Fig. 2.

Generalization. Associative performance indices (PIs) are presented for the generalization task in which animals are trained with an element (X) and tested with a mixture containing it (XY), or vice versa. The numbers denoted at the bottom of each panel refer to the group numbers listed in Table 1. The odours used are listed below the figure. (A) Companion groups of larvae are trained with an element and tested with its double quantity, or vice versa. Performance indices are statistically indistinguishable in all cases (Mann–Whitney U-test: P>0.05/5, N=12 in all cases). (B) Companion groups of larvae are trained with an element and tested with a binary mixture containing it, or vice versa. Performance indices are equal in 19 of 20 of these cases (Mann–Whitney U-test: P>0.05/20 for Groups 11–46 and 49–50) except for 1O and 1O–HA (Group 47 versus 48; Mann–Whitney U-test: P<0.05/20). Sample sizes are 12 in all cases. (A′) Pooled scores of the companion groups from A. A comparison within the data set reveals statistically indistinguishable learnability across odours (Kruskal–Wallis test: P>0.05/2, N=24 in all cases), yielding a baseline of performance indices (dashed grey line represents the median of the pooled data). (B′) Pooled scores of the companion groups from B. The larger the perceptual distance between element and mixture, the less conditioned behaviour one should observe (i.e. the smaller the PI scores should be). Thus, in this task perceptual distance can be approximated by the d_GEN scores, measuring how much smaller PI scores are relative to baseline. A Kruskal–Wallis test (P<0.05/2, N=24 in all cases) reveals that depending on the odours used, perceived distance differs between elements and mixture. Note that because no unambiguous measure of element–mixture similarity could be obtained for the case of 1O and 1O–HA (see B), this stimulus pair cannot be included in this Kruskal–Wallis test. See Fig. 1 legend for odour definitions.