Variation-Set Structure: Extraction of Important Segments in Pigeon Learning.

Abstract The ability to extract reliable segments from the sensory data-stream is necessary for learning and for creating a reliable representation of the environment. Here we focused on segmentation in visual learning associated with foraging. Fifty-one pigeons (Columba livia) were trained to recognize two food-predicting signals, one presented against fixed background elements and the other against varying backgrounds. The pigeons were divided into two groups: 32 pigeons were trained with short intervals of 3-5 minutes between sessions and 19 with long intervals of 60 minutes between sessions. When tested on the signal with a plain background, 24 pigeons trained with the short time intervals preferred the signal presented against a varying background during training and eight preferred the signal presented against a fixed background during training. Nine pigeons trained with the long intervals preferred the signal presented against a varying background during training and ten the signal presented against a fixed background during training. Our findings indicate that pigeons possess an ability to use variation-set structure to segment visual data. To date, the effect of variation-set structure has been demonstrated only in the context of human language acquisition. Our current findings suggest that variation sets structure may be linked to other learning abilities too.