Hoarding titmice predominantly use Familiarity, and not Recollection, when remembering cache locations

Abstract

Scatter-hoarding birds find their caches using spatial memory and have an enlarged hippocampus. Finding a cache site could be achieved using either Recollection (a discrete recalling of previously experienced information) or Familiarity (a feeling of "having encountered something before"). In humans, these two processes can be distinguished using receiver operating characteristic (ROC) curves. ROC curves for olfactory memory in rats have shown the hippocampus is involved in Recollection, but not Familiarity. We test the hypothesis that food-hoarding birds, having a larger hippocampus, primarily use Recollection to find their caches. We validate a novel method of constructing ROC curves in humans and apply this method to cache retrieval by coal tits (Periparus ater). Both humans and birds mainly use Familiarity in finding their caches, with lower contribution of Recollection. This contribution is not significantly different from chance in birds, but a small contribution cannot be ruled out. Memory performance decreases with increasing retention interval in birds. The ecology of food-hoarding Parids makes it plausible that they mainly use Familiarity in the memory for caches. The larger hippocampus could be related to associating cache contents and temporal context with cache locations, rather than Recollection of the spatial information itself.

Keywords: Cognitive evolution; Comparative cognition; Episodic-like memory; Hippocampus; Periparus ater; Scatter-hoarding.

Fig. 1

Graphical illustration of our memory metrics. A: ROC curves in Retrieval (upper panel) and Naïve (lower panel) conditions. The area under the ROC curve, the AUROC, is shaded. To convert AUROC to a measure of Overall performance, we subtract the area corresponding to performance less than chance, shaded in gray, and double the remainder. This is shown as the pink region in D. B: The shaded region is AUROC_R, i.e., the area under the ROC curve which would be obtained with the same value of R (same intercept) but d’ = 0 (no curvature). To obtain a measure of Recollection, we again subtract the area below chance and double the remainder (red region in E). This is in fact equal to the parameter R. C: The shaded region is AUROC_d’, i.e., the area under the ROC curve which would be obtained with the same value of d’ but no recollection. To obtain a measure of Familiarity, we again subtract the area below chance and double the remainder (green region in F). We do this for the Retrieval and Naïve conditions. The difference between the two conditions gives us our measure of enhancement: the enhancement in Overall performance (G), in Recollection specifically (H) and in Familiarity specifically (I)

Fig. 2

Outcomes of human validation trials, comparing Confidence and Order metrics for the Retrieval condition. A: Distribution of participants’ individual Pearson correlation coefficients between the order in which they selected targets and their reported confidence, for the 49/54 participants who used more than one confidence rating. In obtaining these correlations, all searches were used (i.e., we did not truncate if the number of false alarms exceeded the number of objects hidden, unlike what we did for fitting the ROC curves). Error bars show the 95% confidence interval on individual correlation coefficients. 43/49 individual correlation coefficients were significantly negative (p < 0.05, filled symbols). Overall, the whole population was very significantly negative t₄₈ = -17.9, p < 0.001. Results for the Choice and Random group are plotted separately but there was no difference between them (t₂₉ = -1.09, p = 0.29, two-sample t-test; colored horizontal lines show the median for each group and the outline represents the data distribution). B–D Scatterplot of Overall performance, Familiarity, and Recollection derived from Order scores of human participants against Overall performance, Familiarity, and Recollection derived from reported Confidence (gray dashed lines are the identity lines; continuous red lines are linear fits using Type II regression, blue dashed lines show the lines resulting from the 97.5% intercept and slope, and the 2.5% intercept and slope; large crosses show group mean ± 1SEM; pink, blue are for participants in the Choice, Random groups, respectively). Symbol area represents number of coins hidden. Titles report Pearson correlation coefficients and significance. [This figure was generated by file StatsAndFigures_HumanValidation.R] (color figure online)

Fig. 3

A – Distributions of Recollection and Familiarity Enhancement, for birds in the 1-day retention interval and for human participants in Choice and Random groups. Horizontal bars show the mean of each distribution. The horizontal line through 0 represents same performance during the Retrieval and Naïve conditions, i.e., no memory. Asterisks indicate whether the values were significantly different from zero (t-test; ** p < 0.01, *** p < 0.001). B – Average ROC curves, using the mean d’ and R for each group/condition. [This figure was generated with StatsAndFigures_CompareHumansAndBirds.R]

Fig. 4

A. Memory Enhancement for coal tits as a function of retention interval. Colors identify the 9 different Retrieval birds. The size of the symbol represents the number of seeds hidden (nHidden), which was out of our control and determined by the birds’ behavior. Colored lines show the fitted regression for each Retrieval bird; the black line + ribbon shows the regression fitted to all birds, ± SEM. Asterisks above the axis labels indicate whether the Enhancement was significantly different from zero for the given retention interval (* p < 0.05, ** p < 0.01). B: Average ROC curves for bird groups. The thick dashed pink curve shows the ROC curve computed with the mean R and d’ averaged from all Naïve birds in all “delays” (i.e., seed distributions). The thick colored lines show ROC curves computed with the mean R and d’ averaged over all available Retrieval birds at the specified retention interval, i.e., 1, 3, 7, 14, 28, and 42 days after caching, as indicated by the colors. The horizontal lines mark the y-intercept in each case, corresponding to the Recollection parameter R. [This figure was generated by file StatsAndFigures_Birds.R] (color figure online)