SCIENCE ADVANCES | RESEARCH ARTICLE
test, 184 of 253, P < 0.001). The Frequency of response patterns was not equally distributed ⌊chi-square test of goodness of fit: χ2 (df = 2, N = 205) = 293.86, P < 0.001⌋. Again, the crows produced significantly less crossed responses than center-embedded responses [13 versus 184, chi-square test of goodness of fit: χ2 (df = 1, N = 197) = 184.43, P < 0.001; significant at Bonferroni-corrected α = 0.05/2⌉. Recursive responses for both crows in experiment 2 were robust throughout the course of testing (table S2). Thus, the results of experiment 2 mirror and corroborate the results from experiment 1.
With only two pairs used in transfer trials, we had two potential response orders (Figs. 1C and 2B) that both fulfill a center-embedded response structure (fig. S2). Notably, there was a significant association between order and experiment ⌊chi-square test of independence: χ2 (df = 1, N = 265) = 74.76, P < 0.001⌋: In experiment 1, where both inner pairs were presented together, we found that there was no significant difference in the frequency with which the orders were produced ⌊(⌊⌋): 47 of 80, ⌊()⌋: 33 of 80; chi-square test of goodness of fit: χ2 (df = 1, N = 80) = 2.45, P < 0.118; Fig. 2B⌉. However, in experiment 2, where one ourter pair "< >" and one inner pair "( )" were presented together, the crows consistently rpeferred responding in this bounded manner ⌈< () >: 182 of 185, ( < > ): 3 of 185: chi-square test of goodness of fit: χ2 (df = 1, N = 185) = 176.09, P < 0.001, significant at Bonferroni-corrected α = 0.05/3; Fig. 2B⌉. This result suggests that the crows may be sensitive to the hierarchical bounded structure of the center-embedded sequences.
