Transfer from continuous to discrete quantities in honeybees

Abstract

Honeybees can estimate quantities having different dimensions: continuous and uncountable such as the relative size of visual objects in an array, or discrete and countable such as the number of objects of the array. Honeybees can transfer quantity discrimination (i.e., choosing the larger/smaller stimulus) from number to size. Here, we investigated whether honeybees could also generalize from the size (continuous) to the number (discrete) dimension. We trained free-flying foragers to discriminate between large- and small-size elements. At test, bees were presented with a comparison between larger and smaller numerosities controlled for different continuous variables covarying with numerosity such as total area, total perimeter, convex hull, and element size. Results showed that bees generalized from the size to the numerical dimension of the stimuli. This cross-dimensional transfer supports the idea of a universal mechanism for the encoding of abstract magnitudes in invertebrate species comparable to that of vertebrate species.

Keywords: Behavioral neuroscience; Biological sciences; Natural sciences.

Graphical abstract

Figure 1

Schematic representation of the Y-maze

Figure 2

Qualitative overview of the performance at the number tests Results of the number tests (group means with SEM are shown); white dot represents single subject performance; dashed line indicates 50% chance level. No significant effect of the type of test was found (Analysis of variance (ANOVA): type of test (F_(3,72) = 2.44, p = 0.072; no asterisks indicate no significance).

Figure 3

Results of the size and number generalization tests Results of the size and number generalization (overall) tests (group means with SEM are shown; white dot represents single subject performance; dashed line indicates 50% chance level). In the size generalization test, honeybees chose the larger or smaller relative size according to their previous training (size generalization test: 59.89% ± 2.01%, mean% ± SEM%; two-tailed one-sample t test: t₍₁₉₎ = 4.91, p < 0.001). In the number generalization test bees chose the congruent stimulus according to their previous training (i.e., 4 elements if they were trained to choose the smaller size stimulus, and 8 elements if they were trained to choose the larger size stimulus during the training) (number generalization test: 52.84% ± 1.23%, mean% ± SEM%; two-tailed one-sample t test: t₍₁₉₎ = 2.31, p = 0.032) (∗p < 0.05; ∗∗p < 0.01; ∗∗∗p < 0.001).

Figure 4

Spatial frequency analysis of the training and test stimuli Spatial frequency of the stimuli presented during the training and the test phase. Spatial frequency varied directly along the stimulus dimension (i.e., either size or numerousness) during training (smaller size total power = 28878,9 ± 70, mean ± SEM; larger size total power = 31594,8 ± 296,8, mean ± SEM; Wilcoxon rank-sum test: W = 256, p value < 0.001), the size generalization test (smaller size total power = 27725,4; larger size total power = 29760,3; Figure 4), and when the stimulus size was controlled (i.e., number same size test; 4 elements total power = 27954,02; 8 elements total power = 28851,5; Figure 4). Spatial frequency was similar when the total area was equated between arrays (i.e., number total area test; 4 elements total power = 29716,9; 8 elements total power = 29715,0), while it was inversely related with numerosity when the total contour length was controlled (4 elements total power = 33944,7; 8 elements total power = 31020,3) (∗∗∗p < 0.001).