Abstract
Readers tend to skip words, particularly when they are short, frequent, or predictable. Angele and Rayner (2013) recently reported that readers are often unable to detect syntactic anomalies in parafoveal vision. In the present study, we manipulated target word predictability to assess whether contextual constraint modulates the-skipping behavior. The results provide further evidence that readers frequently skip the article the when infelicitous in context. Readers skipped predictable words more often than unpredictable words, even when the, which was syntactically illegal and unpredictable from the prior context, was presented as a parafoveal preview. The results of the experiment were simulated using E-Z Reader 10 by assuming that cloze probability can be dissociated from parafoveal visual input. It appears that when a short word is predictable in context, a decision to skip it can be made even if the information available parafoveally conflicts both visually and syntactically with those predictions.
One striking observation about skilled readers’ eye movements is that they do not fixate every word in a sentence. Instead, quite a few words are skipped. This tends to occur mainly for short words, but longer words are skipped occasionally. In order to read efficiently, skilled readers must decide very quickly – within the first 150 ms of a fixation (Rayner, 1998, 2009) – whether to skip or to fixate the next word and then initiate the appropriate saccade program. During this time, readers have two sources of information to base their decision on: first, they can use parafoveal information that is available about the next word; second, they can use information about the sentence context they have previously read. However, it is not clear whether readers actually use both of these sources of information in making their skipping decisions.
It does seem clear that parafoveal information is important in deciding where to look next. That is, readers of English use parafoveal input in order to identify word boundaries and target words accurately (Pollatsek & Rayner, 1982; Rayner, Fischer, & Pollatsek, 1998). While word length has a strong influence on fixation probability (Rayner & McConkie, 1976; Rayner, Sereno, & Raney, 1996), other properties of the parafoveal word also affect whether it is skipped or fixated: articles and other closed-class words are skipped more often than open-class words of the same length such as three-letter verbs (Angele & Rayner, 2013; Drieghe, Pollatsek, Staub, & Rayner, 2008; Gautier, O’Regan, & Le Gargasson, 2000; O’Regan, 1979, 1980), and high frequency words are skipped more often than low frequency words (Gollan et al., 2011; Rayner et al., 1996). However, there is also much evidence that prior sentence context in the form of predictability of a target word influences skipping. If a word is highly predictable from the sentence context it is skipped more often than a less predictable word (Balota, Pollatsek, & Rayner, 1985; Drieghe, Rayner, & Pollatsek, 2005; Ehrlich & Rayner, 1981; Fitzsimmons & Drieghe, 2013; Rayner, Slattery, Drieghe, & Liversedge, 2011; Rayner & Well, 1996).
Another form of sentence context, namely, syntactic context information, has been examined in studies of word skipping. Specifically, Angele and Rayner (2013) tested whether parafoveal information or sentence syntactic context information had the stronger effect on word skipping by pitting the two influences against each other. They did this by using the gaze-contingent display change paradigm (Rayner, 1975) to manipulate the parafoveal information that readers received about a three-letter verb embedded in a sentence. For example, while fixating the word always in the sentence “They always dim the lights at night,” the parafoveal preview information available to readers about the word dim could be (1) the word dim itself as a control condition, (2) a random letter preview (fda), or (3) the article the; following “They always,” the article the is syntactically illegal or infelicitous. Once the reader’s saccade crossed an invisible boundary location, the preview changed to the target word (i.e., dim). If readers have access to syntactic context information while making their skipping decision, they should not attempt to skip a word that is clearly infelicitous in the sentence context. On the other hand, if readers only rely on the parafoveal information to guide their skipping decision without taking the sentence syntactic context into account, they should be quite likely to skip the article the since it is an extremely high frequency function word. Angele and Rayner (2013) found that the latter was the case: readers were virtually as likely to skip the infelicitous the previews as they were to skip the when it occurred in a felicitous position. Only after moving on to subsequent words did readers experience disruption when they had skipped the infelicitous the previews, evidenced by higher go-past times and regression rates compared to the control condition. Angele, Laishley, Rayner, and Liversedge (2014) showed that this effect was not limited to the article the, but also applied to short high-frequency open-class words. For example, readers were more likely to skip the parafoveal preview of a high frequency word (dog) than the preview of a low-frequency word (dim), even when the sentence syntactic context only allowed the low-frequency word (“The increasingly dim/dog light made it hard to see”).
Taken together, the studies by Angele and Rayner (2013) and Angele et al. (2014) indicate that sentence syntactic context has little effect on word skipping. However, this conclusion is in direct contradiction to the many experiments described above that found clear effects of word predictability on word skipping. One possible explanation for this apparent contradiction is that sentence context may only have an effect on fixation probability when it is highly constraining, that is, when the word to be skipped is highly predictable from the sentence context. In the studies by Angele and Rayner (2013) and Angele et al. (2014), the sentence context largely did not constrain the target word, which might explain the absence of sentence context effects. The present study tested this hypothesis as we manipulated both the preview that readers received of a target word (identical vs. infelicitous the) and the degree of constraint of the sentence context it was embedded in. For example, the target word cut was either presented in a highly predictable context (“Jane used the scissors to carefully cut scraps of paper.”) or in an unconstrained context (“Jane used the machine to carefully cut scraps of paper.”).
If sentence context effects are limited to cases in which the target word is highly constrained, we expected to find an interaction between constraint and preview: readers should be less likely to skip the infelicitous the previews in the high constraint condition than in the low constraint condition. If constraint has no effect on the influence of sentence context, on the other hand, skipping rates for the infelicitous the previews should not differ between the high and the low constraint conditions. The experiment is also simulated using E-Z Reader 10 (Reichle, Warren, & McConnell, 2009) to assess these different possibilities.
Method
Subjects
Forty-four UC San Diego students participated in the experiment for course credit. All were native English speakers with normal or corrected to normal vision, and were naïve to the purpose of the study.
Apparatus
Subjects’ eye movements were recorded with a sampling rate of 1000 Hz using an SR Research Eyelink 1000 eyetracker. Sentences were displayed in 14pt Courier New font on a Hewlett Packard p1230 CRT monitor with a refresh rate of 150 Hz. Viewing distance was approximately 60 cm and each character subtended about .3 degrees of visual angle. Only the right eye was recorded, although viewing was binocular.
Materials
Forty experimental sentence frames comprised four conditions (2 constraint x 2 preview). Each sentence included a 3-character target word that was used as a verb1 (e.g., “Jane used the scissors to carefully cut scraps of paper”, target word in italics; see the Appendix for the complete list of sentences). Pre-target words were chosen to be of sufficient length to avoid skipping (mean length 6.3 characters). We included two preview conditions: a correct preview of the target word (cut) and a the-preview containing an infelicitous preview of an article (see Figure 1). The the-preview always appeared in a position in the sentence in which it was syntactically inappropriate. On average, the frequency for the target words was 18 counts per million in the Francis and Kučera (1982) corpus2. For comparison, the frequency of the in this corpus is 69,971 counts per million.
Figure 1.
Example stimuli. The display changed to the correct target word when readers moved their eyes to the right of the invisible boundary (dashed line).
We manipulated the constraint of the target word such that it was predictable or unpredictable depending on the context of the sentence while holding the position of the target word as constant as possible (see Figure 1; on average the target word was word 9.4 in both the high and low constraint sentences). To determine the predictability of the target words, twenty workers recruited through Amazon Mechanical Turk participated in a cloze norming task in exchange for payment. They were asked to report what they thought the next word in the sentence should be after reading the sentences up until to the target word (e.g., “Jane used the scissors to carefully ___”). Subjects entered the target word 76.8% of the time in the high constraint condition and 5.0% of the time in the low constraint condition. They never entered the as their response.
Procedure
The forty experimental sentences were embedded among sixty filler sentences. Each sentence was presented on the computer screen individually. Following a three-point calibration procedure, subjects were asked to read the sentences silently and to press a button when they finished reading. The gaze-contingent boundary paradigm (Rayner, 1975) was used to present either a correct preview of the target word (cut) or a preview of the that changed to the target word after readers’ eyes crossed the invisible boundary (see Figure 1 for an example). The display change completed within 4ms on average (range 0–7ms) once the tracker detected a saccade crossing the boundary. After 33 of the 100 sentences, subjects responded to a two-alternative forced-choice comprehension question.
Results
Skipping and regression probabilities as well as fixation times were analyzed for three regions: the pre-target word, the target word, and the post-target word. 4.3% of trials were eliminated due to blinks or track loss on the target word or on adjacent fixations, as were 10.2% of trials with display changes that completed after fixation onset. Fixations shorter than 80 ms and within one character of an adjacent fixation were combined (0.3%). 3.7% of fixations shorter than 80 ms or longer than 1000 ms were eliminated. Comprehension accuracy was high (97% correct).
We report linear mixed models (LMMs) with subjects and items as crossed random effects (Baayen, Davidson, & Bates, 2008). These analyses were chosen instead of ANOVAs due to the uneven cell sizes that result from word skipping, where skipped words are treated as missing data in duration measures. LMMs were fit using the lmer function from the lme4 package (Version 1.1–7; Bates, Maechler, Bolker, & Walker, 2014) in the R Environment for Statistical Computing (Version 3.1.0; R Core Team, 2014). Generalized LMMs were fit (using a logit link) to two binomial dependent measures, skipping probability and probability of regressions out, and we report regression coefficients (b), standard errors, and z-values for fixed effects and their interactions (t-values are reported for duration measures). LMMs were fit to the following log-transformed fixation duration measures: single fixation duration (SFD; the mean duration of fixations on a word when that word received just one fixation in the first pass), gaze duration (GD; the sum of all fixations on a word before leaving it, including refixations), and go-past time (the sum of all fixations on a word before leaving it to the right, including regressions to previous words). Two fixed factors were included and contrasts were sum-coded (meaning that the intercept of the model is the grand mean of the dependent measure) for preview (the = 1, target = −1) and for constraint (high = 1, low = −1). An additional sum-coded factor indicating whether or not the target word was skipped (target fixated = 1, target skipped = −1) was also included in a number of the models to investigate how reading behavior is affected just prior to or following a skipping decision.
We determined the random effects structure of our models by starting with the maximal random effects structure (Barr, Levy, Scheepers, & Tily, 2013) but as the maximal models did not converge, we had to reduce the random effects structure. Reported models include random intercepts for subjects and items, random slopes for preview and predictability by subjects (as additive main effects with no interaction; an additional slope for target skipping was included as well for those models where it was included as a fixed effect), as well as random slopes for preview and constraint by items (again with no interaction)3. Because it is not clear how to compute the degrees of freedom for LMMs, we do not report p-values, and use the two-tailed criterion |z| ≥1.96 to correspond to a significance test at the 0.05 α-level (t-values are interpreted in the same way). LMM results are presented in Tables 2, 4, and 6.
Table 2.
Linear mixed effects model analyses on the pre-target word.
| p(Skip) | SFD | GD | |||||||
|---|---|---|---|---|---|---|---|---|---|
|
|
|||||||||
| Effect | b | SE | z | b | SE | t | b | SE | t |
|
|
|||||||||
| (Intercept) | −2.39 | 0.24 | −9.94 | 5.32 | 0.02 | 265.24 | 5.40 | 0.03 | 207.93 |
| Fixed Effects | |||||||||
| Skip target | -- | -- | -- | 0.02 | 0.01 | 2.16 | 0.00 | 0.01 | 0.05 |
| Preview | −0.02 | 0.08 | −0.22 | 0.01 | 0.01 | 0.77 | 0.00 | 0.01 | −0.03 |
| Constraint | −0.01 | 0.08 | −0.13 | −0.02 | 0.01 | −1.68 | −0.01 | 0.01 | −0.84 |
| Skip target * preview | -- | -- | -- | 0.01 | 0.01 | 1.01 | −0.01 | 0.01 | −0.70 |
| Skip target * constraint | -- | -- | -- | 0.00 | 0.01 | 0.05 | −0.01 | 0.01 | −0.68 |
| Preview * constraint | −0.05 | 0.08 | −0.62 | −0.01 | 0.01 | −0.88 | 0.02 | 0.01 | −1.76 |
| Skip target * preview * constraint | -- | -- | -- | −0.01 | 0.01 | −1.44 | 0.02 | 0.01 | −1.62 |
Note. Each column represents a model fit to one of the dependent variables. SFD = single fixation duration; GD = gaze duration; SE = standard error; t = test statistic (b/SE). Cells marked in bold represent |t| ≥ 1.96.
Table 4.
Linear mixed effects model analyses for the target word.
| Duration Measures | Binomial Measures | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
|
|||||||||||||||
| SFD | GD | Go-Past | p(Skip) | Regressions In | |||||||||||
|
|
|||||||||||||||
| Effect | b | SE | t | b | SE | t | b | SE | t | b | SE | z | b | SE | z |
| (Intercept) | 5.40 | 0.03 | 184.04 | 5.43 | 0.03 | 192.84 | 5.51 | 0.03 | 175.48 | 0.30 | 0.14 | 2.15 | −0.53 | 0.15 | −3.45 |
| Fixed Effects | |||||||||||||||
| Skip target | -- | -- | -- | -- | -- | -- | -- | -- | -- | −2.28 | −0.20 | −11.45 | |||
| Preview | −0.03 | 0.02 | −2.06 | 0.04 | 0.02 | 2.31 | 0.04 | 0.02 | 2.53 | 0.62 | 0.04 | 8.34 | 0.28 | 0.17 | 1.69 |
| Constraint | 0.01 | 0.04 | 0.56 | 0.00 | 0.02 | 0.07 | −0.00 | 0.02 | −0.20 | 0.16 | 0.06 | 2.71 | −0.09 | 0.16 | −0.55 |
| Skip target * preview | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | −0.20 | 0.14 | −1.368 |
| Skip target * constraint | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | −0.21 | 0.15 | −1.40 |
| Preview * constraint | −0.01 | 0.01 | −0.67 | 0.00 | 0.01 | 0.34 | −0.01 | 0.02 | −0.55 | 0.03 | 0.06 | 0.50 | −0.07 | 0.13 | −0.58 |
| Skip target * preview * constraint | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | -- | 0.09 | 0.15 | 0.58 |
Note. Each column represents a model fit to one of the dependent variables. SFD = single fixation duration; GD = gaze duration; SE = standard error; t = test statistic (b/SE). Cells marked in bold represent |t| ≥ 1.96.
Table 6.
Linear mixed effects model analyses for the post-target word.
| Duration Measures | Binomial Measures | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
|
|||||||||||||||
| SFD | GD | Go-Past Time | p(Skip) | Regressions Out | |||||||||||
|
|
|||||||||||||||
| Effect | b | SE | t | b | SE | t | b | SE | t | b | SE | z | b | SE | z |
| (Intercept) | 5.33 | 0.02 | 349.31 | 5.24 | 0.02 | 277.58 | 5.59 | 0.03 | 205.49 | −1.13 | 0.27 | −4.17 | −2.37 | 0.31 | −7.77 |
| Fixed Effects | |||||||||||||||
| Skip target | −0.06 | 0.01 | −4.93 | −0.08 | 0.01 | −6.28 | −0.17 | 0.02 | −11.17 | 1.16 | 0.09 | 12.75 | −1.20 | 0.28 | −4.27 |
| Preview | 0.01 | 0.01 | 0.65 | 0.02 | 0.01 | 1.13 | 0.09 | 0.02 | 5.75 | −0.07 | 0.10 | −0.71 | 0.79 | 0.21 | 3.76 |
| Constraint | 0.00 | 0.01 | 0.33 | −0.01 | 0.01 | −0.78 | −0.03 | 0.02 | −1.77 | 0.10 | 0.08 | 1.29 | −0.38 | 0.19 | −2.00 |
| Skip target * preview | −0.03 | 0.01 | −2.66 | −0.03 | 0.01 | −2.36 | −0.07 | 0.02 | −4.78 | −0.17 | 0.09 | −1.95 | −0.06 | 0.20 | −0.30 |
| Skip target * constraint | −0.01 | 0.01 | −0.78 | −0.01 | 0.01 | −0.41 | −0.01 | 0.02 | −0.96 | −0.13 | 0.08 | −1.61 | −0.33 | 0.19 | −1.75 |
| Preview * constraint | 0.00 | 0.01 | 0.31 | 0.02 | 0.01 | 1.39 | 0.01 | 0.02 | 0.73 | −0.08 | 0.08 | −1.04 | 0.16 | 0.18 | 0.91 |
| Skip target * preview * constraint | 0.00 | 0.01 | 0.39 | 0.00 | 0.01 | 0.02 | 0.00 | 0.02 | 0.15 | −0.02 | 0.08 | −0.20 | 0.09 | 0.17 | 0.53 |
Pre-target word
Condition means for the pre-target word are presented in Table 1. Readers fixated on the pre-target word 86% of the time, and fixation probabilities were not affected by either the nature of the parafoveal preview or the predictability of the target word (all zs < .65). There was a small but reliable effect such that single fixation durations on the pre-target word (84% of first-pass fixations) prior to a skip were shorter than prior to making a fixation on the target (210ms vs. 217ms; b = 0.02, SE = 0.01, t = 2.16), replicating a finding reported by Kliegl and Engbert (2005). This effect did not interact with either preview or constraint. Gaze duration showed a similar pattern with shorter fixations prior to skipping (234ms vs. 239ms), but this effect did not reach significance (t = 0.05).
Table 1.
Condition means for the pre-target word.
|
|
||||||||
|---|---|---|---|---|---|---|---|---|
| Skip | SFD | GD | SFD | GD | ||||
|
| ||||||||
| Constraint | Preview | All cases | Target subsequently fixated | Target subsequently skipped | Target subsequently fixated | Target subsequently skipped | ||
| High | target | 0.14 (0.03) | 209 (5.3) | 236 (6.7) | 215 (6.0) | 203 (7.0) | 239 (8.6) | 237 (11.2) |
| High | the | 0.13 (0.02) | 208 (5.5) | 232 (7.4) | 217 (11.2) | 206 (5.9) | 229 (12.3) | 237 (9.3) |
| Low | target | 0.13 (0.02) | 211 (5.1) | 232 (7.2) | 212 (6.7) | 209 (7.4) | 230 (8.0) | 244 (18.7) |
| Low | the | 0.15 (0.02) | 215 (5.1) | 239 (8.1) | 235 (14.2) | 208 (5.9) | 249 (14.5) | 237 (8.9) |
Note. Standard errors are in parentheses. SFD = single fixation duration, GD = gaze duration.
Target word skipping
Condition means for the target word skipping rates are presented in Table 3 and are presented graphically in Figure 2. Readers skipped over the target word on 56.9% of trials. This behavior was influenced both by the nature of the parafoveal preview as well as the constraint for the target word. Readers skipped previews of the more often than correct previews of the target word (0.69 vs. 0.44; b = 0.62, SE = 0.07, z = 8.34). Readers also skipped the word in the target position (either the correct target word or the) more often when the target word was highly constrained from the prior context than when it was not (0.60 vs. 0.53; b = 0.16, SE = 0.06, z = 2.71). These two factors did not significantly interact (z = 0.50). Interestingly, the lack of an interaction between constraint and preview suggests that constraint affected skipping equally for valid and invalid previews.
Table 3.
Condition means for the target word.
| Skip | SFD | GD | Go-Past | Regressions In | |||
|---|---|---|---|---|---|---|---|
|
| |||||||
| Constraint | Preview | All cases | Target initially skipped | Target initially fixated | |||
| High | target | 0.41 (0.04) | 221 (6.3) | 226 (6.9) | 261 (13.7) | 0.74 (0.10) | 0.06 (0.02) |
| High | the | 0.72 (0.03) | 249 (11.1) | 256 (11.2) | 289 (18.3) | 0.83 (0.05) | 0.10 (0.04) |
| Low | target | 0.41 (0.04) | 224 (6.8) | 230 (7.4) | 253 (10.6) | 0.66 (0.09) | 0.09 (0.02) |
| Low | the | 0.66 (0.03) | 242 (11.0) | 254 (10.9) | 293 (14.6) | 0.88 (0.03) | 0.11 (0.03) |
Note. SFD = single fixation duration; GD = gaze duration; SE = standard error.
Figure 2. Target word skipping.
Note: Error bars represent standard error of the mean.
Target word fixations
Fixation time measures on the target word represent the remaining subset of the data when the target word was fixated. Here, readers’ single fixation durations (41% of first-pass fixations) were shorter following correct target word previews than the previews (228ms vs. 248ms; b = −0.03, SE = 0.02, t = −2.06). The same pattern was present in gaze duration, with shorter fixations following valid previews (234ms vs. 257ms; b = −0.04, SE = 0.02, t = −2.31). This reflects the benefit of having valid parafoveal information available prior to fixation (Rayner, 1975). There was no effect of constraint on gaze durations (t = 0.07), and constraint did not interact with preview in this measure (t = −0.34). Go-past times were also shorter following correct vs. the previews (262ms vs. 288ms; b = −0.04, SE = 0.02, t = −2.53). In go-past time there was no effect of constraint (t = −0.20), and no interaction between constraint and preview (t = 0.55).
Readers made considerably more regressions back to the target word when it was initially skipped than when it was fixated (0.85 vs. 0.07; b = −2.28; SE = 0.20; z = −11.45). The effect of parafoveal preview on regression rates was only marginal, with more regressions to the target following the previews than correct target previews (0.50 vs. 0.17; b = 0.28, SE = 0.16, z = 1.69). There was no effect of constraint on regression rates (z = −0.55), and there were no significant interactions amongst any of the fixed factors (all zs < 1.41).
Post-target word
Condition means for the post-target word are presented in Table 5. It is clear that skipping over the target word (versus fixating it) disrupted processing on the post-target word, with longer fixations following a skip in single fixation duration (61% of first-pass fixations; 232ms vs. 207ms), gaze duration (274ms vs. 229ms), and go-past time (381ms vs. 258ms). Critically, this effect interacted with parafoveal preview in all three measures, with a larger cost in terms of elongated fixation times after skipping the previews compared to valid target word previews. See Figure 3 for a depiction of the nature of this interaction in go-past time. In addition to affecting fixation times on the post-target word, skipping affected regression rates out of this region; readers made more regressions out of the post-target word when the target word had been skipped versus fixated (0.33 vs. 0.07; b = −1.20, SE = 0.28, z = −4.27). Regressions were also more frequent following the previews than target previews (0.36 vs. 0.11; b = 0.79, SE = 0.21, z = 3.76). Finally, readers regressed out of this region more often in sentences containing low compared to high constraint target words (0.25 vs. 0.23; b = −0.38, SE = 0.19, z = −2.00).
Table 5.
Condition means on the post-target word.
| Target Skipped
|
||||||
|---|---|---|---|---|---|---|
| Constraint | Preview | Skip | SFD | GD | Go-Past | Regressions Out |
| High | target | 0.16 (0.04) | 225 (7.67) | 253 (10.99) | 300 (17.68) | 0.12 (0.03) |
| High | the | 0.22 (0.04) | 243 (11.40) | 290 (12.24) | 428 (21.62) | 0.44 (0.04) |
| Low | target | 0.11 (0.03) | 221 (9.55) | 287 (19.15) | 327 (20.88) | 0.15 (0.04) |
| Low | the | 0.18 (0.03) | 238 (10.08) | 285 (11.51) | 437 (20.52) | 0.44 (0.05) |
| Target Fixated
|
||||||
| High | target | 0.57 (0.05) | 205 (8.18) | 217 (8.85) | 236 (14.90) | 0.01 (0.01) |
| High | the | 0.44 (0.06) | 212 (12.87) | 230 (13.53) | 273 (30.30) | 0.07 (0.04) |
| Low | target | 0.54 (0.05) | 219 (9.28) | 240 (9.80) | 267 (15.58) | 0.08 (0.04) |
| Low | the | 0.51 (0.05) | 198 (10.87) | 214 (13.65) | 254 (19.95) | 0.12 (0.05) |
Figure 3. Go-past time on the post-target word.
Note. Error bars represent standard error of the mean.
E-Z Reader 10 Simulations
The primary result of our experiment is striking and unexpected: readers skip illicit occurrences of the word the more often when a different word is highly constrained by the sentence compared to when one is not. Why should constraint affect the processing of an altogether different word? This is unexpected given the wealth of evidence that readers are less likely to skip low-constraint words in sentences designed to elicit a higher-constraint alternative (e.g., Balota et al., 1985; Drieghe et al., 2005; Fitzsimmons & Drieghe, 2013; Rayner & Well, 1996) and are just about as likely to skip visually similar non-words as other low-constraint or even semantically anomalous words (Drieghe et al., 2005). Clearly words are processed to an advanced degree parafoveally prior to being skipped, perhaps even to the point of full recognition (Gordon, Plummer, & Choi, 2013). Furthermore, when a word is predictable in context, this pattern indicates a procedure wherein the system checks bottom-up parafoveal input against linguistic expectations before skipping. Given the extant data, our results demand explanation.
In order to uncover the mechanisms underlying our effects, we chose to simulate our experiment using the E-Z Reader model of eye movement control during reading (Reichle, Pollatsek, Fisher, & Rayner, 1998; Reichle et al., 2009). In E-Z Reader a word can be skipped if the fixated (foveal) word has been identified and the parafoveal word is nearly identified before the system is ready to move the eyes to the next word. This requires that the initial stage of lexical processing (L1, the “familiarity check”) complete from the parafovea. The time required to complete L1 for a parafoveal word is influenced by the difficulty of that word in the following ways: (1) a word is “guessed” such that the duration of L1 is set to 0ms with probability equal to its cloze predictability; (2) the duration of L1 is a function of word frequency, cloze probability, and mean eccentricity from the point of fixation. In terms of E-Z Reader, the effect of constraint on skipping any words in the observed data may be accounted for by the first mechanism, and skipping parafoveal words that look like the by the second (due to its very high frequency).
The question is how to incorporate cloze probability to model the joint effects of constraint and parafoveal preview on skipping. Our simulations were conducted to test two possibilities with regard to this issue: (1) the-skipping is driven primarily by its frequency of occurrence (i.e., there is no systematic effect of target word constraint on the-skipping); (2) the-skipping is affected by the cloze probability of the target word (e.g. cut) constrained by the preceding sentence context. If E-Z Reader predicts that skipping of apparent the-previews is affected by the degree of constraint for the target word (e.g. cut), with increased the-skipping in sentences with high versus low constraint, it would suggest that the contribution of constraint to skipping is indeed independent from that of the ease of identifying the parafoveal word.
We conducted the following simulations using the E-Z Reader 10 model4 (Reichle et al. 2009). Our approach was to use the model to estimate the rate at which readers would skip the preview word given its length, frequency of occurrence, and cloze probability. Frequency values for all words used in the experiment were extracted from the Francis and Kučera (1982) corpus5. For convenience we assumed a cloze probability of 0 for all words other than those in the target position (see Staub, 2011). For each simulation we used the default model parameters (as provided by Reichle, Pollatsek, & Rayner, 2012) and predicted data for 10,000 statistical subjects reading the sentences in all four conditions.
Simulation 1
In this simulation we asked whether the very high frequency of the word the alone (69,971 counts per million as compared to an average of 18 counts per million for the target words in our study) accounts for the rate at which it was skipped in our experiment. To do so, cloze probability for the in sentences with illicit previews was assumed to be 0, reflecting the fact that no subject entered the in the cloze norming study (for sentences with valid previews the cloze probabilities equaled those collected in the norming study). Mean simulated skipping rates are presented in Figure 4. For the sentences with correct target word previews, skipping rates were greater for sentences with high compared to low-constraint words (0.42 vs. 0.22). The model estimated that skipping the, however, was unaffected by constraint (0.23 for both high and low constraint conditions), and would actually be skipped less often than the correct target in constraining sentences (0.23 vs. 0.42).
Figure 4.
Simulation 1: Target skipping probabilities by condition.
Simulation 2
In the second simulation, we asked whether the frequency of the parafoveal word and the cloze probability for the expected word have additive effects on skipping. For both preview conditions we assumed that the cloze probability for the preview word was that of the correct target word (e.g., for a target die with cloze 0.8, we assigned 0.8 cloze for the in the false preview condition). The assumption here is that reader expectations are entirely separate from the input received from the parafoveal word. The mean obtained skipping rates for the four different conditions are presented in Figure 5. The pattern here is quite clear: the model simulation replicated the additive effects on skipping in the observed data, with greater skipping for constrained words (0.44 for high vs. 0.23 for low constraint words) and slightly more skipping for the compared to valid target previews (0.35 vs. 0.33). Importantly, E-Z Reader does not predict an interaction between preview and constraint, similar to the observed data. The results from this simulation are plotted alongside the observed skipping data in Figure 6. While the model appears to overestimate the role of constraint and underestimate the role of frequency in skipping these words, the basic pattern was successfully reproduced.
Figure 5.
Simulation 2: Target skipping probabilities by condition.
Figure 6.
Skipping probabilities for the target words from the observed data (left panel) and from Simulation 2 (right panel)
These simulations using E-Z Reader 10 demonstrate that our pattern of skipping data can be explained if we assume that, prior to skipping the, readers took into account the predictability of the word that they expected to see in that position. Importantly, the model predicted this pattern without making any assumptions about the syntactic fit between the preview word and the preceding material. That E-Z Reader predicts this pattern makes sense when we consider that the model permits a word to be “guessed” based solely on its predictability. Note though that in highly constraining sentences the model still predicts that the will be skipped slightly more often than the correct target words (0.45 versus 0.43). The second route of the L1 computation, which assumes that when a word is not “guessed” its identification time is a function of its frequency and cloze probability, accounts this for this difference.
General Discussion
We manipulated target word constraint and parafoveal preview using the gaze-contingent boundary paradigm. The preview of the target word was either a correct preview of a 3-character verb or an invalid preview of the, presented in sentences in which the correct target word was either predictable or unpredictable. Our goal was to determine whether contextual constraint modulates skipping infelicitous occurrences of the, following prior research that has shown that syntactic constraints do not modulate skipping decisions (Angele et al., 2014; Angele & Rayner, 2013). If readers primarily make use of the sentence context when it highly constrains the target word, we expected an interaction between constraint and preview, such that readers would be more likely to catch the violation and decide not to skip the false the previews in high versus low constraint sentences.
In our study, readers were more likely to skip parafoveal previews of the than valid previews of the target words (replicating Angele & Rayner, 2013). This is further evidence that readers are more likely to skip high frequency words, regardless of whether the word fits in the context of the sentence (consistent with Angele et al., 2014). Readers also skipped the word in the target position (either the target word or the) more often in sentences where the target word was highly constrained compared to when it was unconstrained. Thus, even when the sentence context sets a high expectation for a specific word, readers do not seem to integrate parafoveal content with higher-level linguistic information and, as a consequence, are still more likely to skip a parafoveal word that looks like the than one that looks like the expected word. This suggests that the parafoveal preview is processed to an advanced degree (to the point that its frequency, or, at the very least, its orthographic familiarity affects reading behavior) but not to the extent that its fit within the context is assessed. However, it is evident that the false the previews disrupted later processing, especially when they were skipped: while readers spent longer fixating the post-target region when it was skipped compared to when it was fixated, this difference was even greater when the preview word was the.
Given the novelty of the effect that our experimental manipulation had on skipping behavior, we sought to model this pattern of data using the E-Z Reader model. We conducted two simulations of our experiment using assumptions about the cloze probabilities for the word in the target position: (1) the cloze probability was set to 0 in the the-preview condition (corresponding to the actual cloze probability of the in that location; in the target preview condition, the cloze probability was the actual cloze probability of the target word); (2) the cloze probability was always set to the cloze probability for the target word in the sentence, regardless of the preview condition (e.g. when the was the preview for cut, the cloze probability for cut was used). Simulation 2, but not Simulation 1, successfully replicated our pattern of skipping data. Thus, it appears that serial attention shift models of eye movement control are well equipped to handle our results if we assume that reader expectations are computed separately from the visual information they have in front of them. Indeed, this assumption is built-in to the model as a component of the function that handles the initial stage of lexical processing. While it is possible that parallel processing models such as SWIFT (Engbert, Nuthmann, Richter, & Kliegl, 2005) would also predict this pattern, quantitative simulations like those reported here would need to be conducted. Since fixation probabilities in SWIFT are modulated by the amount of parafoveal processing that a word has received (which is influenced both by a word’s frequency and cloze probability), it seems possible that it could replicate this pattern as well.
The results of our study further contribute to issues regarding the interaction of parafoveal information and higher-level cognitive processing. For example the types of context that can influence skipping are limited to those that are highly predictable. This could reflect a limitation of the extent to which words are lexically processed prior to being fixated (i.e., prior to skipping in E-Z Reader the reader makes a hedged bet that recognition is imminent, but the word is not yet fully lexically accessed), and that very early lexical processes are dissociated from later processes that involve syntactic integration (Staub, 2011). Predictability is unique then in that it appears to affect the same early word recognition processes as frequency unlike any other form of context. However, like other effects from the context on fixation durations (e.g., syntactic fit; Staub, 2011) the influences of predictability and frequency on skipping appear to be independent.
It is notable that regardless of the level of constraint for the target word it was very often regressed back to it after having been skipped (regression rate 0.85 following a skip vs. 0.08 following fixation). This seems odd in light of our interpretation of the role of constraint: if constraint buys the reader confidence in the identity of an upcoming word, then they should be less likely to regress back to it when it is highly constrained (even if it is initially skipped). One possibility is that since the target words were typically the main verb of the sentence, adequate comprehension of the sentence necessitates fully processing it. Another possibility is that skipping short words is generally a “risky” (error-prone) behavior. Indeed, this suggestion has been made before (though not exclusively for short words). Rayner et al. (2006) found that older readers skipped more words than younger readers, but as a consequence made more regressions. They modeled this pattern using E-Z Reader by modulating the parameter corresponding to the rate at which words are “guessed” from the parafovea to respond not just to cloze probability, but also to frequency (such that high-frequency words can also be “guessed” in this manner). This could very well be exactly what happened in the present study: readers skipped short words because they were often “guessed” based on their high frequency. However, if the result of this process is less time devoted to parafoveal processing, then it could be that these words are not in fact processed very deeply and require additional viewing time, hence the increased regressions.
It is important for future research to establish whether or not the effects reported here, and the general finding of greater skipping rates for higher-frequency word previews (as in Angele et al., 2014; Angele & Rayner, 2013) is unique to very short (e.g., 3-letter) words or generalizes to longer words. It may be that lexical properties like word frequency play a larger role in skipping shorter (e.g., 3-letter) compared to longer (e.g., 5–6 letter) words. There is some direct support for this idea within the E-Z Reader model, since the rate of parafoveal processing is limited by the mean eccentricity of the parafoveal word from the point of fixation. Since longer words are on average further away from the point of fixation than shorter words, parafoveal processing will be slower. Though this accounts for why long words are unlikely to be skipped in general, it may also predict different effects of parafoveal preview frequency on skipping rates as length increases.
In summary, both contextual constraint and parafoveal information influence word skipping behavior during reading independently. There seems to be no cross talk between the two sources of information, even when there is a conflict between them. When the string in parafoveal vision is more easily identifiable than a highly predictable word of the same length that should appear in that position, readers do not appear to detect the mismatch between these sources of information. E-Z Reader predicts the same pattern if we assume that readers consider the cloze probability of the expected word in lieu of the word actually available in parafoveal vision. If they do detect the mismatch, it is likely at a stage of processing at which the decision to skip the upcoming word cannot be canceled. These results are important for our understanding of the extent to which higher-level cognitive processing interacts with other aspects of lexical processing in parafoveal vision when making the decision to skip.
Acknowledgments
This research was supported by grant HD065829 from the National Institute of Child Health and Human Development and the results were presented at the 27th Annual CUNY Conference on Human Sentence Processing in Columbus, Ohio and the Psychonomic Society’s 55th Annual Meeting in Long Beach, California. We thank Aleksandar Sheringov for assistance with data collection. We would also like to thank Erik D. Reichle and two anonymous reviewers for their comments on an earlier version of this manuscript. In memoriam, we thank Keith Rayner for advising us in our graduate studies (and beyond) and for imparting his wisdom throughout this study and many others. We are very fortunate to have had the opportunity to work with such an astounding and influential scientist, and he is greatly missed.
Appendix: Sentences used in the experiment as well as corresponding cloze probabilities and Francis and Kučera (1982) frequency values (in counts per million) for the target words as submitted to E-Z Reader 10. The target words are italicized. Items marked with an asterisk did not have frequency values in Francis and Kučera (1982) and were excluded from the E-Z Reader simulations
| Sentence Frame | Constraint | Sentence | Cloze | Frequency |
|---|---|---|---|---|
| 1 | High | If you are shot in the heart you will die very quickly unless you are treated. | 0.8 | 73 |
| 1 | Low | If you are shot in the toe you will die very quickly unless you are treated. | 0.0 | 73 |
| 2 | High | To win money at the horse races you must bet your money on the best horse. | 0.8 | 20 |
| 2 | Low | To win the most at the event you must bet your money very wisely. | 0.0 | 20 |
| 3 | High | If the price isn’t listed on the menu you should ask your server when he comes. | 1.0 | 128 |
| 3 | Low | If the price isn’t listed on the website you should ask your server when he comes. | 0.0 | 128 |
| 4 | High | I got a deep gash when a dog violently bit into my thigh. | 0.8 | 101 |
| 4 | Low | I got a deep gash when a man violently bit into my thigh. | 0.0 | 101 |
| 5 | High | There will be more old people as the population is set to rapidly age in coming years. | 0.5 | 227 |
| 5 | Low | There will be more poor people as the population is set to rapidly age in coming years. | 0.0 | 227 |
| 6 | High | To advance in the tournament you must win each game that you play. | 0.8 | 55 |
| 6 | Low | To advance in the system you must win each game that you play. | 0.0 | 55 |
| 7 | High | To compute the sum you must add all of the numbers together. | 0.8 | 88 |
| 7 | Low | To compute the result you must add all of the numbers together. | 0.2 | 88 |
| 8 | High | Jane’s husband isn’t truthful and will often lie about trivial things. | 0.9 | 59 |
| 8 | Low | Jane’s husband isn’t helpful and will often lie about trivial things. | 0.1 | 59 |
| 9 | High | John found his old suit but it no longer fit properly at all. | 1.0 | 75 |
| 9 | Low | John found his old watch but it no longer fit properly at all. | 0.0 | 75 |
| 10 | High | Jane used the scissors to carefully cut scraps of paper into neat shapes. | 0.8 | 192 |
| 10 | Low | Jane used the machine to carefully cut scraps of paper into neat shapes. | 0.2 | 192 |
| 11 | High | After a shower I grab a towel to quickly dry myself off before putting on clothes. | 0.9 | 68 |
| 11 | Low | After a jog I take a minute to quickly dry myself off before going back inside. | 0.0 | 68 |
| 12 | High | When your baby is hungry she will loudly cry until you come to feed her. | 0.9 | 48 |
| 12 | Low | When you cat is hungry she will loudly cry until you come to feed her. | 0.0 | 48 |
| 13 | High | When you are fasting you cannot eat for certain periods of time. | 1.0 | 61 |
| 13 | Low | When you are working you cannot eat for certain periods of time. | 0.0 | 61 |
| 14 | High | When he wanted to say yes, he would silently nod instead of saying “yes”. | 0.9 | 12 |
| 14 | Low | When he wanted to say no, he would silently nod instead of shaking his head. | 0.1 | 12 |
| 15 | High | The author was not sure how his novel should end without upsetting his readers. | 0.6 | 410 |
| 15 | Low | The author was not sure how his day should end without his wife around. | 0.2 | 410 |
| 16 | High | When the lawn got too long my dad asked if I would mow it over the weekend. | 0.8 | * |
| 16 | Low | When it got too long my dad asked if I would mow our lawn over the weekend. | 0.0 | * |
| 17 | High | My pants had holes in them so I asked my mother to kindly sew them back together. | 0.5 | 6 |
| 17 | Low | My shoes had holes in them so I asked my mother to kindly sew them back together. | 0.1 | 6 |
| 18 | High | Using the telescope Jennifer saw galaxies and constellations in the sky. | 0.5 | 352 |
| 18 | Low | Using the device Jennifer saw galaxies and constellations in the sky. | 0.0 | 352 |
| 19 | High | They are going to an alpine resort where they will ski on some very steep slopes. | 0.8 | 5 |
| 19 | Low | They are going to a desert resort where they will ski some unique outdoor slopes. | 0.0 | 5 |
| 20 | High | I like to jog and will go for a short run after work every day. | 0.6 | 212 |
| 20 | Low | I like to swim and will go for a short run after work too. | 0.1 | 212 |
| 21 | High | Each month your landlord expects you to promptly pay your rent and bills. | 1.0 | 172 |
| 21 | Low | Each month your friend expects you to promptly pay your rent and bills. | 0.4 | 172 |
| 22 | High | To change your hair color you can have a stylist dye it any color you like. | 0.6 | * |
| 22 | Low | To change your hair style you can have a stylist dye it any color you like. | 0.0 | * |
| 23 | High | In a theater class you will learn to professionally act in a variety of situations. | 0.7 | 283 |
| 23 | Low | In the evening class you will learn to professionally act in a variety of situations. | 0.0 | 283 |
| 24 | High | As a kid I wore velcro shoes until I could tie laces into a neat bow. | 0.8 | 23 |
| 24 | Low | As a kid I wore ugly shoes until I could tie laces into a neat bow. | 0.0 | 23 |
| 25 | High | If you are happy with your waiter you should tip twenty percent of the bill. | 1.0 | 22 |
| 25 | Low | If you are happy with your experience you should tip twenty percent of the bill. | 0.0 | 22 |
| 26 | High | My dog will lift his leg and quickly pee on every hydrant he sees. | 0.6 | 3 |
| 26 | Low | My cat will lift his tail and quickly pee on the floor when he is scared. | 0.1 | 3 |
| 27 | High | If you stick too much paper in the printer it will jam until it is cleared. | 0.9 | 6 |
| 27 | Low | If you put too much faith in the machine it will jam and break down. | 0.0 | 6 |
| 28 | High | With those strong oars, he could row his boat very far. | 0.6 | 35 |
| 28 | Low | With those strong arms, he could row his boat very far. | 0.0 | 35 |
| 29 | High | A mirror next to a steamy shower will likely fog unless you run the fan. | 1.0 | 25 |
| 29 | Low | A door next to a running shower will likely fog unless you run the fan. | 0.3 | 25 |
| 30 | High | The girl saw the furry puppy and asked if she could pet her and give her a treat. | 0.4 | 8 |
| 30 | Low | The girl saw the furry spider and asked if she could pet her and give her a treat. | 0.1 | 8 |
| 31 | High | The impatient cows will loudly moo when they want to be milked. | 0.9 | * |
| 31 | Low | The impatient animals will loudly moo when they want to be milked. | 0.0 | * |
| 32 | High | I’m tired of renting, and would rather own a nice house or condo now. | 0.7 | 772 |
| 32 | Low | I’m tired of looking, and would rather own a nice house or condo now. | 0.0 | 772 |
| 33 | High | After the concert, the performers humbly bow to their audience. | 0.7 | 15 |
| 33 | Low | After the concert, the guests humbly bow to their Queen. | 0.0 | 15 |
| 34 | High | You should use the shovel to carefully dig your hole to plant the tree. | 1.0 | 10 |
| 34 | Low | You should use the tool to carefully dig your hole to plant the tree. | 0.0 | 10 |
| 35 | High | To clean the floor, fill a bucket and quickly mop until there are no spots. | 0.6 | 3 |
| 35 | Low | To clean the area, take the supplies and quickly mop until there are no spots. | 0.0 | 3 |
| 36 | High | If left out, apples and avocados will quickly rot and turn brown. | 0.5 | 8 |
| 36 | Low | If left out, certain food items will quickly rot and turn brown. | 0.0 | 8 |
| 37 | High | It is fun to take bubble wrap and joyously pop all of the bubbles on it. | 0.9 | 8 |
| 37 | Low | It is fun to take bubble gum and joyously pop all of the bubbles you make. | 0.0 | 8 |
| 38 | High | Instead of a kiss you could hug your children when they leave. | 0.9 | 3 |
| 38 | Low | Instead of a wave you could hug your children when they leave. | 0.0 | 3 |
| 39 | High | To make the doughnuts, heat some oil and quickly fry until they are golden-brown. | 0.4 | * |
| 39 | Low | To make the dish, heat your surface and quickly fry until the food is golden-brown. | 0.1 | * |
| 40 | High | An angry audience will often boo if the performer makes a racist comment. | 0.8 | 1 |
| 40 | Low | An engaged audience will often boo if the performer makes a racist comment. | 0.0 | 1 |
Footnotes
There was one exception where the target word run was used as a noun instead of a verb (see the Appendix).
Four of the target words do not have corresponding entries in the Francis and Kučera (1982).
The model fit to skipping data on the pre-target word that is reported estimates only random intercepts for subjects and items (no random slopes), because the model including random slopes indicated very high correlations between the random slopes and intercepts. The likelihood ratio test (LRT) indicated that removing the random slopes from the model did not hurt model performance (p = 1.00). The model fit to target word go-past time data failed to converge. The predictability term was dropped from the by-items random effects structure, the model was refit, and successfully converged. The same situation occurred when fitting the model to skipping data for the post-target word, and the resulting model including additive random effects slopes for preview and predictability by subjects and preview by items.
E-Z Reader 10 is available as a Java program at http://www.erikdreichle.com/ezreader.html.
Four of the target words do not have corresponding entries in the Francis and Kučera (1982) corpus; these items were not included in the simulations. Frequencies were not available for 11 words (3%); these words were assigned the median frequency of words used in the experiment (937 counts per million). We note that this was done simply in order for the model to run simulations to completion. We did not replace missing frequency counts for the target words with the value that we used to replace other missing frequency counts because this value is much higher than the median frequency of the target words (18), and there is good reason to believe that the actual frequencies of these words would be quite low, given that they do not appear in the corpus.
References
- Angele B, Rayner K. Processing “the” in the parafovea: Are articles skipped automatically? Journal of Experimental Psychology: Learning, Memory, and Cognition. 2013;39:649–662. doi: 10.1037/a0029294. [DOI] [PubMed] [Google Scholar]
- Angele B, Laishley A, Rayner K, Liversedge SP. The effect of high- and low-frequency previews on and sentential fit on word skipping during reading. Journal of Experimental Psychology: Learning, Memory, and Cognition. 2014 doi: 10.1037/a0036396. in press. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Balota DA, Pollatsek A, Rayner K. The interaction of contextual constraints and parafoveal visual information in reading. Cognitive Psychology. 1985;17:364–390. doi: 10.1016/0010-0285(85)90013-1. [DOI] [PubMed] [Google Scholar]
- Bates D, Maechler M, Bolker B. Lme4: Linear mixed-effects models using S4 classes. R package version 0.999999-2. 2013 http://CRAN.R.project.org/package=lme4.
- Drieghe D, Pollatsek A, Staub A, Rayner K. The word grouping hypothesis and eye movements during reading. Journal of Experimental Psychology: Learning, Memory, and Cognition. 2008;34:1552–1560. doi: 10.1037/a0013017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Drieghe D, Rayner K, Pollatsek A. Eye movements and word skipping during reading revisited. Journal of Experimental Psychology: Human Perception and Performance. 2005;31:954–969. doi: 10.1037/0096-1523.31.5.954. [DOI] [PubMed] [Google Scholar]
- Ehrlich SF, Rayner K. Contextual effects on word perception and eye movements during reading. Journal of Verbal Learning and Verbal Behavior. 1981;20:641–655. [Google Scholar]
- Engbert R, Nuthmann A, Richter E, Kliegl R. SWIFT: A dynamical model of saccade generation during reading. Psychological Review. 2005;112:777–813. doi: 10.1037/0033-295X.112.4.777. [DOI] [PubMed] [Google Scholar]
- Fitzsimmons G, Drieghe D. How fast can predictability influence word skipping during reading? Journal of Experimental Psychology: Learning, Memory, and Cognition. 2013;39:1054–1063. doi: 10.1037/a0030909. [DOI] [PubMed] [Google Scholar]
- Francis WN, Kucera H. Frequency analysis of English usage: Lexicon and grammar. Boston: Houghton Mifflin; 1982. [Google Scholar]
- Gautier V, O’Regan JK, Le Gargasson JF. “The-skipping”revisited in French: programming saccades to skip the article “les”. Vision Research. 2000;40:2517–2531. doi: 10.1016/s0042-6989(00)00089-4. [DOI] [PubMed] [Google Scholar]
- Gollan TH, Slattery TJ, Goldenberg D, Van Assche E, Duyck W, Rayner K. Frequency drives lexical access in reading but not in speaking: The frequency-lag hypothesis. Journal of Experimental Psychology: General. 2011;140:186–209. doi: 10.1037/a0022256. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Kliegl R, Engbert R. Fixation durations before word skipping in reading. Psychonomic Bulletin & Review. 2005;12:132–138. doi: 10.3758/bf03196358. [DOI] [PubMed] [Google Scholar]
- Nuthmann A, Engbert R, Kliegl R. Mislocated fixations during reading and the inverted optimal viewing position effect. Vision Research. 2005;45(17):2201–2217. doi: 10.1016/j.visres.2005.02.014. [DOI] [PubMed] [Google Scholar]
- O’Regan JK. Eye guidance in reading: Evidence for the linguistic control hypothesis. Perception & Psychophysics. 1979;25:501–509. doi: 10.3758/bf03213829. [DOI] [PubMed] [Google Scholar]
- O’Regan JK. The control of saccade size and fixation duration in reading: The limits of linguistic control. Perception & Psychophysics. 1980;28:112–117. doi: 10.3758/bf03204335. [DOI] [PubMed] [Google Scholar]
- Pollatsek A, Rayner K. Eye movement control in reading: The role of word boundaries. Journal of Experimental Psychology: Human Perception and Performance. 1982;8:817–833. [Google Scholar]
- R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing; Vienna, Austria: 2013. http://www.R-project.org/ [Google Scholar]
- Rayner K. The perceptual span and peripheral cues in reading. Cognitive Psychology. 1975;7:65–81. [Google Scholar]
- Rayner K. Eye movements in reading and information processing: 20 years of research. Psychological Bulletin. 1998;124:372–422. doi: 10.1037/0033-2909.124.3.372. [DOI] [PubMed] [Google Scholar]
- Rayner K. The Thirty-Fifth Sir Frederick Bartlett Lecture: Eye movements and attention in reading, scene perception, and visual search. The Quarterly Journal of Experimental Psychology. 2009;62:1457–1506. doi: 10.1080/17470210902816461. [DOI] [PubMed] [Google Scholar]
- Rayner K, Fischer MH, Pollatsek A. Unspaced text interferes with both word identification and eye movement control. Vision Research. 1998;38:1129–1144. doi: 10.1016/s0042-6989(97)00274-5. [DOI] [PubMed] [Google Scholar]
- Rayner K, McConkie GW. What guides a reader’s eye movements. Vision Research. 1976;16:829–837. doi: 10.1016/0042-6989(76)90143-7. [DOI] [PubMed] [Google Scholar]
- Rayner K, Reichle ED, Stroud MJ, Williams CC, Pollatsek A. The effect of word frequency, word predictability, and font difficulty on the eye movements of older and younger readers. Psychology and Aging. 2006;21:448–465. doi: 10.1037/0882-7974.21.3.448. [DOI] [PubMed] [Google Scholar]
- Rayner K, Sereno SC, Raney GE. Eye movement control in reading: A comparison of two types of models. Journal of Experimental Psychology: Human Perception and Performance. 1996;22:1188–1200. doi: 10.1037//0096-1523.22.5.1188. [DOI] [PubMed] [Google Scholar]
- Rayner K, Slattery TJ, Drieghe D, Liversedge SP. Eye movements and word skipping during reading: Effects of word length and predictability. Journal of Experimental Psychology: Human Perception and Performance. 2011;37:514–528. doi: 10.1037/a0020990. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Rayner K, Well AD. Effects of contextual constraint on eye movements in reading: A further examination. Psychonomic Bulletin & Review. 1996;3:504–509. doi: 10.3758/BF03214555. [DOI] [PubMed] [Google Scholar]
- Reichle ED, Drieghe D. Using E-Z Reader to examine word skipping during reading. Journal of Experimental Psychology: Learning, Memory, and Cognition. 2013;39:1331–1320. doi: 10.1037/a0030910. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Reichle ED, Pollatsek A, Fisher DL, Rayner K. Toward a model of eye movement control in reading. Psychological Review. 1998;105:125–157. doi: 10.1037/0033-295x.105.1.125. [DOI] [PubMed] [Google Scholar]
- Reichle ED, Pollatsek A, Rayner K. Using EZ Reader to simulate eye movements in nonreading tasks: A unified framework for understanding the eye–mind link. Psychological review. 2012;119:155–185. doi: 10.1037/a0026473. [DOI] [PubMed] [Google Scholar]