Eye Movement Control During Reading : Effects of Word Frequency and Orthographic Familiarity

Word frequency and orthographic familiarity were independently manipulated as readers’ eye movements were recorded. Word frequency influenced fixation durations and the probability of word skipping when orthographic familiarity was controlled. These results indicate that lexical processing of words can influence saccade programming (as shown by fixation durations and which words are fixated). Orthographic familiarity, but not word frequency, influenced the duration of prior fixations. These results provide evidence for orthographic, but not lexical, parafoveal-on-foveal effects. Overall, the findings have a crucial implication for models of eye movement control in reading: There must be sufficient time for lexical factors to influence saccade programming before saccade metrics and timing are finalized. The conclusions are critical for the fundamental architecture of models of eye movement control in reading— namely, how to reconcile long saccade programming times and complex linguistic influences on saccades during reading.

As we read, the linguistic characteristics of words influence the duration of fixations and which words are fixated (Rayner, 1998).
The present study provides a detailed examination of the influences of orthographic familiarity and word frequency on eye movements during reading.It thus provides a critical assessment of the relationship between linguistic text processing and the systems that control when and where the eyes move.Specifically, whether lexical processing can have an immediate influence on saccade programming is examined.The issues addressed here have crucial implications for the architecture of models of eye movement control in reading.This article adds to a growing number of recent studies specifically aiming to test and develop such accounts (Inhoff, Eiter, & Radach, 2005;Kliegl, Nuthmann, & Engbert, 2006;Rayner, Ashby, Pollatsek, & Reichle, 2004;Rayner, Juhasz, & Brown, 2007;Rayner, Liversedge, White, & Vergilino-Perez, 2003;Reingold & Rayner, 2006).Previous studies of the effects of word frequency and orthography are discussed first.Models of eye movements in reading, which may account for such effects, are summarized.Finally, the issue of whether processing of parafoveal information can influence prior fixations (parafoveal-on-foveal effects) and where words are fixated (saccade specification) is considered.

Word Frequency Effects
The influence of word frequency on word processing is an established finding both for isolated word response time tasks (Monsell, 1991) and sentence reading (Rayner, 1998).Inhoff and Rayner (1986; see also Rayner & Duffy, 1986) first demonstrated that for words in sentences with word length controlled, first fixation durations and gaze durations (the sum of fixations before leaving a word) are longer on infrequent than on frequent words.Word frequency effects during sentence reading are usually spatially localized to the word that induced those effects (Henderson & Ferreira, 1990, 1993;Raney & Rayner, 1995), and in other cases there are effects of word frequency both on the word itself and on subsequent spillover fixations (Kennison & Clifton, 1995;Kliegl et al., 2006;Rayner & Duffy, 1986).Frequent words are also more likely to be skipped than infrequent words (e.g., Rayner, Sereno, & Raney, 1996; see also Brysbaert, Drieghe, & Vitu, 2005;Brysbaert & Vitu, 1998).Note that other factors, such as age of acquisition and concreteness, have been shown to have independent influences on reading behavior even though they are correlated with word frequency (Juhasz & Rayner, 2003).
The influence of word frequency on eye movement behavior during reading suggests that lexical word recognition processes can influence when and where the eyes move.However, of the large number of studies that have manipulated word frequency, a very small proportion have attempted to control for the orthographic characteristics of the words.It is easy to confound word frequency with orthographic familiarity because frequent words are necessarily orthographically familiar, whereas many infrequent words are orthographically unfamiliar.A few studies have attempted to control for orthographic characteristics using the measure of type frequency, which is the number of words that contain a particular letter sequence.Rayner and Duffy (1986) undertook post hoc analyses that showed higher trigram-type frequencies for frequent than for infrequent target words, but they found that word frequency effects still held for those items in which differences in type frequency were reversed.Bertram and Hyo ¨na ¨(2003) showed word frequency effects on first fixation and gaze durations when bigram type frequency was controlled.Rayner et al. (1996) reported effects of word frequency on fixation durations and word skipping for items that had equal monogram-and bigram-type frequency counts.
Importantly, previous reading studies only attempted to control for orthography by using type frequency counts.As type frequency is simply the number of words that contain a particular letter sequence, it is effectively a measure of lexical informativeness or redundancy.For example, the trigram pne at the word beginning has a very low type frequency and is very informative because it highly constrains the number of possible word candidates ( pneumatic, pneumonia).Importantly, type frequency does not reflect letter sequence familiarity.For example, there are a number of words that begin with the letter sequence irr, but very few of these are very frequent; hence, although irr has a relatively high type frequency, it actually has quite low orthographic familiarity.Manipulations of only type frequency or informativeness should therefore reflect processing at the level of lexical candidates.
A better measure of orthographic familiarity is token frequency, which is the sum of the frequencies of words that contain a particular letter sequence. 1Critically, effects of orthographic familiarity may involve processing at a sublexical or even a visual level (see the Orthographic Effects section below).Note that other studies have tested for effects of word frequency, and the type and token frequency of word-initial letters, in multiple isolated wordprocessing tasks (Kennedy, 1998(Kennedy, , 2000;;Kennedy, Pynte, & Ducrot, 2002) and sentence reading (Kennedy & Pynte, 2005;Pynte & Kennedy, 2006).However, the orthographic characteristics beyond the word-initial letters were not controlled in these studies and, consequently, they did not test whether word frequency effects occur independently of differences in orthographic familiarity.To summarize, although previous studies of word frequency have attempted to control for orthography to some extent, these studies did not eliminate the possibility that differences in orthographic familiarity could have produced, or at least influenced, the word frequency effect.
The issue of whether orthographic familiarity may be contributing to the word frequency effect is of particular importance for the case of word skipping.Critically, the linguistic characteristics of skipped words must be processed in parafoveal vision (where stimuli are degraded due to acuity limitations), which reduces the speed of linguistic processing of those words (Rayner & Morrison, 1981;Schiepers, 1980).In addition, fixations prior to word skipping are likely to involve processing of the fixated word too, which may reduce the time (or resources) available within a fixation for processing of the parafoveal word (e.g., Morrison, 1984).Of the studies that have shown word frequency effects on word skipping, only one study controlled for type-bigram frequency (Rayner et al., 1996).Not only was orthographic familiarity not controlled but orthographic processing of trigrams, not just bigrams, could have contributed to the effect.Furthermore, the fact that some studies have not shown word frequency effects on word skipping (Calvo & Meseguer, 2002;Henderson & Ferreira, 1993) raises the possibility that other factors, such as orthographic familiarity, may be important.Visual familiarity may play an important role in word processing (Martin, 2004), and some models suggest that fixation durations (McDonald, Carpenter, & Shillcock, 2005) and especially word skipping (Engbert, Nuthmann, Richter, & Kliegl, 2005) are not necessarily driven by full word identification.Con-sequently, it is particularly feasible that processing of orthographic familiarity may have produced the previously reported effects of word frequency on word skipping.
To summarize, previous studies have not provided a proper test of whether word frequency influences fixation durations and word skipping when orthographic familiarity is fully controlled.Differences in orthographic familiarity may have caused, or at least inflated, effects that have been attributed to word frequency.Consequently, previous studies of word frequency do not categorically demonstrate lexical influences on word recognition and eye movement behavior.The issue of whether word frequency effects may be explained by differences in orthographic processing is crucial.If the effects are due to differences in orthography, this raises the critical question of whether lexical factors can have an immediate impact on when and where the eyes move during reading.Basically, given that this issue is so critical for models of eye movement control during reading, it is absolutely essential that a thorough analysis of the effects of word frequency and orthographic familiarity be undertaken.The present study did this by testing the effects of word frequency on eye movement behavior while controlling for monogram, bigram, and trigram token orthographic frequencies.

Orthographic Effects
Studies using isolated word tasks have investigated a wide range of factors related to orthographic processing of words (L.Henderson, 1982).A number of early studies were undertaken into the effects of bigram frequency and word frequency during reading (Gernsbacher, 1984).These studies used isolated word methods such as tachistoscopic presentation of words, naming, and lexical decision, and bigram frequency was controlled only by type frequency counts such as those reported by Mayzner and Tresselt (1965).However, the findings of such studies using low-frequency words have been contradictory, with some indicating that words with high-frequency bigrams are more difficult to process than those with low-frequency bigrams (Broadbent & Gregory, 1968;Rice & Robinson, 1975), while others have shown the opposite result (Biederman, 1966;Seidenberg, Waters, Barnes, & Tanenhaus, 1984;Waters & Seidenberg, 1985).However, as noted above, letter sequences can have different levels of informativeness and familiarity, and it is possible that such differences may be one explanation for the inconsistent results (Grainger & Dijkstra, 1996).
A number of studies using eye-tracking methodologies have examined the role of the informativeness of letter sequences (confounded with orthographic familiarity).Pynte, Kennedy, and Murray (1991) showed longer fixation durations on informative parts of words (see also Holmes & O'Regan, 1987).In sentence reading, Lima and Inhoff (1985) showed that first fixation durations were longer on words that began with constraining (e.g., dwarf) compared with less constraining (e.g., clown) letter sequences.Other experiments have shown that the orthographic familiarity of the initial letters of long words can influence where they are first fixated (Hyo ¨na ¨, 1995;Radach, Inhoff, & Heller, 2004;Vonk, Radach, & van Rijn, 2000;White & Liversedge, 2004, 2006a, 2006b).
Orthographic familiarity may affect visual, sublexical, or lexical levels of processing.Processing of text at a visual, rather than linguistic, level may modulate the familiarity of visual information such that frequent letter strings may develop higher visual familiarity than infrequent letter strings (Findlay & Walker, 1999).Therefore, any effects of orthographic familiarity must be interpreted as a reflection of processing at least at the level of visual familiarity.However, differences in orthographic familiarity may also be associated with differences in informativeness or constraint (type frequency), and orthographically unfamiliar words may also tend to have more irregular phonology and fewer orthographic neighbors.Consequently, any effects of orthographic familiarity could also be driven by sublexical or even lexical processes.The present study manipulated the orthographic familiarity of the entire word, and it examined whether orthographic familiarity influences fixation durations and word skipping.
In contrast, it has been suggested (Deubel, O'Regan, & Radach, 2000;Nazir, 2000), and models have proposed (McConkie & Yang, 2003;O'Regan, 1990O'Regan, , 1992;;Reilly & O'Regan, 1998;Suppes, 1990;Yang, 2006;Yang & McConkie, 2001), that eye movements are largely controlled by visual and oculomotor factors and that linguistic processes have a smaller influence, for example by delaying or cancelling programmed saccades.However, such models cannot necessarily account for some linguistic influences on reading behavior, such as the following frequency effects: (a) on the first of multiple first-pass fixations on a word (Rayner et al., 1996), (b) in the absence of visual information (Liversedge et al., 2004;Rayner, Liversedge, et al., 2003), (c) and on gaze durations that cannot be explained by a number of very long fixations or gaze durations (Rayner, 1995;Rayner, Liversedge, et al., 2003).Linguistic factors, therefore, have a critical role in influencing eye movement behavior during reading, and visual-and/or oculomotor-based models cannot fully account for reading eye movement behavior.However, the issue of precisely how linguistic processing influences when and where the eyes move during reading is far from resolved.
The most comprehensive implemented models of eye movement behavior during reading-the E-Z reader model (Pollatsek et al., 2006;Reichle et al., 1998Reichle et al., , 1999Reichle et al., , 2003Reichle et al., , 2006) ) and the SWIFT ([Autonomous] Saccade-Generation With Inhibition by Foveal Targets) model (Engbert et al., 2002(Engbert et al., , 2005;;Kliegl & Engbert, 2003;Richter et al., 2006)-include a period of saccade programming during which the planned timing and metrics of the subsequent saccade are prepared (see also Salvucci, 2001).The time necessary to program a saccade has been estimated to be 175-200 ms (Rayner, Slowiaczek, Clifton, & Bertera, 1983), or 125 ms once 50 ms of mandatory visual processing has been taken into account (Pollatsek et al., 2006).Fixation durations in reading are on average 200 -250 ms long (Rayner, 1998), and so a substantial portion of the time during a fixation could involve programming the next saccade.Although linguistic processing can continue during this period, it may not necessarily influence saccade programming.
Saccade programming is composed of a labile period, during which the saccade can be cancelled, and a nonlabile period, during which the saccade metrics are finalized and the saccade cannot be cancelled (Becker & Ju ¨rgens, 1979;Deubel et al., 2000).In the E-Z reader model, the mean durations of the labile and nonlabile stages are 100 ms and 25 ms, respectively (Pollatsek et al., 2006).In the SWIFT model, the labile stage is between 50 and 150 ms, and the nonlabile stage is between 5 and 50 ms (Engbert et al., 2005) or else these values are fixed at 150 ms and 50 ms, respectively (Richter et al., 2006).In the E-Z reader and SWIFT models, linguistic factors can determine, or influence, when the labile stage of saccade programming commences.In both cases, the time at which the saccade program is triggered effectively influences when the subsequent saccade is executed and, therefore, the duration of fixations.Critically, given the time required for saccade programming (ϳ125 ms) and the average duration of fixations (ϳ250 ms), such linguistic influences on when saccade programming is initiated would necessarily have to occur relatively early during fixations.
Linguistic processing during the labile stage of saccade programming may also influence current fixation durations or the next saccade target.In both the E-Z reader and SWIFT models, such mechanisms have been adopted to account for word skipping.In the E-Z reader model, the labile stage can be cancelled and restarted such that the word target is changed and a new saccade program can commence.In contrast, in the SWIFT model, the saccade target location is always specified at the end of the labile stage.Note that even when linguistic factors impact late during a fixation at the end of the labile stage of saccade programming, there will still be some delay before the saccade can be executed due to the subsequent nonlabile stage.
The issue of how linguistic processing can influence saccade programming, taking account of saccade programming times, is one of the most critical issues for the design of models of eye movements in reading.The present study tested whether lexical factors (word frequency) influence saccade programming (fixation durations and word skipping) independent of sublexical factors (orthographic familiarity).If the lexical factor of word frequency influences fixation durations or word skipping, this must be explained by either an early linguistic influence on when saccade programming is initiated or a later influence during, or at the end of, the labile stage of saccade programming.
Both the E-Z reader and SWIFT models suggest that word frequency affects saccade programming, and so they predict that this influences both fixation durations and word skipping.However, both models are vague about precisely how word frequency affects the difficulty of word processing.Given the timing constraints of the proposal that linguistic factors trigger the initiation of saccade programming, it is quite plausible that the sublexical factor of orthographic familiarity, and not the lexical factor of word frequency, might influence saccades within such a framework.Nevertheless, if the present study shows that word frequency does have an influence-for example, on how long words are first fixated-then this would suggest that the linguistic processes that occur prior to saccade programming in these models must be of a lexical nature.

Parafoveal-on-Foveal Effects of Word Frequency and Orthography
Many studies have shown that parafoveal processing of words to the right of fixation can influence the probability of word skipping and can also facilitate subsequent processing of those words when they are later fixated, a phenomenon known as preview benefit (Rayner & Pollatsek, 1981).However, the issue of whether parafoveal processing of words can influence fixations prior to fixation of those words (fixation n Ϫ 1)-a phenomenon known as parafoveal-on-foveal effects-is controversial.Previous research testing whether orthographic familiarity and word frequency produce parafoveal-on-foveal effects is reviewed below.The investigation of parafoveal-on-foveal effects is especially critical because it has important implications for models of eye movements in reading.Serial processing models suggest that words are lexically processed one at a time (Morrison, 1984;Pollatsek et al., 2006;Reichle et al., 1998Reichle et al., , 1999Reichle et al., , 2003Reichle et al., , 2006)).In contrast, parallel accounts suggest that multiple words can be processed at once (Engbert et al., 2002(Engbert et al., , 2005;;Kennedy, 2000;Kliegl & Engbert, 2003;McDonald et al., 2005;Reilly & Radach, 2003, 2006;Richter et al., 2006).If the characteristics of a parafoveal word influence fixation n Ϫ 1, this could be indicative of parallel processing of words.There are alternative explanations for these effects, such as inaccurate saccade targeting, which are explained below and further in the Discussion.To be clear, parallel-based models predict both sublexical and lexical parafoveal-on-foveal effects, whereas the core assumptions of serial-based models do not predict such effects, although additional assumptions or explanations have been offered to account for these.Importantly, the present study provided an important opportunity to test for the presence of parafoveal-on-foveal effects in a carefully controlled experiment.
Many fixations are mislocated during reading (McConkie, Kerr, Reddix, & Zola, 1988;Nuthmann, Engbert, & Kliegl, 2005), and it has been suggested that parafoveal-on-foveal effects may arise because of such inaccurately targeted saccades (Drieghe, Rayner, & Pollatsek, in press;Rayner, Warren, Juhasz, & Liversedge, 2004;Rayner, White, et al., 2003).That is, saccades intended to land on the critical word mistakenly land at the end of the previous word, but attention is still allocated to the originally intended location such that processing of the critical word influences the fixation duration on the previous word.Importantly, these effects would be expected to be in a conventional direction.For example, lower frequency and less familiar parafoveal words would produce longer fixations on foveal words than would higher frequency and more familiar parafoveal words.Reports of postlexical parafovealon-foveal effects have shown effects in a conventional direction that might, therefore, be explained by inaccurate saccade targeting (Inhoff, Radach, Starr, & Greenberg, 2000;Murray, 1998;Rayner, Warren, et al., 2004).Nevertheless, Starr and Inhoff (2004) showed that parafoveal-on-foveal effects of orthographic familiarity held even when prior fixations at the very end of the foveal word were eliminated from analysis.Furthermore, other studies that have shown that parafoveal information influences the probability of refixating on the foveal word, or that the parafoveal-onfoveal effects are opposite to the conventional direction, cannot be explained by the inaccurate saccade targeting account (Kennedy, 1998(Kennedy, , 2000;;Kennedy & Pynte, 2005;Kennedy et al., 2002).
The present study provided an additional test of whether sublexical (orthographic familiarity) and lexical (word frequency) characteristics of parafoveal words influence foveal fixations in the form of parafoveal-on-foveal effects.The question of whether there are lexical parafoveal-on-foveal effects in sentence reading under carefully controlled experimental conditions is particularly important because, so far, the only sentence reading studies to have shown parafoveal-on-foveal effects of word frequency have been based on a corpus of reading data (Kennedy & Pynte, 2005;Pynte & Kennedy, 2006) or data from a combination of different studies (Kliegl et al., 2006).It is important to assess whether lexical parafoveal-on-foveal effects arise under standard experimental conditions when the stimuli are designed to control for other variables.In the present study, sentence beginnings were identical across the conditions, and none of the critical words were predictable from the sentence context.In addition, both foveal and parafoveal words were short, therefore providing optimal conditions for parafoveal preview and, consequently, for parafoveal-on-foveal effects to occur.

Saccade Specification and Regressions
The current study also provided an opportunity to test whether orthographic familiarity and word frequency influence saccade length into short words or where short words are first fixated.As noted above, previous research suggests that long words with orthographically unfamiliar word beginnings are first fixated nearer to the beginning of the word than are long words with orthographically familiar beginnings (Hyo ¨na ¨, 1995;Radach et al., 2004;Vonk et al. 2000;White & Liversedge, 2004, 2006a, 2006b), whereas Rayner et al. (1996) showed no effect of word frequency on where words are first fixated.Importantly, previous studies have not tested whether orthographic familiarity can influence where short words are first fixated.Note that models of eye movement control during reading generally predict that the only parafoveal information used to influence saccade specification to word targets is word length; hence, they would predict no effects of orthographic familiarity or word frequency on initial fixation positions (e.g., Engbert et al., 2005;Reichle et al., 1999).Reichle et al. (2003) suggested that low spatial frequency information (such as ascenders and descenders) might influence saccade programming; however, they did not specify exactly what influence this information would have on where words are fixated (Liversedge & White, 2003).

Summary
The present study provided the first comprehensive test of the independent effects of word frequency and (whole word) orthographic familiarity on word recognition processes, as shown by eye movement behavior during reading.The main focus of the study was to test whether lexical processes affect saccade programming such that fixation durations and word skipping are influenced by the lexical characteristics of words.The study also provided an opportunity to test whether the orthographic or lexical characteristics of parafoveal words might be processed in parallel with the fixated word, or at least produce parafoveal-on-foveal effects.Measures of saccade specification, such as where words are initially fixated, are also reported, as are analyses of whether orthographic familiarity or word frequency influence regressions out of, or into, words.
To investigate the independent effects of word frequency and orthographic familiarity, an eye movement experiment was undertaken with three conditions.Word frequency effects were examined by comparing frequent (e.g., town) and infrequent (e.g., cove) words that were equally orthographically familiar.Orthographic familiarity effects were examined by comparing orthographically familiar (e.g., cove) and unfamiliar (e.g., quay) words that were equally infrequent.That is, the orthographically familiar low word frequency condition was used for both analyses.It was not possible to orthogonally manipulate the variables, because words with high frequencies cannot be orthographically unfamiliar.

Participants
Thirty students at the University of Durham, Durham, United Kingdom, participated in the experiment.They all had normal or corrected-to-normal vision and were naive regarding the purpose of the experiment.

Materials and Design
The critical words were (a) frequent and orthographically familiar, (b) infrequent and orthographically familiar, or (c) infrequent and orthographically unfamiliar.These three conditions were manipulated within participants and items.
Word frequencies and n-gram frequencies were calculated in counts per million using the CELEX English word form corpus (Baayen, Piepenbrock, & Gulikers, 1995).The word form, rather than lemma, corpus was selected to ensure inclusion of letter sequences within all words.For example, the word form corpus includes words such as went, whereas in the lemma corpus, frequencies for such words would be associated instead with the base form (e.g., go).For the orthographically familiar words, the frequent words had higher word frequencies (M ϭ 297, SD ϭ 166) than the infrequent words (M ϭ 1.7, SD ϭ 0.9), t(38) ϭ 11.1, p Ͻ .001.For the infrequent words, there were no differences in word frequency between the orthographically unfamiliar (M ϭ 1.5, SD ϭ 1.0) and familiar (M ϭ 1.7, SD ϭ 0.9) conditions (t Ͻ 1).
Previous research indicates that letter-position coding is quite flexible (e.g., Humphreys, Evett, & Quinlan, 1990).Consequently n-gram frequencies were calculated both specific to position and nonposition specific regardless of word length.Orthographic familiarity was measured using n-gram token frequencies, which represent the sum of the frequencies of the words that contain a particular letter sequence.Token frequencies for each of the n-grams within the word (e.g., for a four-letter word, there are two trigrams, three bigrams, and four monograms) were summed together.Table 1 shows the mean monogram, bigram, and trigram summed frequencies for each of the conditions.The table shows that n-gram frequencies in the two orthographically familiar conditions were the same (or even slightly higher in the infrequent condition).For the infrequent words, n-gram frequencies were always significantly lower in the orthographically unfamiliar condition than in the orthographically familiar condition.The initial trigram is most important in preprocessing of words (Rayner, Well, Pollatsek, & Bertera, 1982).In line with the summed n-gram data, position-specific token-initial trigram frequencies of the critical words were significantly smaller for the unfamiliar (M ϭ 166, SD ϭ 303) than for the familiar (M ϭ 1,307, SD ϭ 3,087) infrequent words, t(39) ϭ 2.28, p Ͻ .05,whereas there was no difference for the familiar frequent (M ϭ 992, SD ϭ 903) and infrequent (M ϭ 1,307, SD ϭ 3,087) words (t Ͻ 1).Type frequencies followed similar patterns to token frequencies for all of the measures described above.
Orthographic familiarity is also associated with differences in the number of orthographic neighbors.The number of orthographic neighbors was calculated using the entire English Lexicon Project database (Balota et al., 2002).There was no difference in the number of orthographic neighbors between the frequent (M ϭ 7.5, SD ϭ 4.5) and infrequent (M ϭ 8.1, SD ϭ 4.8) orthographically familiar words (t Ͻ 1), but there were more orthographic neighbors for the orthographically familiar (M ϭ 8.1, SD ϭ 4.8) than for the orthographically unfamiliar (M ϭ 2.2, SD ϭ 2.7) infrequent words, t(38) ϭ 7.37, p Ͻ .001.The number of higher frequency neighbors for each of the items was also calculated for the frequent and infrequent orthographically familiar words (as noted above, the orthographically unfamiliar infrequent words had very few neighbors).The infrequent orthographically familiar words had significantly more higher frequency neighbors (M ϭ 4.8, SD ϭ 3.8) than the frequent words (M ϭ 0.7, SD ϭ 0.8), t(38) ϭ 6.66, p Ͻ .001.These differences reflect the fact that the frequent words are so highly frequent that there are few words that are more frequent than they are (compared with the infrequent words). 2  To ensure that the critical words were not predictable within the context of the sentence, sentence completion norms were obtained.Twelve participants were given the beginning portions of the sentence up to the critical word and were asked to provide a single word that they felt could fit as the next word in the sentence.Of all of the completions, only two (0.4%) were correct.Therefore, none of the critical words were predictable from the sentence context.
There were 39 critical words in each condition; all were either 4 or 5 letters long and matched for length across the three conditions, with a mean word length of 4.5 characters (SD ϭ 0.5).Each set of critical words was embedded in the same neutral sentential frame up to and including the word after the critical word.Each of the sentences was no longer than one line of text (80 characters).The word preceding the critical word (word n Ϫ 1) was either 5 or 6 letters long, with a mean word frequency of 147 counts per million (SD ϭ 170).A full list of materials is provided in the Appendix.
Three lists of 123 sentences were constructed, and 10 participants were randomly allocated to each list.Each list included all 117 experimental sentences with an additional 6 practice sentences at the beginning.For each third of the lists, the 2 Previous studies that have manipulated the number of higher frequency neighbors during sentence reading have shown either no effects (Sears, Campbell, & Luper, 2006) or late effects (Perea & Pollatsek, 1998; but see Pollatsek, Perea, & Binder, 1999;Sears et al., 2006).It is therefore unlikely that differences in the number of higher frequency neighbors could have produced the differences between the frequent and infrequent word conditions reported here.Nevertheless, any effects of the number of higher frequency neighbors would reflect a lexical level of processing, consistent with the reported conclusions related to effects of word frequency.conditions were rotated across the three lists. 3The order of the items within each third of the list was different, but overall the sentences were presented in a fixed, pseudorandom order, ensuring that repeated items were widely distributed throughout the experiment.Thirty-eight of the sentences were followed by a comprehension question.

Procedure
Eye movements were monitored using a Dual-Purkinje-Image eye tracker.Viewing was binocular, but only the movements of one eye were monitored (the right eye was monitored for 18 participants, and the left eye was monitored for 12 participants). 4 The sentences were presented in light cyan on a black background, with characters presented in Courier font.The viewing distance was 80 cm, and 3.7 characters subtended 1°of visual angle.The resolution of the eye tracker was 10 min of arc, and the sampling rate was every millisecond.
Participants were instructed to understand the sentences to the best of their ability.A bite bar and head restraint were used to minimize head movements.The accuracy of the eye tracker was checked (and recalibrated when necessary) before each trial.After each sentence, participants pressed a button box to continue and to respond yes or no to comprehension questions.The entire experiment lasted approximately 30 min.

Analyses
Fixations shorter than 80 ms that were within 1 character of the next or previous fixation were incorporated into that fixation.Any remaining fixations shorter than 80 ms or longer than 1,200 ms were discarded; 3.7% of trials were excluded due to either no first-pass fixations on the sentence prior to word n Ϫ 1 or tracker loss or blinks on first-pass reading of word n Ϫ 1 or the critical word.

Results
The influence of orthographic familiarity and word frequency on saccade programming was assessed by examining fixation durations on, and the probability of skipping, the critical word.The duration of the first fixation, single fixation durations, gaze duration (the sum of fixations on a word before leaving it), and total time (the sum of all fixations within a word) were calculated.Parafoveal-on-foveal effects were assessed by examining the duration of the fixation directly before fixation of the critical word.In addition, analyses of saccade targeting to the critical word and the effect of the critical word on interword regressions are reported.
Repeated measures analyses of variance (ANOVAs) were undertaken across the three conditions for both participants' (F 1 ) and items' (F 2 ) means.Note that the items are the sentences, such that the sentence beginnings were identical for each item, and the three word conditions were manipulated using a repeated measures design.Such a design not only provides control over sentence context across the conditions, it also enables the items' statistical analyses to be undertaken with the more powerful repeated measures procedures. 5For cases in which the ANOVAs showed a main effect across the three conditions, paired-samples t tests were undertaken.Comparisons between the frequent and infrequent orthographically familiar conditions were used to test for an effect of word frequency.Comparisons between the orthographically familiar and unfamiliar infrequent conditions were used to test for an effect of orthographic familiarity.The mean error rate on the comprehension questions was 4%, indicating that participants properly read and understood the sentences; trials were included regardless of question responses.

Parafoveal-on-Foveal Effects
Table 2 shows mean reading times on word n Ϫ 1 and fixation durations prior to fixating the critical word.There were no effects of the condition of the critical word on first fixation durations, single fixation durations, or gaze durations on word n Ϫ 1 (Fs Ͻ 1).Because of acuity limitations, the characteristics of a parafoveal word are only likely to influence processing on previous fixations for saccades launched from near launch sites (Kennison & Clifton, 1995;Lavigne, Vitu, & d'Ydewalle, 2000;Rayner, 1975;Rayner, Binder, Ashby, & Pollatsek, 2001;White & Liversedge, 2006b).Therefore, influences of the critical word on the fixation duration prior to fixation of it were calculated for all of the data and for only the 88% of cases in which saccades were launched from word n Ϫ 1.
There was a main effect of condition on the duration of the fixation prior to fixation of the critical word for all data, F 1 (2, 58) ϭ 3.24, MSE ϭ 174, p Ͻ .05,partial 2 ϭ .101;F 2 (2, 76) ϭ 3.19, MSE ϭ 148, p Ͻ .05,partial 2 ϭ .077,and for saccades launched from word n Ϫ 1, F 1 (2, 58) ϭ 4.55, MSE ϭ 162, p Ͻ .05, 3 The effects of linguistic variables on word skipping are often quite small.Therefore, to increase the power of the analyses, each participant was shown all three versions of each experimental item.Consequently, the repeated sentence beginnings were specifically designed to be very bland, and counterbalancing procedures ensured that the repeated items were spaced throughout the stimuli lists.The order in which the participants saw the different conditions for each item was also counterbalanced across the three lists.After the experiment, participants were asked if they noticed any kind of repetition between the sentences, and none of them did. 4The measures were also calculated separately for participants for whom the left and right eyes were recorded; the two groups showed the same patterns of results. 5There is an issue about whether within-item or between-items designs should be adopted when the critical words are different across conditions.A between-items design is used when there are no controls for matching other variables across the conditions.In contrast, when an experiment has a withinitem design, the different items are matched across the conditions-for example, by using the same sentence frame and matching for other variables such as length of the critical words.In the present study, the overall pattern of results was similar when the items analyses were undertaken with between-items tests.For example, for orthographically familiar words, reading times on the critical word were significantly longer for infrequent compared with frequent words for first fixation durations, t 2 (76) ϭ 5.55, p Ͻ .001;single fixation durations, t 2 (76) ϭ 6.14, p Ͻ .001;gaze durations, t 2 (76) ϭ 7.30, p Ͻ .001;and total time, t 2 (76) ϭ 6.24, p Ͻ .001.Other analyses (such as the main effects of condition on word skipping, F 2 (2, 114) ϭ 2.23, MSE ϭ 0.014, p ϭ .112,partial 2 ϭ .038,and fixation durations prior to fixating the critical word launched from word n Ϫ 1, F 2 (2, 114) ϭ 2.40, MSE ϭ 227, p ϭ .095,partial 2 ϭ .04)did not reach significance for the between-items analyses, though the same patterns of effects as for the within-item tests clearly held.Overall, the between-items analyses have larger p values because of the less powerful design (i.e., because between-items analyses do not take account of the controls within the repeated measures design).These analyses highlight the importance of adopting careful matching procedures in studies of word recognition in natural reading.partial 2 ϭ .136;F 2 (2, 76) ϭ 3.13, MSE ϭ 174, p Ͻ .05,partial 2 ϭ .076.For infrequent critical words, there was a small (6-ms) parafoveal-on-foveal effect such that prior fixation durations were longer when the critical word was orthographically unfamiliar compared with when it was orthographically familiar, and these effects were significant for saccades launched from word n Ϫ 1, t 1 (29) ϭ 2.37, p Ͻ .05;t 2 (38) ϭ 2.49, p Ͻ .05,but significant only across items, t 2 (38) ϭ 2.33, p Ͻ .05,and not participants, t 1 (29) ϭ 1.85, p ϭ .075,for all of the data.In contrast, for both sets of analyses, for orthographically familiar words there was no difference in prior fixation duration between the frequent and infrequent conditions (ts Ͻ 1). 6 It has been suggested that parafoveal-on-foveal effects can sometimes occur as a result of inaccurate saccade targeting (Drieghe et al., in press;Rayner, Warren, et al., 2004;Rayner, White, et al., 2003).However, if the parafoveal-on-foveal effects reported above had arisen because of mistargeting of saccades, there perhaps also should have been a parafoveal-on-foveal effect of word frequency.The following analysis is aimed at identifying whether parafoveal-on-foveal effects occurred when cases in which there were most likely to have been mistargeted saccades were eliminated.Saccades that were intended to land on the critical word but that undershot it were most likely to have landed on the three final characters of word n Ϫ 1.For fixations launched from word n Ϫ 1, but not from the three final characters of word n Ϫ 1, the main effect of condition was marginal, F 1 (2, 58) ϭ 2.84, MSE ϭ 223, p ϭ .067,partial 2 ϭ .089;F 2 (2, 76) ϭ 2.70, MSE ϭ 266, p ϭ .074,partial 2 ϭ .066.Although there was no effect of parafoveal word frequency on these fixation durations (ts Ͻ 1), the results indicate that prior fixation durations in this region were longer prior to orthographically unfamiliar compared with orthographically familiar infrequent words, t 1 (29) ϭ 2.22, p Ͻ .05;t 2 (38) ϭ 2.74, p Ͻ .01.
Together, the absence of a parafoveal-on-foveal effect of word frequency and the indication of a parafoveal-on-foveal effect of orthography more than three characters from the critical words suggest that the orthographic parafoveal-on-foveal effects shown here may not simply be explained by inaccurate saccade targeting (see Starr & Inhoff, 2004, for a similar finding).Importantly, the present study may have provided very favorable conditions for parafoveal-on-foveal effects to arise.That is, both the foveal (word n Ϫ 1) and parafoveal (critical) words were short, and the orthographic familiarity of the entire parafoveal word was strongly manipulated.(Note, however, that it could be that only the famil-iarity of some of the letters, such as those at the word beginning, could have produced the effect.)Consequently, the parafoveal-onfoveal effects shown here may not hold under less favorable conditions-for example, when there are long parafoveal words or only the orthographic familiarity of the word beginning is manipulated (White & Liversedge, 2004, 2006b).Critically, despite the favorable conditions, and consistent with prior research (Rayner et al., 1998), there was no effect of word frequency on prior fixation durations for orthographically familiar words. 7

Reading Measures for the Critical Word
The mean reading times, refixation probabilities, and spillover fixation durations for the critical word are shown in Table 3.There were significant main effects of condition for all of the reading measures on the critical word (Fs Ͼ 15, ps Ͻ .001).For orthographically familiar words, reading times for the critical word were shorter on frequent than on infrequent words for first fixation durations (t 1 (29) ϭ 8.96, p Ͻ .001;t 2 (38) ϭ 5.34, p Ͻ .001),single fixation durations (t 1 (29) ϭ 8.01, p Ͻ .001;t 2 (38) ϭ 6.01, p Ͻ 6 In an additional analysis, prior fixations were divided into those with near (M ϭ 3.0, SD ϭ 1.4) and far (M ϭ 6.7, SD ϭ 3) launch sites prior to fixation of the critical word for each participant.There was no difference between the frequent and infrequent conditions for either near (frequent: M ϭ 237, SD ϭ 58; infrequent: M ϭ 238, SD ϭ 64) or far (frequent: M ϭ 231, SD ϭ 64; infrequent: M ϭ 231, SD ϭ 61) launch sites.Similar to the data for saccades launched from word n Ϫ 1, prior-fixation durations were numerically longer prior to orthographically unfamiliar (near: M ϭ 242, SD ϭ 65; far: M ϭ 238, SD ϭ 63) than orthographically familiar (near: M ϭ 238, SD ϭ 64; far: M ϭ 231, SD ϭ 61) words, but none of the effects were statistically reliable.There was also no significant effect of condition on prior fixation durations launched three or fewer characters from the critical word (Fs Ͻ 1.43, ps Ͼ .24).
The distribution of single fixation durations on the critical word is shown in Figure 1.Note that the longer fixation durations in the infrequent compared with the frequent word conditions are characterized by a rightward shift in the distribution, consistent with previous studies (Rayner, 1995;Rayner, Liversedge, et al., 2003).Numerical differences in the distributions between the frequent and infrequent orthographically familiar conditions occur no earlier than differences in the distributions between the orthographically familiar and unfamiliar infrequent conditions.These numerical patterns indicate that the effect of word frequency during single fixation durations occurs no later than that of orthographic familiarity.
Overall, the results show that word frequency has a robust and long-lasting effect on word processing.In contrast, orthographic familiarity has a numerically smaller influence for all of the reading time measures on the critical word.The absence of significant orthographic familiarity effects in later measures such as spillover indicates that orthographic familiarity exerts a relatively small and short-lived influence on word processing compared with effects of word frequency.
Compared with the results for saccades launched from word n Ϫ 1, a similar pattern of effects held for saccades launched from three or fewer characters from the critical word.There was  38) ϭ 1.31, p ϭ .20.These results show that word frequency, but not orthographic familiarity, significantly influences saccade programming such that this determines whether words are skipped. 8Critically, the effect of word frequency on word skipping indicates that the lexical characteristics of words are processed in the parafovea and that they can have an early influence on word processing and eye movement behavior.Furthermore, the fact that orthographic familiarity produces a parafoveal-on-foveal effect, but does not influence word skipping, indicates that although these two measures both reflect early parafoveal processing of words, they may be determined by qualitatively different aspects of eye movement control.

Saccade Metrics for the Initial First Pass Fixation on the Critical Word
Mean landing positions, launch positions, and saccade lengths for first-pass saccades launched from word n Ϫ 1 to the critical word are shown in Table 4.There were no effects of condition on landing positions (Fs Ͻ 1); launch positions, F 1 (2, 58) ϭ 2.25, MSE ϭ 0.174, p ϭ .114,partial 2 ϭ .072;F 2 (2, 76) ϭ 1.87, MSE ϭ 0.095, p ϭ .162,partial 2 ϭ .047;or saccade lengths, F 1 (2, 58) ϭ 1.88, MSE ϭ 0.16, p ϭ .162,partial 2 ϭ .061;F 2 (2, 76) ϭ 2.25, MSE ϭ 0.09, p ϭ .113,partial 2 ϭ .056.These findings suggest that neither the orthographic familiarity nor the word frequency of four-and five-letter words influences saccade targeting to words.These results contrast with evidence showing that orthographic familiarity does influence where longer words are first fixated (Hyo ¨na ¨, 1995;Radach et al., 2004;Vonk et al. 2000;White & Liversedge, 2004, 2006a, 2006b).However, note that initial trigrams used in experiments with longer words were often much more infrequent than those used here.Therefore, it is possible that the orthographic familiarity of shorter words may influence saccade targeting with much stronger manipulations of orthographic familiarity.It is also possible that orthographic influences on saccade targeting have no, or much smaller, effects for shorter words.
For cases in which the critical word was fixated on first pass, there was no effect of condition on the probability of making a regression into the critical word, F 1 (2, 58) ϭ 2.16, MSE ϭ 0.004, p ϭ .124,partial 2 ϭ .069;F 2 (2, 76) ϭ 1.65, MSE ϭ 0.006, p ϭ .199,partial 2 ϭ .042.In contrast, for cases in which the critical word was skipped on first pass, there was a main effect of condition on the proportion of regressions made into the critical word, F 1 (2, 54) ϭ 9.09, MSE ϭ 0.056, p Ͻ .001,partial 2 ϭ .252;F 2 (2, 74) ϭ 10.01, MSE ϭ 0.085, p Ͻ .001,partial 2 ϭ .213.For cases in which the critical word was skipped on first pass, for orthographically familiar words there were more regressions into infrequent than into frequent critical words, t 1 (27) ϭ 2.92, p Ͻ .01;t 2 (37) ϭ 3.5, p Ͻ .01.However, there was no difference in the proportion of regressions made into the critical word for infrequent orthographically unfamiliar compared with familiar words (ts Ͻ 1.1).The finding that word frequency influences the probability of making a regression back to a skipped word supports results reported by Vitu and McConkie (2000).It is particularly interesting that the present findings show that word frequency, but not orthographic familiarity, significantly influenced regressions.The results suggest that the linguistic influence on regressions is at a lexical rather than a sublexical level.The effect of lexical processing difficulty on regressions supports the suggestion made by Engbert et al. (2005) that interword regressions can be caused by incomplete lexical access.However, note that processing difficulty related to postlexical semantic integration may also be related to such an effect of word frequency.The present findings are also consistent with previous research suggesting that regressive saccades can target areas of processing difficulty quite accurately (Frazier & Rayner, 1982;Kennedy, Brooks, Flynn, & Prophet, 2003;Meseguer, Carreiras, & Clifton, 2002).

Discussion
The results clearly show that word frequency, independent of orthographic familiarity, influences the probability of skipping words and fixation durations on words.However, there were no effects of word frequency on prior fixation durations.In contrast, orthographic familiarity had a small effect on prior fixation durations and fixation durations on words but no influence on the probability of word skipping.The key implications are, therefore, that lexical processing of fixated words can influence saccade programming, as shown by fixation durations, and that lexical processing of parafoveal words can influence saccade programming, as shown by word skipping.The results have important implications for models of eye movement control in reading, and these are discussed below.In addition, there was a small orthographic parafoveal-on-foveal effect but no evidence of lexical parafoveal-on-foveal effects.The implications of these findings are considered at the end of this Discussion.

Saccade Programming: Fixation Durations and Word Targeting
The findings reported here demonstrate that models of eye movement control in reading should include a saccadeprogramming mechanism that is sensitive to lexical (or even postlexical) processing of word frequency.Therefore, there must be sufficient time during a fixation for both lexical processing and saccade programming to occur, before the metrics and timing of the subsequent saccade are finalized.The present findings are particularly crucial for models in which linguistic factors modulate when saccade programming is initiated because, due to the time required for saccade programming, such linguistic influences may have to occur early during fixations (Engbert et al., 2002(Engbert et al., , 2005;;Kliegl & Engbert, 2003;Pollatsek et al., 2006;Reichle et al., 1998Reichle et al., , 1999Reichle et al., , 2003Reichle et al., , 2006;;Richter et al., 2006).
In the E-Z reader model (Pollatsek et al., 2006;Reichle et al., 1998Reichle et al., , 1999Reichle et al., , 2003Reichle et al., , 2006)), completion of an initial stage of word processing (L 1 ) triggers the initiation of a saccade program, which subsequently takes approximately 125 ms to complete.Therefore, the time at which this first stage of word processing is completed effectively produces the linguistic influence on the duration of fixations.For a word to be skipped, both the fixated word and the parafoveal word would have to have reached the L 1 stage before the skipping program could be finalized.In the SWIFT model, a random stochastic process influences when saccade programming begins, and this is delayed or cancelled by linguistic processing using a mechanism referred to as foveal inhibition (Engbert et al., 2002(Engbert et al., , 2005;;Kliegl & Engbert, 2003;Richter et al., 2006).Therefore, similar to the E-Z reader model, in the SWIFT model, linguistic influences on fixation durations must also occur prior to saccade programming.Crucially, for such accounts to explain the results reported here, the lexical variable of word frequency must have an influence early during a fixation, before saccade programming commences.For example, for a fixation lasting 250 ms with 125 ms of saccade programming, lexical influences on saccade programming would have to occur in the initial 125 ms of the fixation.Given such an account, either lexical processing must be very fast or a significant amount of this processing may be undertaken on previous fixations (Findlay & White, 2003).
An alternative possibility is that linguistic factors may influence the saccade program later during a fixation but before the nonlabile stage of saccade programming.For example, in the SWIFT model, determination of the saccade target (which word is fixated) occurs at the end of the labile stage of saccade programming.If linguistic processing could influence saccade programming during the labile stage, there would be more time during the fixation for lexical processing to occur such that it could then influence when or where the eyes moved.For example, for a fixation lasting 250 ms with 50 ms for the nonlabile stage of saccade programming, 200 ms of linguistic processing time may be available before the linguistic influence on the saccade program is finalized at the end of the labile stage.
In the E-Z reader model, once the first stage of word processing is complete (L 1 ) and saccade programming has been initiated, a second stage of word processing (L 2 ) is achieved before attention shifts to the following word (Pollatsek et al., 2006;Reichle et al., 1998Reichle et al., , 1999Reichle et al., , 2003Reichle et al., , 2006)).Because saccade-programming time is relatively constant, the duration of L 2 determines the time between the attention shift and when the eyes move, during which the following word can be preprocessed.For the E-Z reader model, the influence of L 2 on the degree of parafoveal processing is crucial to explaining both spillover effects (e.g., Rayner & Duffy, 1986) and modulation of parafoveal processing by foveal load (Henderson & Ferreira, 1990;Kennison & Clifton, 1995;White, Rayner, & Liversedge, 2005).If L 1 is of a lexical nature, this raises the question of exactly how L 2 is different from L 1 .Future models based on multiple word-processing stages may need to specify the types of processing entailed in each stage more precisely.For a more detailed discussion of how even postlexical factors may possibly influence the first stage of word processing, see Reichle, Pollatsek, and Rayner (2007).
The present study also raises the issue of precisely how the orthographic familiarity of words influences fixation durations on those words.In the E-Z reader model (Pollatsek et al., 2006;Reichle et al., 1998Reichle et al., , 1999Reichle et al., , 2003Reichle et al., , 2006)), it could be argued that orthographic familiarity may influence the initial stage of word processing (L 1 ), which influences the current fixation duration, but not the second stage of word processing (L 2 ), which can influence subsequent fixation durations.Such a suggestion would be consistent with the finding that orthographic familiarity had a short-lived effect on reading times on the critical word in the present study (for similar reasoning, see Reingold, 2003;Reingold & Rayner, 2006).
The results also showed that the lexical characteristics of words influenced the probability of making a regression back to words if they were skipped on first pass.In line with these findings, the SWIFT model of eye movement control is unique in that it predicts that regressions are directed to words that are not completely processed on first pass (Engbert et al., 2002(Engbert et al., , 2005;;Kliegl & Engbert, 2003;Richter et al., 2006; see also Reilly & Radach, 2003, 2006).It seems especially noteworthy that orthographic familiarity did not significantly influence word skipping or regression probabilities, while word frequency did influence these measures.These results suggest that lexical factors have a more critical role than sublexical factors in determining which words are fixated and refixated.However, note that although lexical factors have been shown here to influence both when and where the eyes move, research suggests that these two processes may in fact be controlled in qualitatively different ways.For example, while fixation durations are influenced by both foveal and parafoveal linguistic information, word skipping may be influenced by linguistic information only in the parafovea (White, 2007; see also Drieghe, Rayner, & Pollatsek, 2005).

Parafoveal-on-Foveal Effects
The present study also has important implications for the contentious issue of parafoveal-on-foveal effects.The results show that, at least when both foveal and parafoveal words are short and the orthographic familiarity of the entire parafoveal word is manipulated, the orthographic characteristics of the critical word have a small (6-ms) influence on prior fixation durations.These results are in line with previous sentence reading studies showing parafoveal-on-foveal effects at least at the level of orthographic familiarity (Inhoff, Starr, & Shindler, 2000;Pynte et al., 2004;Rayner, 1975;Starr & Inhoff, 2004;Underwood et al., 2000).It could be that only the orthographic familiarity of the initial letters of the word is critical.Furthermore, other characteristics associated with orthographic familiarity, such as the informativeness of the component letter sequences, may have driven the effects.Fixations are often mislocated (McConkie et al., 1988;Nuthmann et al., 2005), and some parafoveal-on-foveal effects may be explained by such mislocated fixations (Drieghe et al., 2005(Drieghe et al., , 2007;;Rayner, Warren, et al., 2004;Rayner, White, et al., 2003).However, the present results indicate that the orthographic parafovealon-foveal effects shown here are not simply isolated to fixations three or fewer characters from the critical word.Therefore, either a substantial proportion of mislocated fixations must land more than three characters from the critical word or, as concluded by Starr and Inhoff (2004), there are parafoveal-on-foveal effects of orthography that are not caused by mislocated fixations.
Given that the explanation of mislocated fixations does not seem feasible in this case, there are two alternative accounts for the small orthographic parafoveal-on-foveal effects shown here.First, attention may be allocated to multiple words in parallel such that processing of the foveal word may occur simultaneously with processing of the orthographic characteristics of the parafoveal word (Engbert et al., 2002(Engbert et al., , 2005;;Kennedy, 2000;Kliegl & Engbert, 2003;McDonald et al., 2005;Reilly & Radach, 2003, 2006;Richter et al., 2006).Second, attention may be allocated to words one at a time (as in serial models), but the orthographic characteristics of the parafoveal words may be concurrently processed in a manner that does not require attention.Such early preattentive visual processing of the parafoveal word may affect processing of the foveal word to generate orthographic parafoveal-on-foveal effects (Pollatsek et al., 2006;Reichle et al., 2003Reichle et al., , 2006)).Note that in the E-Z reader model, preattentive processes that may induce parafoveal-on-foveal effects are quite separate from the linguistic processes associated with the L 1 and L 2 stages of word recognition, which have a more primary role in influencing saccade programming during reading.
To be clear, the parafoveal-on-foveal effects shown here were not necessarily mediated by the same attention-based processes that are generally allocated for reading of words.That is, parafoveal-on-foveal effects may not necessarily affect the reading process (see Drieghe et al., in press).Therefore, although there may be temporal overlap between processing of the foveal word and processing related to the orthographic characteristics of the parafoveal word, these separate processes may be independent and perhaps even qualitatively different.For example, the orthographic parafoveal-on-foveal effects may be mediated by visual, not linguistic, processes.Overall, the orthographic parafoveal-on-foveal effects reported here cannot distinguish between the parallel attention (Engbert et al., 2002(Engbert et al., , 2005;;Kennedy, 2000;Kliegl & Engbert, 2003;McDonald et al., 2005;Reilly & Radach, 2003, 2006;Richter et al., 2006) and preattentive (Pollatsek et al., 2006;Reichle et al., 2003Reichle et al., , 2006) ) explanations provided by the models.
The present study revealed no effect of word frequency on prior fixation durations.This was despite the study providing favorable conditions for lexical parafoveal-on-foveal effects to occur by using short foveal and parafoveal words.Note that the foveal words n Ϫ 1 were frequent words (see the Method section), so there should have been sufficient time or processing resources to process the parafoveal word (Henderson & Ferreira, 1990;Kennison & Clifton, 1995;White et al., 2005).The results are in line with previous experimental studies showing no evidence of lexical parafoveal-on-foveal effects (Altarriba et al., 2001;Calvo & Meseguer, 2002;Henderson & Ferreira, 1993;Inhoff, Starr, & Shindler, 2000;Rayner, 1975;Rayner et al., 1998;Rayner et al., 2007;Schroyens et al., 1999), but they contrast with studies that have shown lexical parafoveal-on-foveal effects using very large data sets across a corpus of data or a combination of different studies

Appendix Experimental Sentences
The critical word is shown in italics.Sentences a, b, and c refer to the frequent and orthographically familiar, infrequent and orthographically familiar, and infrequent and orthographically unfamiliar conditions, respectively.1a.He loved to visit the local town near to where his grandparents lived.
1b.He loved to visit the local cove near to where he learnt to swim.
1c.He loved to visit the local quay near to his father's fish shop.
2a.He thought the awful party was a waste of time.
2b.He thought the awful lager was really not good enough.
2c.He thought the awful vinyl was a lot worse than the old carpet.
3a.He tried to lift the heavy door onto its hinges but he needed more help.
3b.He tried to lift the heavy gong onto the table without it making a noise.
3c.He tried to lift the heavy tusk onto the truck but it was just too awkward.
4a.She thought the dusty glass might be very valuable.
4b.She thought the dusty cello might need re-tuning.
4c.She thought the dusty yucca might need watering.
5a.He saw the famous place in the city where the pop star was born.
5b.He saw the famous adder in the reptile section of the zoo.
5c.He saw the famous crypt in the old church.
6a.The photograph showed the young child sitting on top of the climbing frame.
6b.The photograph showed the young heron sitting on its nest.
6c.The photograph showed the young koala sitting in the eucalyptus tree.
7a.She liked the basic home because it was practical and well designed.
7b.She liked the basic loom because she could easily weave beautiful fabrics.
7c.She liked the basic yoga because it provided good but gentle exercise.
8a.She admired the unique city before she found out about the drug problem.
8b.She admired the unique dart before she aimed it at the dartboard.
8c.She admired the unique oboe before she began to play her favourite composition.
9a. Eventually the strong wife managed to move the concrete slab.
9b. Eventually the strong lout managed to push his way to the bar.
9c. Eventually the strong oxen managed to finish ploughing the field.
10a.She laughed at the funny sound that was coming from the radiator.
10b.She laughed at the funny hound that was bounding across the field.
10c.She laughed at the funny scowl that the child was making.
11a.She knew that the modern chair was perfect for her bedroom.
11b.She knew that the modern grate was of very poor quality.
11c.She knew that the modern kiosk was not suitable for the business.
12a.She looked at the awful fire which was spreading across the forest.
12b.She looked at the awful lice which were causing so many problems.
12c.She looked at the awful acne which she desperately wanted to be better.
13a.He used the decent table for the dinner party at his house.
13b.He used the decent plank for the building work on the garage.
13c.He used the decent opium for the very last time.

221
WORD FREQUENCY AND ORTHOGRAPHY (Appendix continues) 14a.He knew that the cheap paper might not be acceptable for the office.
14b.He knew that the cheap crate might not be strong enough to hold the bottles.
14c.He knew that the cheap khaki might not be quite the right colour.
15a.She chose the normal bank for the new business account.
15b.She chose the normal sari for her daughter to wear for the visit.
15c.She chose the normal kiln for the pots to be fired in.
16a.She liked the pretty hair that her sister had always had.
16b.She liked the pretty glen that she saw with her family in Scotland.
16c.She liked the pretty loch that was so quiet and peaceful.
17a.He gave the spare money to his brother after school.
17b.He gave the spare cress to his mother for the recipe.
17c.He gave the spare myrrh to the chemist at the university.
18a.She wanted the clean room ready for when her parents visited.
18b.She wanted the clean lint ready for the first aid session.
18c.She wanted the clean toga ready for the big party at the start of term.
19a.She liked the clever idea despite the expensive cost of the project.
19b.She liked the clever dame despite the annoying high pitched voice.
19c.She liked the clever guru despite his rather extreme beliefs.
20a.She worried about the major case that she had been asked to work on.
20b.She worried about the major ford that she would have to cross tomorrow.
20c.She worried about the major feud that had arisen within the family.
21a.She hated the awful view which she looked out on from her window.
21b.She hated the awful lisp which she had had since she was a child.
21c.She hated the awful levy which the society had decided to charge.
22a.He needed some normal water for the special mixture.
22b.He needed some normal mince for the special meal that he was cooking.
22c.He needed some normal gauze for the machine he was constructing.
23a.He disliked the boring woman despite her admirable achievements.
23b.He disliked the boring chime despite his love of clocks.
23c.He disliked the boring polka despite the fact that he was winning.
24a.The doctor looked at the unique blood under the microscope.
24b.The doctor looked at the unique finch under the old tree.
24c.The doctor looked at the unique algae under the water.
25a.She hated the yellow house which belonged to the dentist.
25b.She hated the yellow froth which clung to the edge of the bowl.
25c.She hated the yellow fudge which her grandmother gave her to eat.
26a.She admired the great work that the charity had done for the homeless.
26b.She admired the great lark that sang so beautifully in the tree.
26c.She admired the great judo that she saw the children doing in the arena.
27a.He planned the entire story before he spoke to the publisher.
27b.He planned the entire prank before he told his friends what to do.
27c.He planned the entire haiku before he wrote it for his girlfriend.
28a.He was impressed by the great game that everyone wanted to play.
28b.He was impressed by the great solo that the singer had performed.
28c.He was impressed by the great hoax that they had managed to pull off.29b.He examined the clean quill before he began to write.
29c.He examined the clean anvil before the smith used it to shape the metal.
30a.He wanted the quiet night to last forever because it was so peaceful.
30b.He wanted the quiet tramp to move away from the shop entrance.
30c.He wanted the quiet hyena to come back so that he could see him again.
31a.He thought the simple light was perfect for the office.
31b.He thought the simple plait was very appropriate on the little girl.
31c.He thought the simple humus was a bit boring for the special sandwiches.
32a.He cleaned the dirty food before giving it to the animals.
32b.He cleaned the dirty leek before he chopped it up for the stew.
32c.He cleaned the dirty ruby before he took it to the jewellers.
33a.She knew that the modern road would be safest in the winter conditions.
33b.She knew that the modern cork would keep the wine in good condition.
33c.She knew that the modern tyre would last a long time.
34a.She found the small book under the bed in her room.
34b.She found the small dice under the board game and cards.
34c.She found the small kiwi under the oranges and apples in the fridge.
35a.He watched the quiet class whilst they worked on the project.
35b.He watched the quiet panda whilst he hid amongst the bamboo bushes.
35c.He watched the quiet waltz whilst he waited at the dance hall.
36a.He knew that the single girl would not want his phone number.
36b.He knew that the single pear would be good for the fruit crumble.
36c.He knew that the single buoy would be positioned near the stranded ship.
37a.She wanted a decent world for the poor people in the developing countries.
37b.She wanted a decent puree for the fantastic meal she was trying to prepare.
37c.She wanted a decent dowry for herself and her husband after the marriage.
38a.She used the fresh fish for the family dinner.
38b.She used the fresh lard for frying the meatballs.
38c.She used the fresh suet for making the dumplings.
39a.He examined the small group using a basic written test for each child.
39b.He examined the small louse using a special microscope.
39c.He examined the small ulcer using key-hole surgery.

Figure 1 .
Figure 1.The distribution of single fixation durations on the critical word for each of the three experimental conditions (35-ms bins).
examined the clean stage before the performance.

Table 1 Non
-Position-Specific (NPS) and Position-Specific (PS) Summed Token n-Gram Frequency Counts (per Million) for Each of the Conditions, With Differences in Mean Frequency Counts for the Orthographically Familiar and Infrequent Conditions * p Ͻ .05. ** p Ͻ .01. *** p Ͻ .001.

Table 3
Mean Reading Times on the Critical Word for Each Condition, Probabilities of Refixating the Critical Word on First Pass, and Mean Duration of the Spillover Fixation Except for the probabilities, all values represent milliseconds.Standard deviations are shown in parentheses.

Table 4
Probabilities of Skipping the Critical Word for All Data, for Only Saccades Launched From Word n Ϫ 1, and for Only Saccades Launched From Three or Fewer Characters From the Critical Word; Mean Landing Positions, Launch Positions, and Saccade Lengths for the Critical Word; and Probabilities of First-Pass Regressions Out of the Critical Word and Into the Critical Word When the Critical Word Was Fixated (Fix n) and Skipped (Skip n) on First Pass Except as indicated, all values represent probabilities.Standard deviations are shown in parentheses.