Reading about a RELO-VUTION

Abstract

Pseudowords created by transposing two letters of words (e.g., MOHTER; CHOLOCATE) are highly confusable with their base word; this is known as the transposed-letter similarity effect. In this work, we examined whether transposed-letter effects occur when words span more than one line (e.g., CHOLO- in one line and CATE in another line; note that the transposed letters L and C are in different lines). While this type of presentation is not the canonical format for reading in alphabetic languages, it is widely used in advertising, billboards, and street signs. Transposed-letter pseudowords and their replacement-letter controls were written in the standard one-line format versus a two-line format (Experiments 1–2) or a syllable-per-line format (Experiment 3). While results showed some decrease in the transposed-letter effect in the two-line and syllabic formats, the transposed-letter effect was still substantial in the accuracy of responses. These findings demonstrate that even when the letters being transposed are relatively far apart in space, the transposed-letter effect is still robust. Thus, a major component of letter position coding occurs at an abstract level.

Appendix 1: 2-D overlap account

As stated in “Introduction”, the overlap model’s assumptions of location uncertainty can be easily extended to the 2-dimensional stimuli presentation used in the present experiments—for simplicity, we focused on the two-line manipulation of Experiments 1–2. Specifically, we assumed that the position uncertainty could be explained as a bivariate Gaussian function as depicted by the blobs/ellipses in Fig. 3. This two-dimensional position uncertainty is centered in the middle of the letter. Because the horizontal and vertical dimensions are independent, they have zero covariance and independent sigmas (i.e., for all letters, the position uncertainty in the vertical dimension is independent of the position uncertainty in the horizontal dimension).

We made basic assumptions in line with the Gómez et al.’s (2008) paper, namely that the first location in each line would benefit from the empty space and would have a smaller position uncertainty along the horizontal dimension (s₁ = 0.5), and the position uncertainty would be higher for subsequent letters (s_>1 = 1.2). Similarly, we assumed that along the vertical dimension the position uncertainty would be equal for both lines (s₁ = s₂ = 0.5). The panels in Fig. 3 were drawn under those assumptions. Using this 2-D model for the two-line presentation, along with the standard overlap model, we calculated the overlap between pseudowords and the base word for all letter lengths (the columns labeled “Overlap” in Table 2; for comparison, we also present transposed-letter effects for all word lengths in this experiment).

Fits of the data

While the empirical transposed-letter effect in the two-line nonwords (Experiments 1–2) was more than half of that in the one-line nonwords, the 2-D overlap model predicts a rather miniscule transposed-letter effect for the two-line presentation which is orders of magnitude smaller than the overlap for one-line presentation.

It is important to note that the overlap model is not a model of the lexical decision task. Hence, the overlap measurement must be scaled somehow to make predictions about RT and accuracy. Nonetheless, the difference in overlap between one- and two-line presentations is so large for all letter lengths that there is no realistic scaling factor that could account for the sizable empirical transposed-letter effects in the two-line condition.