.68
-0.03
0.09
-0.38
-0.003
0.001
-2.88*
Longrange
regression
1.08
0.20
5.41 *
-0.64
0.10
-6.73 *
-0.20
0.10
-2.02 *
-0.003
0.001
-5.25 *
Rereading
time
845.62
81.51
10.37 *
-120.44
49.27
-2.44 *
-71.26
53.26
-1.34
-2.58
0.36
-7.17*
Re-reading
fixation
duration
243.32
6.04
40.28 *
6.62
4.37
1.52
15.39
5.86
2.63 *
-0.09
0.03
-3.18 *
Note: * Denotes statistical significance (t > 1.96). For Sentence reading time, Regression to wordn-1 and Re-reading time, see main
text for interactions with trial order. The following maximal models failed to converge: sentence reading time (display condition
removed from item random effects structure, correlation removed from subject random effects structure); re-reading fixation
duration (correlation removed from item random effects structure).
Spontaneous re-reading within sentences
Table 3. Experiment 1: Mean Reading Times for High and Low Frequency Critical Words. Standard Deviations Shown in
Parentheses.
Measure
Word frequency
Control
Beyond wordn
Beyond wordn-1
First fixation duration
High
210 (65)
217 (65)
213 (68)
Low
241 (81)
245 (76)
244 (80)
High
220 (79)
232 (83)
221 (79)
Low
264 (118)
264 (95)
264 (101)
High
280 (160)
271 (136)
285 (179)
Low
378 (255)
334 (200)
326 (178)
Gaze duration
Total time
57
Spontaneous re-reading within sentences
58
Table 4. Experiment 2: Means for Question Response Accuracy, Question Response Time, Sentence Reading Time, First-pass
and Re-reading Measures. Standard Deviations Shown in Parentheses.
Experiment 2
Control
Word length replacement
Word shape replacement
Temporary
Permanent
Temporary
Permanent
Question response accuracy
0.99 (0.11)
0.96 (0.19)
0.96 (0.19)
0.95 (0.21)
0.97 (0.18)
Question response time
2001 (901)
2112 (1021)
2122 (880)
2101 (908)
2121 (885)
Sentence reading time
3124 (980)
2977 (890)
2952 (829)
3174 (943)
3242 (1240)
Number of first-pass fixations
10.54 (2.26) 10.60 (2.30)
10.68 (2.36)
10.71 (2.35)
10.57 (2.25)
First-pass fixation duration
215 (32)
223 (42)
219 (35)
219 (32)
216 (32)
Regressiona to wordn-1
0.51 (0.50)
0.47 (0.50)
0.45 (0.50)
0.55 (0.50)
0.57 (0.50)
Long-range regressiona
0.62 (0.49)
0.40 (0.49)
0.42 (0.49)
0.58 (0.49)
0.60 (0.49)
Re-reading timeb
705 (608)
545 (491)
546 (446)
718 (535)
842 (934)
Re-reading fixation duration
232 (80)
235 (92)
231 (84)
240 (84)
248 (90)
Note. a Proportion of trials with at least one regression
b
Re-reading time for trials for which re-reading occurred
Spontaneous re-reading within sentences
59
Table 5. Experiment 2: Linear Mixed Effect Model Statistics (Baseline Analyses) for Question Response Accuracy, Question
Response Time, Sentence Reading Time, First-pass and Re-reading Measures.
Experiment 2
Question
Question Sentence Number of First-pass Regression LongReRe-reading
response
response reading
first-pass
fixation
to wordn-1
range
reading
fixation
accuracy
time
time
fixations
duration
regression time
duration
Intercept
b 5.19
2001.40 3134.14 10.56
215.26
0.06
0.61
680.66
230.36
se 0.45
98.07
141.26
0.38
4.31
0.18
0.23
57.44
6.73
t/z 11.43 *
20.41 *
22.19 *
27.88 *
49.90 *
0.34
2.70 *
11.85 *
34.23 *
Control vs.
b -1.18
121.02
-139.16
0.08
7.99
-0.16
-1.10
-160.44
2.62
Word length se 0.45
42.17
49.76
0.09
2.89
0.13
0.13
48.97
5.52
temporary
t/z -2.62 *
2.87 *
-2.80 *
0.86
2.77 *
-1.24
-8.18 *
-3.28 *
0.47
Control vs.
b -1.26
121.31
-174.93
0.14
3.50
-0.27
-1.00
-160.23
-0.33
Word length se 0.45
42.18
46.45
0.10
1.90
0.13
0.13
45.75
5.55
permanent
t/z -2.81 *
2.88 *
-3.77 *
1.35
1.85
-2.11 *
-7.51 *
-3.50 *
-0.06
Control vs.
b -1.47
103.96
46.25
0.15
3.81
0.21
-0.20
15.32
8.24
Word shape
se 0.44
42.23
41.58
0.09
1.82
0.13
0.13
38.93
5.36
temporary
t/z -3.32 *
2.46 *
1.11
1.73
2.09 *
1.69
-1.49
0.39
1.54
Control vs.
b -1.18
132.50
128.63
0.06
1.36
0.31
-0.08
136.76
15.80
Word shape
se 0.46
42.42
75.90
0.08
1.71
0.13
0.13
68.55
5.34
permanent
t/z -2.59 *
3.12 *
1.70
0.72
0.79
2.44 *
-0.61
2.00 *
2.96 *
Trial order
b -1.27
-2.71
-0.003
-0.05
-0.004
-1.48
se 0.33
0.30
0.001
0.01
0.001
0.30
t/z -3.89 *
-9.03 *
-5.84 *
-4.02 *
-4.12 *
-5.02 *
Note: * Denotes statistical significance (t > 1.96). The following maximal models failed to converge: question response time and
re-reading fixation duration (display condition removed from subject and item random effects structures); re-reading time (display
condition removed from item random effects structure).
Spontaneous re-reading within sentences
60
Table 6. Experiment 2: Linear Mixed Effect Model Statistics (2X2 Analyses) for Question Response Accuracy, Question
Response Time, Sentence Reading Time, First-pass and Re-reading Measures.
Experiment 2
Intercept
Text
replacement
type
Text
replacement
permanence
Type X
Permanence
Trial order
b
se
t/z
b
se
t/z
b
se
t/z
b
se
t/z
b
se
t/z
Question
response
accuracy
3.87
0.24
15.97 *
-0.10
0.22
-0.45
0.11
0.22
0.48
0.37
0.45
0.82
-
Question
response
time
2121.25
94.48
22.45 *
-3.65
36.16
-0.10
14.84
36.36
0.41
24.03
64.07
0.38
-0.83
0.36
-2.28 *
Sentence
reading
time
3098.95
141.06
21.97 *
244.73
40.66
6.02 *
23.15
37.76
0.61
118.65
88.49
1.34
-2.67
0.34
-7.90 *
Number of
first-pass
fixations
10.67
0.39
27.11 *
-0.003
0.05
-0.06
-0.01
0.07
-0.19
-0.15
0.12
-1.21
-0.003
0.001
-5.38 *
First-pass
fixation
duration
219.39
4.29
51.15 *
-3.20
1.50
-2.14 a
-3.52
1.54
-2.28 *
2.03
2.90
0.70
-0.06
0.01
-4.10 *
Regression
to wordn-1
0.08
0.16
0.49
0.47
0.09
5.29 *
-0.01
0.09
-0.08
0.21
0.18
1.16
-0.003
0.001
-3.02 *
Longrange
regression
0.01
0.20
0.06
0.89
0.09
9.57 *
0.10
0.09
1.12
0.02
0.18
0.12
-
Rereading
time
641.64
47.35
13.55 *
235.02
36.42
6.45 *
58.09
38.16
1.52
116.13
91.99
1.26
-1.36
0.32
-4.21 *
Re-reading
fixation
duration
237.19
6.23
38.08 *
10.95
4.75
2.31 *
2.39
4.06
0.59
9.99
7.96
1.26
-
Note: * Denotes statistical significance (t > 1.96). a Not significant for model based on log transformed data. The following
maximal models failed to converge: question response time, number of first-pass fixations and re-reading fixation duration
(correlation removed from item random effects structures).
Spontaneous re-reading within sentences
61
Table 7. Experiment 3: Example Experimental Items.
Sentence
Difficulty Question (Answer in parentheses)
The doctor who the receptionist phoned asked the
Easy
Did the receptionist send an email? (N)
nurse for the prescriptions.
Difficult
Was it the receptionist that asked for the prescriptions? (N)
The accountant who the chairman visited asked the
Easy
Did someone ask the statistician for help? (Y)
statistician for some help.
Difficult
Was it the accountant that asked the statistician for help? (Y)
The model who the woman photographed told the
Easy
Were the actors messing around? (Y)
actors to stop messing around.
Difficult
Was it the model who told the actors to behave? (Y)
The babysitter who the mother trusted scolded the
Easy
Were the children playing quietly? (N)
children for being too noisy.
Difficult
Was it the mother who scolded the children? (N)
Spontaneous re-reading within sentences
62
Table 8. Experiment 3: Means for Question Response Accuracy, Question Response Time, Sentence Reading Time, First-pass
and Re-reading Measures. Standard Deviations Shown in Parentheses.
Experiment 3
Question
Control
type
Question response accuracy
Word length replacement
Word shape replacement
Temporary
Permanent
Temporary
Permanent
Easy
0.93 (0.26)
0.95 (0.22)
0.94 (0.24)
0.96 (0.20)
0.95 (0.22)
Difficult
0.78 (0.41)
0.75 (0.43)
0.76 (0.43)
0.75 (0.43)
0.75 (0.44)
Easy
1930 (1088) 1874 (906)
1849 (877)
1798 (790)
1902 (821)
Difficult
2601 (1242) 2699 (1318)
2667 (1351)
2739 (1412)
2734 (1383)
Sentence reading time
-
3608 (1371) 3393 (1216)
3297 (1062)
3753 (1588)
3614 (1199)
Number of first-pass fixations
-
10.58 (2.02) 10.66 (2.05)
10.66 (2.12)
10.64 (2.06)
10.68 (2.04)
First-pass fixation duration
-
231 (38)
238 (42)
237 (41)
239 (42)
239 (42)
Regressiona to wordn-1
-
0.64 (0.48)
0.56 (0.50)
0.52 (0.50)
0.58 (0.49)
0.58 (0.49)
Long-range regressiona
-
0.60 (0.49)
0.37 (0.48)
0.40 (0.49)
0.57 (0.50)
0.57 (0.49)
Re-reading timeb
-
1100 (1008) 853 (834)
739 (619)
1181 (1183)
984 (799)
Re-reading fixation duration
-
230 (63)
241 (89)
237 (71)
251 (85)
Question response time
Note. a Proportion of trials with at least one regression
238 (78)
b
Re-reading time for trials for which re-reading occurred
Spontaneous re-reading within sentences
63
Table 9. Experiment 3: Linear Mixed Effect Model Statistics (Baseline Analyses) for Question Response Accuracy, Question
Response Time, Sentence Reading Time, First-pass and Re-reading Measures.
Experiment 3
Intercept
b
Control vs.
Word length
temporary
Control vs.
Word length
permanent
Control vs.
Word shape
temporary
Control vs.
Word shape
permanent
Trial order
se
t/z
b
se
t/z
b
se
t/z
b
se
t/z
b
se
t/z
b
se
t/z
Question response
accuracy
Easy
Difficult
Question response
time
Easy
Difficult
Sentence
reading
time
Number of First-pass
first-pass
fixation
fixations
duration
Regr. to
wordn-1
Longrange
regr.
Rereading
time
Re-reading
fixation
duration
3.07
0.28
11.12 *
0.46
0.31
1.49
0.17
0.30
0.57
0.61
0.33
1.87
0.38
0.31
1.25
-
1936.60
101.50
19.08 *
-62.32
74.35
-0.84
-84.37
73.41
-1.15
-128.95
73.22
-1.76
-28.45
69.50
-0.41
-1.33
0.18
-7.24 *
3626.81
171.95
21.09 *
-240.38
55.08
-4.36 *
-326.96
67.74
-4.83 *
139.89
75.99
1.84
0.06
73.05
0.001
-1.92
0.17
-11.52 *
10.59
0.26
40.22 *
0.06
0.10
0.61
0.05
0.10
0.56
0.06
0.08
0.78
0.10
0.10
1.02
-0.002
0.0002
-6.84 *
0.75
0.55
1047.31
229.44
0.22
3.49 *
-0.45
0.11
-3.92 *
-0.64
0.12
-5.52 *
-0.33
0.12
-2.82 *
-0.36
0.12
-3.10 *
-0.003
0.0004
-7.69 *
0.21
2.65 *
-1.18
0.12
-10.25 *
-1.03
0.12
-8.94 *
-0.20
0.11
-1.74
-0.16
0.11
-1.41
-0.003
-0.0004
-7.13 *
96.67
10.83 *
-276.68
54.19
-5.11 *
-368.32
65.68
-5.61 *
32.17
58.80
0.55
-103.93
61.60
-1.69
-0.99
0.16
-6.08 *
4.98
46.08 *
8.50
5.16
1.65
11.13
5.57
2.00 a
7.63
4.71
1.62
22.42
4.95
4.53 *
-
1.66
0.21
7.71 *
-0.19
0.19
-1.03
-0.13
0.19
-0.71
-0.21
0.19
-1.11
-0.24
0.19
-1.30
-
2611.03
115.54
22.60 *
87.66
83.88
1.05
54.06
84.27
0.64
135.18
88.22
1.53
128.65
87.44
1.47
-1.36
0.28
-4.81 *
231.14
5.17
44.71 *
6.78
1.50
4.52 *
6.14
1.51
4.06 *
7.43
1.51
4.92 *
7.84
1.51
5.20 *
-
Note: “Regr.” = Regression. * Denotes statistical significance (t > 1.96). The following maximal models failed to converge:
difficult question response time, sentence reading time and re-reading time (display condition removed from item random effects
structures); number of first-pass fixations (display condition removed from item random effects structure, correlation removed
from subject random effects structure), re-reading fixation duration (correlation removed from item random effects structure).
Spontaneous re-reading within sentences
64
Table 10. Experiment 3: Linear Mixed Effect Model Statistics (2X2 Analyses) for Question Response Accuracy, Question
Response Time, Sentence Reading Time, First-pass and Re-reading Measures.
Experiment 3
Intercept
Text
replacement
type
Text
replacement
permanence
Type X
Permanence
Trial order
b
se
t/z
b
se
t/z
b
se
t/z
b
se
t/z
b
se
t/z
Question response
accuracy
Easy
Difficult
Question response
time
Easy
Difficult
Sentence
reading
time
Number of First-pass
first-pass
fixation
fixations
duration
Regr. to
wordn-1
Longrange
regr.
Rereading
time
Re-reading
fixation
duration
3.40
0.24
14.30 *
0.18
0.24
0.78
-0.26
0.24
-1.12
0.06
0.47
0.13
-
1861.25
71.04
26.20 *
-6.03
41.49
-0.15
40.77
40.09
1.02
123.68
78.52
1.58
-1.32
0.19
-6.81 *
3520.94
153.18
22.99 *
354.57
53.76
6.60 *
-114.53
52.13
-2.20 a
-57.21
84.41
-0.68
-1.99
0.18
-10.98 *
10.66
0.27
39.35 *
0.03
0.06
0.41
0.02
0.06
0.30
0.04
0.11
0.37
-0.002
0.0003
-6.74 *
0.30
0.20
1.52
0.20
0.08
2.51 *
-0.11
0.08
-1.35
0.16
0.16
0.98
-0.003
0.0004
-7.11 *
-0.09
0.20
-0.45
0.93
0.08
11.36 *
0.09
0.08
1.17
-0.12
0.16
-0.72
-0.003
0.0004
-6.01 *
872.45
72.69
12.00 *
279.80
45.76
6.11 *
-115.00
50.39
-2.28 a
-44.81
71.27
-0.63
-0.99
0.18
-5.62 *
241.75
4.94
48.91 *
5.06
4.11
1.23
8.46
3.61
2.34 a
12.21
8.58
1.42
-
1.44
0.17
8.51 *
-0.06
0.13
-0.45
0.01
0.13
0.08
-0.10
0.26
-0.40
-
2712.89
128.78
21.07 *
61.97
59.48
1.04
-21.39
64.89
-0.33
24.73
117.31
0.21
-1.24
0.32
-3.89 *
238.22
5.17
46.09 *
1.11
1.46
0.76
-0.12
1.15
-0.10
1.01
2.44
0.41
-
Note: “Regr.” = Regression. * Denotes statistical significance (t > 1.96). a Not significant for model based on log transformed
data. For re-reading time, see main text for interactions with trial order. The following maximal models failed to converge: firstpass fixation duration (interaction removed from item random effects structure).
Spontaneous re-reading within sentences 65
Figure captions
Figure 1. Example experimental sentence and illustration of the trailing mask technique
employed in Experiment 1. In this example the word “room” is the high frequency critical
word. In the low frequency condition the critical word was “crib” and the sentence read
“He knew that the small crib would be ideal for his baby nephew.”. In the example, the
asterisk above each line represents the location of eye fixation, there is a regression from
“room” followed by re-reading from the word “He”. In the control condition (A) there is
no text replacement. In the “Beyond wordn” condition (B) all text is replaced by visually
similar letters to the left of the fixated word. In the “Beyond wordn-1” condition (C) text is
replaced by visually similar letters to the left of wordn-1. In both “Beyond wordn” and
“Beyond wordn-1” conditions the text replacements are permanent, text remains
unavailable during re-reading.
Figure 2. Experiment 1. Loess curves for Sentence reading time and Re-reading time, for
each of the three display conditions as a function of trial order.
Figure 3. Illustration of the gaze contingent change technique employed in Experiments 2
and 3. During first-pass, the word to the left of the fixated word (wordn-1) is always
displayed correctly. The asterisk above each line represents the location of eye fixation. In
all examples there is a regression from “teacher” followed by re-reading from the word
“solve”. In the control condition (A) there is no text replacement. In the temporary
replacement conditions (B, D) text is replaced beyond wordn-1 and re-appears during re-
Spontaneous re-reading within sentences 66
reading. In the permanent replacement conditions (C, E) text is replaced beyond wordn-1
and remains unavailable during re-reading. Text replacements preserved only word length
(B, C) or word length and shape (D, E).
Figure 4. Experiment 3. Loess curve for Re-reading time for word length and word shape
text replacement conditions.
Figure 5. Experiment 3. Boxplots for temporary (“Temp”) and Permanent (“Perm”)
replacement conditions for sentence reading time, re-reading time and re-reading fixation
duration.
Spontaneous re-reading within sentences 67
Figure 1.
(A) Control:
*
He knew that the small room would be really useful for storage.
*
He knew that the small room would be really useful for storage.
*
He knew that the small room would be really useful for storage.
(B) Beyond wordn:
*
Lo lrav fbol fba cnohb room would be really useful for storage.
*
Lo lrav fbol fba cnohb room would be really useful for storage.
*
Lo lrav fbol fba cnohb room would be really useful for storage.
(C) Beyond wordn-1:
*
Lo lrav fbol fba small room would be really useful for storage.
*
Lo lrav fbol fba small room would be really useful for storage.
*
Lo lrav fbol fba small room would be really useful for storage.
Spontaneous re-reading within sentences
Figure 2.
68
Spontaneous re-reading within sentences 69
Figure 3.
(A) Control:
*
The boy could not solve the tricky anagram alone so he asked the teacher for help.
*
The boy could not solve the tricky anagram alone so he asked the teacher for help.
*
The boy could not solve the tricky anagram alone so he asked the teacher for help.
(B) Word length temporary replacement:
*
XXX XXX XXXXX XXX XXXXX XXX XXXXXX XXXXXXX XXXXX XX XX XXXXX the teacher for help.
*
XXX XXX XXXXX not solve the tricky anagram alone so he asked the teacher for help.
*
XXX XXX XXXXX XXX XXXXX the tricky anagram alone so he asked the teacher for help.
(C) Word length permanent replacement:
*
XXX XXX XXXXX XXX XXXXX XXX XXXXXX XXXXXXX XXXXX XX XX XXXXX the teacher for help.
*
XXX XXX XXXXX XXX XXXXX XXX XXXXXX XXXXXXX XXXXX XX XX XXXXX the teacher for help.
*
XXX XXX XXXXX XXX XXXXX XXX XXXXXX XXXXXXX XXXXX XX XX XXXXX the teacher for help.
(D) Word shape temporary replacement:
*
Iba haq eavib ral caiuc fbc fnlehq oroqnon oiarc ca lc ochcl the teacher for help.
*
Iba haq eavib not solve the tricky anagram alone so he asked the teacher for help.
*
Iba haq eavib ral caiuc the tricky anagram alone so he asked the teacher for help.
(E) Word shape permanent replacement:
*
Iba haq eavib ral caiuc fbc fnlehq oroqnon oiarc ca lc ochcl the teacher for help.
*
Iba haq eavib ral caiuc fbc fnlehq oroqnon oiarc ca lc ochcl the teacher for help.
*
Iba haq eavib ral caiuc fbc fnlehq oroqnon oiarc ca lc ochcl the teacher for help.
Spontaneous re-reading within sentences 70
Figure 4.
Spontaneous re-reading within sentences 71
Figure 5.