Cumulative Repetition Effects Across Multiple Readings of a Word: Evidence From Eye Movements

Abstract

When a word is read more than once, reading time generally decreases in the successive occurrences. This Repetition Effect has been used to study word encoding and memory processes in a variety of experimental measures. We studied naturally occurring repetitions of words within normal texts (stories of around 3,000 words). Using linear mixed models to analyze the evolution of fixations over successive repetitions, we observed an interaction between corpus word frequency and repetition. Specifically, we found a decrease in fixation durations in words with low frequency but not with high frequency, and both values converged after five or six repetitions. Furthermore, we showed that repetition of a lemma is not enough to evoke this effect. Our results are in agreement with predictions formulated by the context-dependent representation model, and this adds new arguments to the discussion of the sources of the repetition effect.

Acknowledgments

We thank Yamila Sevilla, Daniel M. Low, Carolina Gattei, and Gerardo Fernández for their thoughtful reading. We thank Thomas A. Gavin, Professor Emeritus, Cornell University, for help with editing the English in this article. This work is funded by the Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET), M.S. is sponsored by the James McDonnell Foundation 21st Century Science Initiative in Understanding Human Cognition Scholar Award.

Supplemental material

Supplemental material, data, and R scripts for statistical analyses and the generation of figures are available at http://reading.liaa.dc.uba.ar/.

Notes

¹ As expected from results on Gaze and First-Fixation Durations, Single-Fixation Duration also showed significant effects of both NREP (t = 8.5) and the interaction between NREP and frequency (t = − 5.8).

² Because the probability of skip, as defined in our dataset has only two possible values, that is, 0 if the participant fixated on a given word in that text, and 1 if the participant did not, we used a Generalized LMM with kernel = binomial, instead of the LMM that was used for continuous dependent variables.