How Effective is Example Generation for Learning Declarative Concepts?

Facilitating transfer through incorrect examples and explanatory feedback

This article examines whether studying correct versus incorrect examples produces differential learning. A prediction that follows from behaviourism is that learning should be best from studying correct examples. A contrasting prediction is that incorrect examples can highlight a concept's properties that are missing in the example, and thereby enable concept learning. We test these ideas across two experiments, wherein subjects were shown hypothetical study scenarios and were asked to determine whether each was a true experiment. In Experiment 1, some subjects were only presented correct examples, some were only presented incorrect examples, and others were presented both. In addition, example type was crossed with feedback type: Some subjects were given explanatory feedback and some were not given any feedback; a control condition was also included, wherein subjects were not shown any study scenarios. All subjects completed a posttest involving novel scenarios; some questions asked subjects to indicate whether they were true experiments (classification questions), and some asked them to specify what was lacking in the design or to indicate how it could be fixed (application questions). Experiment 2 used a similar design, but compared explanatory feedback with corrective feedback. In both experiments, as long as some form of feedback was provided, subjects in the mixed example condition achieved the best classification performance. Furthermore, subjects in the incorrect and mixed example conditions performed best on application questions, particularly when explanatory feedback was provided. These findings offer insights into the mechanisms that might underlie learning from incorrect examples.

Cognitive Prerequisites for Generative Learning: Why Some Learning Strategies Are More Effective Than Others

This study examined age-related differences in the effectiveness of two generative learning strategies (GLSs). Twenty-five children aged 9-11 and 25 university students aged 17-29 performed a facts learning task in which they had to generate either a prediction or an example before seeing the correct result. We found a significant Age × Learning Strategy interaction, with children remembering more facts after generating predictions rather than examples, whereas both strategies were similarly effective in adults. Pupillary data indicated that predictions stimulated surprise, whereas the effectiveness of example-based learning correlated with children's analogical reasoning abilities. These findings suggest that there are different cognitive prerequisites for different GLSs, which results in varying degrees of strategy effectiveness by age.

Lighting the wick in the candle of learning: generating a prediction stimulates curiosity

Curiosity stimulates learning. We tested whether curiosity itself can be stimulated-not by extrinsic rewards but by an intrinsic desire to know whether a prediction holds true. Participants performed a numerical-facts learning task in which they had to generate either a prediction or an example before rating their curiosity and seeing the correct answer. More facts received high-curiosity ratings in the prediction condition, which indicates that generating predictions stimulated curiosity. In turn, high curiosity, compared with low curiosity, was associated with better memory for the correct answer. Concurrent pupillary data revealed that higher curiosity was associated with larger pupil dilation during anticipation of the correct answer. Pupil dilation was further enhanced when participants generated a prediction rather than an example, both during anticipation of the correct answer and in response to seeing it. These results suggest that generating a prediction stimulates curiosity by increasing the relevance of the knowledge gap.

Why are children overconfident? Developmental differences in the implementation of accessibility cues when judging concept learning

Children are often overconfident when monitoring their learning, which is harmful for effective control and learning. The current study investigated children's (N=167, age range 7-12years) judgments of learning (JOLs) when studying difficult concepts. The main aims were (a) to investigate how JOL accuracy is affected by accessibility cues and (b) to investigate developmental changes in implementing accessibility cues in JOLs. After studying different concepts, children were asked to generate novel sentences and then to make JOLs, select concepts for restudy, and take a final test. Overconfidence for incorrect and incomplete test responses was reduced for older children in comparison with younger children. For older age groups, generating a sentence led to greater overconfidence compared with not being able to generate a sentence, which indicates that older children relied more on accessibility cues when making JOLs. This pattern differed in the youngest age group; younger children were generally overconfident regardless of whether they had generated sentences or not. Overconfidence was disadvantageous for effective control of learning for all age groups. These findings imply that instructions to encourage children to avoid metacognitive illusions need to be adapted to children's developmental stage.