Learning to discriminate simple sinusoidal gratings is task specific

Abstract. Learning transfer effects of a modified spatial frequency discrimination task with simple sinusoidal gratings on various untrained test tasks, testing performance in relative position, local width and global size discrimination, have been investigated. Six subjects were exposed to grating stimuli of varying spatial frequency and size but constant relative position while they had to respond only to spatial frequency. In the course of 7 consecutive days all subjects showed significant reduction of spatial frequency discrimination thresholds. Comparison of discrimination thresholds for the untrained test tasks, taken before and after the learning epoch, reveals complete learning transfer to spatial frequency discrimination with gratings of constant size and variable relative position, but complete lack of transfer when grating spatial frequency is shifted about one octave lower. Further, there is improvement of relative position discrimination and width discrimination of single luminance bars, which is not very specific for the learnt spatial frequency. Although size variation was part of the learning procedure, discrimination of global stimulus size did not improve. Generally, the observed scheme of learning transfer reveals that there is learning only for stimulus attributes that are behaviorally relevant in the learning task. The differential scheme of improvement and code usage in the test tasks strongly indicates involvement of higher stages capable of independent access to different coding domains, as well as attentionally guided attribute selection and suppression. Supported by other recent findings, it is suggested that discrimination learning can be understood as a higher level process of gradual refinement of code selection out of a rich code base provided by lower level stages.