aspect of a prior learning trial improves performance, even though this procedure is not sufficient to evoke the trained response. Retrieval cues are generally presented near the beginning of a trial; presumably they activate a memory of events from prior trials. If so, then the memory of past events must take the form of an anticipation or "representation" ( Roitblat, 1982) of events which will occur in the current trial. I want to suggest in this essay that a similar process of retrieval and representation enters into acquisition, or associative memory, as well, particularly, when there are delays between the important events within trials.
In Capaldi's work on delayed alternation of reward, rats learned to use the outcome of one trial as the cue for the reinforcement condition in the next trial. Capaldi and Stanley ( 1963) showed that rats learned to run quickly after an unrewarded trial (N) and slowly after an unrewarded one (R) in the NRNRNR. . .sequence. Performance was rather insensitive to the ITI, and an ITI of 20 min was mastered with little difficulty. The paradigm is not restricted to the alternation of reward and non-reward. Pschirrer ( 1972) ran rats with a fixed repeating sequence of milk, pellets, and non- reward. They learned to runs slowly only on those trials which followed pellets as the prior reward. Thus they learned to use the type of reward from a prior trial as a cue for non-reward in the current trial following a 15-min ITI. Petrinovich and Bolles ( 1957) showed that rats could learn to alternate the arms of a T-maze on successive runs during a session. Some rats could perform well with ITI's of a few hours.
It is of course possible that memories for appropriate events remain active between trials in situations such as these. But a retrieval mechanism provides a more credible account. Presumably the memory of each trial became "inactive" during the long ITI, and cues from the apparatus retrieve it at the beginning of the next trial. The rats would need to learn to retrieve the memory only of the preceding trial; otherwise interference would result. An interpretation in terms of retrieval receives further support from work by Capaldi ( 1971) on a double-alternation problem (NNRRNNRR. . .). In this procedure, non-rewarded trials follow rewarded and non-rewarded trials equally often. Rats find it almost impossible to learn this problem in the standard way. Capaldi overcame the difficulty by alternating the color of the alley (black or white) so that the problem was reduced to two independent single-alternation problems differentiated by color: BN, WN, BR, WR, BN, WN, BR, WR. . . Alley color presumably provided a retrieval cue which enabled the rat to remember the outcome of the previous trial with the same color. When alley color was randomized with respect to the double alternation, the rats did not learn the problem.