Abstract
Crowded letter recognition can be enhanced by training on character-based tasks or by non-task-based training (stimulus exposure only). Here, we asked whether the variation in learning benefits across training methods can be accounted for by stimulus-dependent factors such as letter exposure, English-language letter frequency, and letter spatial complexity. We analyzed the data of four training groups sourced from two learning studies (Yu, Legge et al., 2010; Yu, ARVO 2013). All groups completed pre- and post-tests including a letter-recognition task (identifying trigrams, strings of three random letters, presented at varying distances left and right of the midline 10° below fixation). Training comprised four or five daily one-hour sessions. The four training methods are trigram training (identifying three letters), lexical-decision training (discriminating three-letter words from non-words), task-absent trigram exposure with repetition (having the option of repeating the same trigram stimulus for as many times as needed), and task-absent trigram exposure without repetition. We modeled post-pre improvement of crowded letter recognition (examining middle letter of the trigrams only) as a function of letter exposure, frequency, complexity, and the interaction terms. The best-fit model (p<.0001; R2=.19) included two significant predictors—letter exposure, and interaction of exposure and frequency. Across training groups and individual letters, letter exposure ranged from 0 to 3515 occurrences. Letter frequency ranged from 0.09% to 12.55% (Jones & Mewhort, 2004). The model suggests that despite differences in training protocols across groups, globally, letter exposure is the best predictor of post-pre improvement (positive correlation) among the variables investigated, and that the slope for exposure is varied depending on letter frequency (negative correlation). Our results indicate that perceptual learning protocols can benefit from taking into account the amount of letter exposure and letter frequency.
Meeting abstract presented at VSS 2014