Purchase this article with an account.
Jiajuan Liu, Zhong-lin Lu, Barbara Dosher; Similar perceptual learning in 10-alternative letter identification in external noise with and without feedback supervision. Journal of Vision 2020;20(11):1237. doi: https://doi.org/10.1167/jov.20.11.1237.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Perceptual learning in n-alternative identification (nAFC) has been found with accuracy feedback (correct or incorrect) for faces, filtered texture patterns (Gold et al. 1999), and peripheral letter identification in reduced vision (Chung et al., 2005). Here we examined learning without feedback and with response feedback (providing the correct response) in a 10AFC letter identification task, and compared the results with those of prior studies which showed (a) faster learning with response feedback but small amounts of learning without feedback (especially compared to the often robust unsupervised learning in 2AFC tasks) in 8AFC orientation identification (Liu et al, VSS 2018), and (b) both results can be successfully modeled via reweighting (improved readout) from a hierarchy of sensory representations in an n-AFC integrated reweighting theory (IRT) (Dosher et al. 2013). In the current study, observers were trained either without feedback or with response feedback to identify spatial-frequency filtered letters (CDHKNORSVZ) (Hou et al., 2015), temporally embedded (NSN at 60 Hz) in white external noise (sigma=0.33), with letter contrasts of 0.25, 0.5, or 1.0 intermixed randomly over trials. Observers practiced in five 1200-trial sessions and then one session without external noise and with feedback (for internal noise estimation). Learning was robust in both response feedback and no feedback conditions, and the rate of learning was only slightly faster for response feedback. These results contrasted noticeably with those for 8-alternative identification of unidimensional orientation, where response feedback yielded much faster learning rates, and learning without feedback was very modest (Liu et al, VSS 2018). IRT model simulations indicated that this reduced sensitivity to full feedback supervision likely reflects the multidimensional similarity structure of the letters (seen in multi-dimensional scaling of letter confusion data) and the role of pre-existing letter templates in learning to identify the letters in external noise.
This PDF is available to Subscribers Only