Investigating the Relationship Between Perceptual Learning and Statistical Learning in Human Vision
- Author(s): Bufford Funk, Carolyn Ann
- Advisor(s): Kellman, Philip J
- et al.
Perceptual learning (PL) and statistical learning (SL) both seek to explain how humans learn through experience. However, research lacks clear, consistent definitions, causing collective confusion about the relationship between SL and PL: some researchers view SL and PL as distinct learning processes, while others view them as part of a unified learning process. We describe two distinct learning concepts that cohere with most work: PL is improving perception and, in a psychophysical sense, improving sensitivity. SL is recording co-occurrences in memory, and perhaps, in a psychophysical sense, changing criterion or bias. These concepts allowed us to test the relationship between PL and SL. We developed a novel psychophysical assessment to measure incidental PL in a well-known visual SL paradigm of shape pairs presented simultaneously. Experiments 1 and 3 used the paradigm’s SL familiarization and SL familiarity test, then our PL assessment. We also tested incidental SL in PL: Experiment 2 used PL training on stimuli from the SL paradigm, then a brief SL familiarity test for an incidental pair, then the assessment. Experiments 1 and 3 showed familiarity, but Experiment 2 showed no evidence of familiarity in the SL test. Experiments 1 and 3 showed PL effects: transfer in increased accuracy and sensitivity and decreased false alarming relative to baseline, evidence of incidental PL in an SL paradigm. Reducing SL by reducing familiarization session length caused weaker PL, observed only for the longest exposure duration. Experiment 2 showed stronger PL effects on the assessment, but no SL. We discuss several ways SL and PL could be related and compare each possibility to our results. Several of our results are consistent with SL and PL being part of a unified learning process, or at least occurring under overlapping conditions, but the relationship may be more nuanced and asymmetric: PL may occur more under conditions designed for SL, but SL may be less likely to occur during focused PL tasks. These results help clarify and unite rich but separate literatures on perceptual and statistical learning.