Date Published: July 19, 2017
Publisher: Public Library of Science
Author(s): Matthew G. Wisniewski, Milen L. Radell, Barbara A. Church, Eduardo Mercado, Stefan Elmer.
Individuals learn to classify percepts effectively when the task is initially easy and then gradually increases in difficulty. Some suggest that this is because easy-to-discriminate events help learners focus attention on discrimination-relevant dimensions. Here, we tested whether such attentional-spotlighting accounts are sufficient to explain easy-to-hard effects in auditory perceptual learning. In two experiments, participants were trained to discriminate periodic, frequency-modulated (FM) tones in two separate frequency ranges (300–600 Hz or 3000–6000 Hz). In one frequency range, sounds gradually increased in similarity as training progressed. In the other, stimulus similarity was constant throughout training. After training, participants showed better performance in their progressively trained frequency range, even though the discrimination-relevant dimension across ranges was the same. Learning theories that posit experience-dependent changes in stimulus representations and/or the strengthening of associations with differential responses, predict the observed specificity of easy-to-hard effects, whereas attentional-spotlighting theories do not. Calibrating the difficulty and temporal sequencing of training experiences to support more incremental representation-based learning can enhance the effectiveness of practice beyond any benefits gained from explicitly highlighting relevant dimensions.
Two perceptual events that are difficult or impossible for an individual to distinguish can become discriminable through a training procedure that starts with easy distinctions and gradually progresses to more subtle differences [1–2]. This phenomenon has been referred to as the easy-to-hard effect or transfer along a continuum, while the progressive procedures used to induce the effect have been termed fading or progressive training. Pavlov  demonstrated the easy-to-hard effect in dogs learning to discriminate visual, auditory, and somatosensory stimuli. Early studies in humans also showed easy-to-hard effects for simple images and sounds [3–5]. More recent work has established that fading influences not only acquisition, but also perceptual generalization [6–7] and cortical plasticity . The easy-to-hard effect was once a major focus of associative learning research because of extensive debates about whether effects were due to an increase in dimensional salience , or acquired gradients of association (reviewed by ). Recently, similar debates have arisen in the context of perceptual learning/perceptual category learning studies, with some researchers again arguing that this effect is due to increases in dimensional salience [11–12], whereas others argue that the effect can be explained in terms of gradual changes in stimulus representations and/or their associations [13–20].
In Experiment 1, participants were trained simultaneously in progressive and constantly difficult training regimens to categorize FM sweep trains with different rates of frequency modulation as ‘Fast’ or ‘Slow’. Progressive and Constant regimens were assigned to different frequency ranges and were counterbalanced. Participants were tested post-training with both frequency ranges. Effects of regimen were examined.
That a progressive advantage was found within-subjects when comparing conditions that had the same critical relevant dimension suggests that learning involved processes beyond dimensional-highlighting. Representation-based theories where learning occurs because of changes to the stimulus representations themselves, or in the read-out connections from those representations, predict that benefits should be restricted at least partially to the trained sounds (i.e., the sounds that elicit those representations). Representation-based accounts of perceptual learning are thus more consistent with the current data.
Experiment 2 investigated whether or not there is a progressive advantage in the generalization of learning [6–8]. Two types of generalization were examined. First we examined whether the effect would remain when participants were tested in an untrained task. After training similar to Experiment 1, participants were tested in a two-interval two-alternative forced-choice (2i-2afc) task on their ability to discriminate rate in both the progressive and constant trained frequency ranges. Two sounds of differing FM rates were presented, and participants were asked to indicate which was faster. We also tested whether or not there would be a progressive advantage when stimulus contrasts were made more difficult than training. Here, we shortened sweep trains in order to make the task more difficult. Shorter novel sweep trains with less repetitions were tested in addition to trained sweep trains to characterize potential differences in generalization of learning to discriminate more difficult contrasts. There were two reasons for the methodological changes from Experiment 1 to Experiment 2. If the progressive advantage found in Experiment 1 extends to an untrained task (i.e., the 2i-2AFC task) and to untrained stimulus contrasts (i.e., shorter sweep trains), this would provide support for a representation-based account of the progressive advantage. It would also suggest that these perceptual learning mechanisms are potentially relevant for real world training applications because they generalize beyond the exact circumstances of training.
Experiment 2 replicated the Experiment 1 finding that the progressively trained range showed enhanced performance post-training. Further, the progressive advantage was observable in an untrained 2i-2afc task, suggesting that the processes involved in learning are task-general rather than task-specific. The progressive advantage was also not stimulus specific. That is, the effect occurred for both trained and untrained sounds that contained less FM sweep repetitions compared to training.
The notion that focusing attention on a relevant perceptual dimension explains perceptual learning and easy-to-hard effects has been around for over a century [1,9,11]. In early writing on the topic, James  reported that most scientists/philosophers of the time had dismissed perceptual learning as a topic of study, assuming that the theoretical mechanisms were determined on the basis that: “what we attend to we perceive more minutely”. Unsatisfied with this as the sole explanation for perceptual learning, James offered a theory in which sensory impressions are discriminated based upon their associations with memories of past events, which could also account for easy-to-hard effects. Later, in a classic easy-to-hard effect demonstration, Lawrence  found that rats given 30 initial easy trials performed better than rats trained in a constantly hard regimen to discriminate stimulus brightness. He concluded that dimensional discovery should play some role in learning theory in addition to associative mechanisms (also see ). Simulation work later revealed that such effects could be accounted for by both associative (e.g., [20, 47]) and non-associative [18–19] representation-based learning mechanisms without assuming dimensional discovery. Even so, attentional-spotlighting views have continued to be popular accounts of the easy-to-hard effect and perceptual learning.
Although attentional spotlighting can in some cases be useful for learning to make fine perceptual distinctions, it alone is not a sufficient explanation of easy-to-hard effects. Attentional-spotlighting accounts incorrectly predict that easy-to-hard sequencing should aid discrimination performance all along the discrimination relevant dimension. They also incorrectly predict that when a participant’s attention is explicitly and repeatedly drawn to relevant dimensions early in training (e.g., by the presentation of easy contrasts in one range of that dimension), then he or she should show no within-subject benefits of progressive training (e.g., ). In contrast to the attentional-spotlighting explanation of easy-to-hard effects, proposed representational modification/reweighting learning mechanisms (e.g., [16,19,30]) are able to account for the specificity of easy-to-hard effects to trained sounds and the presence of an easy-to-hard effect when relevant dimensions are clearly revealed. Future theoretical and applied work may benefit from consideration of how multiple processes contribute, and possibly interact, to modify perceptual acuity.