Midsession Reversal Task with Variable Trial Spacings: Further Tests of the Timing Hypothesis with Starlings

This study examined how starlings (Sturnus unicolor) adapt to a serial learning task with a predictable reversal in the reinforcement contingencies at midsession. The birds learned a simultaneous discrimination between two options, S1 and S2 (red and green key light colors). Choices of S1 were rewarded during the first 40 trials and choices of S2 were rewarded during the last 40 trials, with variable exponentially distributed ITIs separating the trials. Then, to test the hypothesis that starlings anticipate the midsession reversal based on time into the session, we changed the average of the ITIs during a test session. The hypothesis predicted that with ITIs twice as short during testing, preference would shift from S1 to S2 twice as many trials later than in training, and with ITIs twice as long during testing, preference would shift twice as many trials earlier than in training. Results showed that preference shifted in the predicted direction, but the shifts were smaller in magnitude than predicted. Cumulative difference records plotting choices across time- or trial-into-the-session revealed a variety of adjusting strategies, some consistent with the use of temporal cues, others consistent with the use of local or numerical cues. The variability of strategies occurred both between and within subjects and suggests that multiple cues combine to control behavior in the midsession reversal task.