Generalisatiekloof

Generalisatiekloof is a Dutch term used in cognitive science, education and artificial intelligence to describe the gap between generalizable knowledge or rules and their reliable application to novel or diverse situations. The word combines generalisatie (generalization) and kloof (gap), and is often used to discuss why people or systems succeed in some settings yet fail when contexts change.

In cognitive science, the concept refers to transfer-of-learning challenges: individuals may learn a rule in one task but apply it poorly to different tasks that require the same underlying principle. In machine learning and AI, it denotes the generalization gap: models that perform well on training data may underperform on out-of-distribution data because the training regime does not capture the full variability of real-world scenarios.

Causes of the generalisatiekloof include limited training data, overfitting to superficial cues, context-specific representations, and mismatches between training and deployment environments. Psychological factors, such as reliance on memorization rather than abstraction, also contribute, as can biases in evaluation that overemphasize in-domain performance.

Addressing the generalisatiekloof involves strategies such as promoting abstraction and principled reasoning in education, using data augmentation and regularization in ML, designing evaluation protocols that test cross-domain generalization, and incorporating causal or invariant representations to improve transfer.

See also: generalization, transfer of learning, domain adaptation, distribution shift, overfitting