Many theories of conditioning describe learning as a process by which stored information about the relationship between a conditioned stimulus (CS) and unconditioned stimulus (US) is progressively updated upon each occasion (trial) that the CS occurs with, or without, the US. These simple trial-based descriptions can provide a powerful and efficient means of extracting information about the correlation between two events, but they fail to explain how animals learn about the timing of events. This failure has motivated models of conditioning in which animals learn continuously, either by explicitly representing temporal intervals between events, or by sequentially updating an array of associations between temporally distributed elements of the CS and US. Here, I review evidence that some aspects of conditioning are not the consequence of a continuous learning process but reflect a trial-based process. In particular, the way that animals learn about the absence of a predicted US during extinction suggests that they encode and remember trials as single complete episodes rather than as a continuous experience of unfulfilled expectation of the US. These memories allow the animal to recognise repeated instances of non-reinforcement and encode these as a sequence which, in the case of a partial reinforcement schedule, can become associated with the US. The animal is thus able to remember details about the pattern of a CS’s reinforcement history, information that affects how long the animal continues to respond to the CS when all reinforcement ceases.
The Importance of Trials