Untangling letter confusability and word length effects in pure alexia

ABSTRACT

Pure alexia is an acquired neuropsychological disorder that follows damage to the occipito-temporal lobe. This brain damage results in a severe reading impairment in which previously literate individuals are no longer able to efficiently read words, but are still able to perform other language tasks. The present study sought to identify factors of words that make it more difficult for pure alexic individuals to read, such as letter confusability and word length. Eye-tracking methodology was paired with a naming task to examine whether word length or letter confusability is a better predictor of processing difficulty. It was found that word length was a significant predictor of reading time, while summed letter confusability was not significant. This study contradicts some previous research and shows that when an orthogonal set of stimuli is used, letter confusability is not a significant factor driving this reading impairment in all individuals with pure alexia.

KEYWORDS:

• Pure alexia
• eye tracking
• word length effect
• letter confusability

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Items were selected as carefully as possible with an emphasis that none of the correlations among word length, summed letter confusability, frequency, age of acquisition, and imageability be strong or statistically significant, while still maintaining large variation within each of these variables to assess their predictive ability across the continuum. This is particularly difficult given that these variables are naturally strongly correlated with one another. It should be noted that although the correlation between word length and summed letter confusability approached significance (p = .073) in this experiment, this is driven by the large sample size used as the magnitude of the correlation was weak (r = .136). G.J. also participated in another naming study that explored only word length (from 5 to 7 letters) and summed letter confusability in a set of 240 items where the correlation between the two variables was even weaker, r(238) = .026, p = .686. When the log naming latencies of the 185 correctly named items were regressed onto both word length and summed letter confusability from Loomis (1982), word length was a significant predictor in the model (β = .340, t = 4.881, p < .001), but letter confusability was not (β = .060, t = .866, p = .388). When regression analyses were conducted using each of the other confusability matrices, similar results were found; word length was found to be a significant predictor (all ps < .027), but summed letter confusability was not (all ps > .402). This supports the results of the current study in showing that word length accounts for a greater proportion of the variability in visual word processing than many different measures of summed letter confusability when these two variables can be differentiated.

2 Dependent measures were log transformed so that they would be more normally distributed, an assumption necessary in multiple regression techniques. However, analyses on the raw naming times showed the same pattern of findings.

3 D. Fiset et al. (2005) use the finding that the word length effect disappeared when words were equated for summed letter confusability as evidence for parallel processing of letters. However, as Barton et al. (2014) point out, this conclusion from Fiset et al. seems surprising. As they write:

If all letters of a word are being processed simultaneously, then processing time should reflect either the average time to decipher each letter, or the time taken to decipher the most difficult letter . . . which [D. Fiset et al., 2005] showed was not the case. (Barton et al., 2014, 401)

In contrast, the fact that D. Fiset et al. (2005) found summed confusability to predict naming latencies during whole-word reading is actually evidence in favour of serial processing of letters, rather than parallel processing. Also, as Leff and Starrfelt (2014) note, even when the individuals in Fiset et al.’s study were presented with words with low summed letter confusability, their naming times were still quite elevated (averaging between 3000 and 3500 ms) relative to normal participants (averaging between 500 and 600 ms; Schilling, Rayner, & Chumbley, 1998) indicating that even in words with low levels of letter confusability, normal parallel processing cannot be achieved. Finally, it is crucial to mention that the calculation of a word's summed letter confusability can be reduced to the product of a word's average letter confusability multiplied by its word length. Hence, at best, summed letter confusability “is a more sophisticated measure of the effect of word length, rather than an explanation of the effect” (Barton et al., 2014, 402).