Mental rotation

Compare two pictures of a bent block. Are they the same object viewed from different angles, or mirror images? People often take longer as the required turn grows. Do they turn an internal image continuously, change viewpoints in separate jumps, or recognize the shape without turning it?[1]

Compare two pictures of a bent arrangement of cubes: one may be the same object turned around, or a mirror image. How do people bring the views into correspondence to decide? A longer response at a larger angle could reflect continuous internal turning, separate view-changing operations, or a different comparison of shape.

A model informed by interactive VR experiments uses large quarter-turn or half-turn transformations before comparing descriptions. It predicts important behavior, but often takes the same number of operations at 120° and 180°. Human responses without overt rotation continue to slow across that interval, a difference the model does not directly explain. Evaluation includes 58 held-out objects, meaning objects excluded from training, in a fixed-axis cube-object set. This is evidence of transfer within that regime, rather than arbitrary shape and rotation-axis generalization.[1]

A successful 2026 model uses large discrete view changes. Yet it does not directly explain why human responses keep slowing from 120° to 180° when its operation count is often unchanged.[1]

One operation can take more time at a larger angle, so counting operations does not settle their duration. The model authors propose a possible neural implementation with such a time cost, but do not establish it in humans. Intermediate neural states and performance on new shapes and axes could distinguish the accounts more sharply. A competing multi-view model’s abstract reports successful behavioral prediction; that correspondence also does not identify the human algorithm.[1] [2]

A bent-block object keeps its shape when it is turned, but its images can look very different. Recognizing the common shape connects seeing with thinking about how things are arranged in space. Explaining how people compare those views would clarify what imagining a turn or comparing the parts contributes to that ability. A model can make the right shape judgment without necessarily using the same steps as a person.[1]

Turning a shape in thought is one way to bring different views into correspondence; recognizing their common structure could solve the comparison another way. The distinction matters for understanding what internal spatial reasoning contributes beyond a correct answer. Predicting choices, time and intermediate states on unseen cases would constrain that contribution more strongly than task accuracy alone, without establishing a unique human algorithm.[1] [2]