Frequency and predictability effects in the Dundee Corpus: An eye movement analysis

Abstract

Analyses carried out on a large corpus of eye movement data were used to comment on four contentious theoretical issues. The results provide no evidence that word frequency and word predictability have early interactive effects on inspection time. Contrary to some earlier studies, in these data there is little evidence that properties of a prior word generally spill over and influence current processing. In contrast, there is evidence that both the frequency and the predictability of a word in parafoveal vision influence foveal processing. In the case of predictability, the direction of the effect suggests that more predictable parafoveal words produce longer foveal fixations. Finally, there is evidence that information about word class modulates processing over a span greater than a single word. The results support the notion of distributed parallel processing.

Keywords:

• Eye movements
• Fixation duration
• Gaze
• Reading
• Corpus studies

Acknowledgments

This research was supported by a grant from the UK Economic and Social Research Council (Project Number RES-000-23-1388-ECRP05). Thanks are due to Matei Vladeanu who designed much of the cloze procedure software and supervised the data collection. We are grateful to Dennis Drieghe, Reinhold Kliegl, Keith Rayner, Richard Shillcock, and an anonymous referee for comments on earlier versions of this paper.

Notes

¹ In a case of particular difficulty, the program to shift the eyes to the next word may be executed before the foveal word has been completely processed. This would produce a spillover effect, but the situation will be rare.

² Since this paper was prepared, a study by Dimigen, Sommer, Hohlfeld, Jacobs, and Kliegl (2011) has appeared, using coregistration of eye movements and fixation-related brain potentials. A regression model on N400 amplitude revealed main effects of frequency and predictability (with a positive coefficient) and also a significant interaction between these two factors (see Dimigen et al., 2011, Table 3). Nonetheless, interpretation of this outcome is complicated by the fact that the N400 effects occur rather late relative to their apparent equivalents in the eye movement record.

³ The English-language part of the Dundee Corpus, together with the complete data frame incorporating predictability scores, can be obtained from the first author.

⁴ A referee has pointed out that Kliegl et al. (2006) used single-fixation duration as a dependent variable. The difference is unlikely to be critical in the present context. First-fixation duration is possibly a noisier measure but, set against this, restricting selection to single-fixation cases tends to bias the data away from longer words.

⁵ It is possible to interpret the effect as a high-level anticipatory process, driven by parafoveal lexical access, in which case the outcome does not bear directly on the serial versus parallel distinction. However, such an interpretation is incompatible with current instantiations of the EZ Reader model and would involve the completion of substantially more high-level processing before the eyes leave a word than is implemented in the SWIFT model.