Fixation Durations Before and After Word Skipping: Evidence From Eye Movements

1. Glasgow Language Processing Fixation Durations Before and After Word Skipping: Evidence From Eye Movements Christopher J. Hand, Patrick J. O’Donnell, and Sara C. Sereno Department of Psychology, University of Glasgow Corresponding author: c.hand@psy.gla.ac.uk • INTRODUCTION • During normal reading, words are often skipped, thus receiving no foveal processing. • Word Length • Word length massively influences the likelihood of skipping a word. 2-3 letter words are skipped around 75% of the time, whereas words of 8 letters or longer are almost never skipped. • Contextual Predictability • Predictable words are skipped more often than unpredictable words. • Word Frequency • The effect of frequency on word skipping is unclear. Word frequency and word length are highly correlated in natural text – short words tend to be much more frequent than longer words. • Some evidence has been found to suggest an effect of frequency on word skipping (Rayner, Sereno, & Raney, 1996), but only when the eyes were fixated close to the beginning of the target word on prior fixation: a word tends to be skipped when the eyes are fairly close to that word on prior fixation. • Models of eye movement control in reading • Three broad categories • Sequential attentional shift (SAS) models • Guided by attentional gradient (GAG) models • Primary oculomotor control (POC) models • These models can be distinguished by the relative weighting of oculomotor and linguistic factors and how the interactions between these factors are conceptualised and implemented (Kliegl & Engbert, 2005). • All three types of model predict that short, high-frequency, and highly predictable words will be skipped more often than longer, lower-frequency, and lower predictability words, and this has been supported by experimental evidence (Brysbaert & Vitu, 1998). • One fundamental difference between the predictions of these models concerns fixation durations before and after skipping a word. • Presently, there remains controversy as to whether the fixation duration prior to a skip is inflated. • Some evidence suggests inflated fixation duration prior to a skip (Pollatsek, Rayner, & Balota, 1986). • Fixations prior to skipping low-frequency words were longer than those before skipping higher-frequency words (Kliegl & Engbert, 2005). • Others find no effect of skipping on fixation duration prior to a skip (McConkie, Kerr, & Dyre, 1994; Radach & Heller, 2000) • Two processes predict increased fixation duration prior to a skip: • Saccade cancellation • Parafoveal pre-processing • SAS models • e.g., E-Z Reader Model of eye movement control in reading (Reichle, Pollatsek, Fisher, & Rayner, 1998) • Word n+1 is the default target of saccade from word n • Skipping to word n+2 should incur a cost in terms of fixation duration on word n • Saccade must be cancelled and reprogrammed • Such values have been obtained (Reichle, Pollatesk, & Rayner, 2003) • E-Z Reader predicts that the cost associated with the cancellation of a saccade should diminish as the frequency of word n+1 increases • GAG models • e.g., Saccade-generation With Inhibition by Foveal Targets (SWIFT; Engbert et al., 2005) • Unlike SAS models, GAG models allow for parallel processing of words. • Longer fixations prior to skips imply a longer accumulation of parafoveal information • Parafoveal pre-processing increases the probability of complete lexical access – thus, the skipping of word n+1 (Engbert et al., 2005; Reilly & Radach, 2003) • GAG models view longer fixations as a cause of skipping rather than a consequence • POC models • Assume that saccades are generated from a distribution of saccade amplitudes adjustable dependent on the difficulty of the text (McConkie, Kerr, Dyre, 1994) • Do not predict a strong modulation of fixations before skipping • The present study was carried out to examine whether fixation durations prior to skips were inflated compared to when target words were fixated. Additionally, the frequency and predictability of target words were orthogonally varied in order to establish the extent to which the lexical attributes of target words might influence fixation durations before and after target word skipping. RESULTS HF-PHF-ULF-PLF-U Before Skip 278 256 252 264 No Skip 251 260 257 262 After Skip 249 262 264 268 No Skip 269 280 274 272 Main effect of skipping The results of the analyses revealed that there was no significant effect of word skipping on fixation durations before skipping the target word F1(1,63)= 1.45, p>0.20; F2(1,43)= 1.88, p>0.15. Fixations before the target word There was a significant three-way interaction between fixation location (i.e., before vs. after the target word), skipping condition (i.e., skip vs. no skip), and target word frequency on fixation durations before the target word. F1(1,63)= 6.80, p<0.05; F2(1,43)= 9.86, p<0.01. Planned follow-up comparisons 11ms effect of skipping on fixation durations before HF words was significant F1(1,63)= 5.69, p<0.05; F2(1,43)= 2.23, p=0.14; see Figure 1 No significant effect of skipping on fixation durations before LF words both Fs<1; see Figure 1. There was a significant three-way interaction between fixation location, skipping condition, and target word predictability on fixation durations before the target word. F1(1,63)= 3.79, p=0.5; F2(1,43)= 11.7, p<0.01. Planned follow-up comparisons 10ms effect of skipping on fixation durations before P words was significant F1(1,63)= 7.28, p<0.01; F2(1,43)= 3.09, p=0.08; see Figure 2. No significant effect of skipping on fixation durations before LF words both Fs<1; see Figure 2. Figure 1 Figure 2 Fixations after the target word HF target word 19ms shorter when target skipped than when the target word had been fixated F1(1,63)= 16.0, p<0.001; F2(1,43)= 13.1, p<0.001.; see Figure 3. LF target word 7ms shorter when target skipped than when the target word had been fixated F1(1,63)= 2.23, p=0.14; F2(1,43) <1; see Figure 3. P target word 14ms shorter when target skipped than when the target word had been fixated F1(1,63)= 13.0, p<0.001; F2(1,43)= 13.4, p<0.001.; see Figure 4. U target word 11 ms shorter when target skipped than when the target word had been fixated F1(1,63)= 7.43, p<0.01; F2(1,43) <1; see Figure 4. Figure 3 Figure 4 • METHOD • Participants64 native English speakers, normal vision, not diagnosed as dyslexic. • ApparatusFourward Technologies Generation 5.5 dual-Purkinje eyetracker. • ProcedureParticipants read short passages of text (comprising 2 sentences) while their eye movements were monitored. Y/N comprehension questions were presented on half the trials. • Materials and Design • 4 conditions: Frequency (HF,LF) X Predictability (P,U) • 88 experimental passages => 22 items per condition • Target selection: HF & LF targets matched pair-wise on word length (5-8 letters). • Passage construction: 2 possible targets per passage (HF-P | LF-U or HF-U | LF-P). Passages comprised two lines of text, up to 60 characters per line. • Word rating task: Predictability of targets in passages judged (1=low to 7=high). • Cloze probability task: Passages up to (but not including) target presented. Participants generated next word in each passage. • HF-PHF-ULF-PLF-U • Frequency (BNC, per million) 145 145 4 4 • Predictability (1-7) 6.19 4.07 6.11 3.69 • Cloze probability 0.57 0.02 0.50 0.01 • ConditionExample Item • HF-P | LF-U None of the baker's plans for the wedding cake had satisfied [lb] • the bride. He had completely run out of ideas | yeast and was irate. • LF-P | HF-U Amy's bread dough for the dinner wouldn't rise and the shops [lb] • were now closed. She had run out of yeast | ideas and was irate. DISCUSSION The non-significant main effect of skipping is misleading. Skipping differentially affects fixation durations before and after target words, dependent on the frequency and predictability of that word. The presence of inflated fixation durations prior to skips is consistent with both SAS and GAG models of eye movement control in reading. However, the pattern of effects observed in the present study are inconsistent with the predictions of E-Z Reader. Previous research has demonstrated that the extraction of information parafoveally is influenced by the frequency and predictability of that parafoveal word (Balota, Pollatsek, & Rayner, 1985; Inhoff & Rayner, 1986). The differential effects of word frequency and predictability on fixation durations prior to skips may be more consistent with GAG models of eye movement control such as SWIFT, which posit that longer parafoveal processing during the prior fixation causes inflated fixation durations prior to skips. REFERENCES Balota, D.A., Pollatsek, R., & Rayner, K. (1985). The interaction of contextual constraints and parafoveal visual information in reading. Cognitive Psychology, 17, 364-390. British National Corpus (1995). Available on-line via the British National Corpus world-wide web site: http://www.natcorp.ox.ac.uk. Brysbaert, M., & Vitu, F. (1998). Word skipping: Implications for theories of eye movement control during reading. In G. Underwood (Ed.), Eye guidance in reading and scene perception (pp. 125-147). Oxford: Elsevier. Engbert, R., Nuthmann, A., Richter, E.M, & Kliegl, R. (2005). SWIFT: A dynamical model of saccade generation during reading. Psychological Review, 112, 777-813. Inhoff, A.W., & Rayner, K. (1986). Parafoveal word processing during eye fixations in reading: Effects of word frequency. Perception & Psychophysics, 40, 431-439. Kliegl, R., & Engbert, R. (2005). Fixation durations before word skipping in reading. Psychonomic Bulletin & Review, 12, 132-138. McConkie, G.W., Kerr, P.W., & Dyre, B.P. (1994). What are “normal” eye movements during reading: Toward a mathematical description. In J. Ygge & G. Lennestrand (Eds.) Eye movements in reading (pp. 315-327). Oxford: Pergammon. Radach, R., & Heller, D. (2000). Relations between spatial and temporal aspects of eye movement control. In A. Kennedy, R. Radach, D. Heller, & J. Pynte (Eds.) Reading as a perceptual process (pp. 165-192). Oxford: Elsevier. Rayner, K., Sereno, S.C., & Raney, G.E. (1996). Eye movement control in reading: A comparison of two types of models. Journal of Experimental Psychology: Human Perception and Performance, 22, 1188-1200. Reichle, E.D, Pollatsek, A., Fisher, D.L., & Rayner, K (1998). Toward a model of eye movement control in reading. Psychological Review, 105, 125-157. Reichle, E.D., Rayner, K., & Pollatsek, A. (2003). The E-Z Reader model of eye movement control in reading: Comparisons to other models. Behavioural & Brain Sciences, 26, 445-526. This work has been performed as part fulfilment for a PhD for Christopher J. Hand, supported by the ESRC.