Eye movements and attention in reading, scene perception, and visual search

Abstract

Eye movements are now widely used to investigate cognitive processes during reading, scene perception, and visual search. In this article, research on the following topics is reviewed with respect to reading: (a) the perceptual span (or span of effective vision), (b) preview benefit, (c) eye movement control, and (d) models of eye movements. Related issues with respect to eye movements during scene perception and visual search are also reviewed. It is argued that research on eye movements during reading has been somewhat advanced over research on eye movements in scene perception and visual search and that some of the paradigms developed to study reading should be more widely adopted in the study of scene perception and visual search. Research dealing with “real-world” tasks and research utilizing the visual-world paradigm are also briefly discussed.

Keywords:

• Eye movements
• Attention
• Reading
• Scene perception
• Visual search

Acknowledgments

Preparation of this article was supported by Grant HD26765 from the National Institute of Health. Thanks to Jane Ashby, Chuck Clifton, Denis Drieghe, Harold Greene, John Henderson, Barbara Juhasz, Simon Liversedge, Alexander Pollatsek, and Adrian Staub for helpful comments on a prior draft.

Notes

¹ While across a large passage of text, the correlation between fixation duration and saccade length tends to be close to zero (Rayner & McConkie, 1976), it is the case that for certain segments of text one can find reasonable-sized correlations. Thus, when the text is difficult, readers make long fixations and short saccades, leading to a significant correlation.

² Within task, there are very high correlations for eye movement characteristics. Thus, Castelhano and Henderson (2008a) found that fixation durations for a group of viewers tended to correlate highly independently of whether a photo or line drawing was used as the stimulus. Likewise, saccade length was highly correlated. But, again, there was little correlation between fixation duration and saccade length per se.

³ The fact that words are skipped obviously means that readers do not invariably move forward in the text fixating on each successive word in its canonical order. However, some type of inner speech code presumably aids the reader to maintain the correct word order.

⁴ In the most extreme situation, the window contains only the fixated letter, thereby creating a situation in which the reader is literally forced to read letter by letter. In this situation, normal readers' eye movement data are very much like the eye movement data of brain-damaged pure alexic or letter-by-letter readers (Johnson & Rayner, 2007; Rayner & Johnson, 2005).

⁵ In some situations, readers can obtain information further to the left of fixation (Binder, Pollatsek, & Rayner, 1999). For example, when a word is skipped, attention may often be directed to the left of fixation following the skip.

⁶ In the boundary paradigm, the display change from the preview to the target word occurs during a saccade when vision is suppressed. Typically, readers are not aware of the change. But, can the results of boundary experiments be attributed to artefacts associated with the change? Inhoff, Starr, Liu, and Wang (1998) directly tested this by varying the speed of the display change and the refresh rate of the display monitor. They found no evidence to suggest that results of gaze-contingent change experiments were artefacts of the paradigm. Thus, when the timing of the display change is such that the change occurs during the saccade, researchers can be confident that the only thing that influences the data is the experimental manipulation. However, when the timing is such that the display change is too slow, a different pattern of data will appear. Thus, readers who are aware of the display change show a different pattern of data than do readers who are not aware (White et al., 2005a).

⁷ Chinese readers obtain preview benefit from characters and words to the right of their fixation (Inhoff & Liu, 1998; Liu, Inhoff, Ye, & Wu, 2002; Pollatsek, Tan, & Rayner, 2000b; Tsai, Lee, Tzeng, Hung, & Yen, 2004; J. Yang, Wang, Xu, & Rayner, in press; Yen, Tsai, Tzeng, & Hung, 2008), and they can sometimes obtain preview benefit from word n + 2.

⁸ A number of studies using either multiple isolated word-processing tasks, in which participants must look at, for example, three words in succession or tasks that mimic reading, have also reported evidence for parafoveal-on-foveal effects (Kennedy, 2000; Kennedy et al., 2004; Kennedy et al., 2002).

⁹ Although most research has indicated that word length is an important cue in deciding where to look next, Epelboim, Booth, and Steinman (Epelboim, Booth, Askkenazy, Taleghani, & Steinman, 1997; Epelboim, Booth, & Steinman, 1994, 1996) argued that word length per se is not a critical cue for eye guidance. Indeed, they claimed that reading unspaced text is “relatively easy”. However, analyses by Rayner and Pollatsek (1996) demonstrated that even in their experiments, most readers slowed down when reading unspaced text.

¹⁰ The interesting question about Chinese readers is how they segment words given the lack of space information (Inhoff & Wu, 2005; Li, Rayner, & Cave, in press; Wu, Slattery, Pollatsek, & Rayner, 2008).

¹¹ It is not possible to cite all of the many studies demonstrating the effects that are described here. Thus, in general the articles originally documenting the demonstrations and some recent demonstrations of the effects are listed (with an emphasis on studies from my lab).

¹² See also Filik (2008) for slightly different data on this issue, and Ferguson and Sanford (2008) for data consistent with Warren et al. (2008).

¹³ For a comprehensive overview of these models, see the special issue of Cognitive Systems Research (Reichle, 2006), and see Reichle et al. (2003) for a comparison of the models. See also Reichle and Laurent (2006) for a model describing how learning influences eye movements.

¹⁴ A recent neurophysiologically inspired model (Heinzle, Martin, & Hepp, 2009) has many of the same properties as those inherent in E-Z Reader, including serial lexical processing.

¹⁵ More densely packed scenes lead to longer fixations and shorter saccades.

¹⁶ A very promising variation on gaze-contingent moving-windows and moving-masks paradigms, discussed in the context of visual search, that has not yet been fully exploited is to use multiresolution displays (Loschky & McConkie, 2002; Reingold & Loschky, 2002; Reingold, Loschky, McConkie, & Stampe, 2003). With these types of display, a clear view of the scene can be provided around the fixation point with increasing degradation of the scene outside of the window.

¹⁷ The research on advertisements is quite interesting in the context of examining how viewers alternate their attention between pictorial and written information. The research indicates that the strategy of the viewer and their goal very much influence where they look.

¹⁸ The task originated with Cooper (1974) but has been effectively utilized by Tanenhaus, Altmann, and others to study a number of topics ranging from auditory word recognition to syntactic parsing. Some recent work by Altmann (2004), Richardson and Spivey (2000), and Ferreira, Apel, and Henderson (2008) is very interesting in that they show, perhaps surprisingly, that listeners fixate on now-empty regions that had previously been occupied by relevant objects. Ferreira et al. suggested that the “looking at nothing” finding perhaps provides some clues about how the visual system creates and stores internal memory representations and that looking at nothing aids retrieval of these representations.