Lexical quality and eye movements: Individual differences in the perceptual span of skilled adult readers

Abstract

Two experiments used the gaze-contingent moving-window paradigm to investigate whether reading comprehension and spelling ability modulate the perceptual span of skilled adult readers during sentence reading. Highly proficient reading and spelling were both associated with increased use information to the right of fixation, but did not systematically modulate the extraction of information to the left of fixation. Individuals who were high in both reading and spelling ability showed the greatest benefit from window sizes larger than 11 characters, primarily because of increases in forward saccade length. They were also significantly more disrupted by being denied close parafoveal information than those poor in reading and/or spelling. These results suggest that, in addition to supporting rapid lexical retrieval of fixated words, the high quality lexical representations indexed by the combination of high reading and spelling ability support efficient processing of parafoveal information and effective saccadic targeting.

Keywords:

• Reading
• Eye movements
• Individual differences
• Lexical quality
• Perceptual span

Notes

¹ The Nelson–Denny Reading Test also includes a measure of reading speed. When this variable was included in the analysis, the pattern of significant results was virtually identical. However, it was decided not to include this measure due to concerns with the validity of self-reported reading speed measures (Carver, 1985).

² The relatively low correlations between reading and spelling do not reflect ceiling effects in the assessment instruments. No participant in either experiment achieved a perfect score on the Nelson–Denny Reading Test. In Experiment 1, total scores ranged from the 23rd to the 99th percentile for U.S. first-year college students, with mean percentile ranks of 63 and 92 for the lower and upper half of the sample. In Experiment 2, participants' scores ranged from the 46th to the 99th percentile, and the average percentile rank of the upper and lower halves of the sample were 69 and 92, respectively.

³ In the analysis of both experiments, models that included random slopes for subjects and items corresponding to each of the fixed effects, interactions, and correlations between these effects (i.e., maximal random effect structures) failed to converge. Simpler models that removed random slopes for the interactions did converge but some of the correlations were 1, indicating that their inclusion would risk overfitting. A model specifying subject random slopes for the window factor converged but also yielded correlations very close to 1. The data reported are therefore for models that only include subject and item random intercepts. The pattern of significant effects did not differ between the final model and the converged models that specified random slopes.

⁴ We also conducted an analysis in which average reading rate from the full-line condition was a continuous predictor in the models, mirroring Ashby et al. (2012) and Rayner et al. (2010). The results of this analysis were clear: Reading ability and spelling ability no longer predicted average reading speed but all of the other significant effects remained. That is, even when reading speed was controlled for, lexical processing ability significantly modulated the size and use of the perceptual span.

⁵ It is possible that the use of upper-case Xs outside the moving window in the two experiments may have contributed to this finding. This type of mask may have been more perceptually salient in the periphery than, for example, a random letter mask. Nevertheless, the use of upper-case X masks is consistent with the majority of the moving-window literature.