DIANA A. LIAO HTTPS://ORCID.ORG/0000-0003-0910-9363 , KATHARINA F. BRECHT HTTPS://ORCID.ORG/0000-0001-8500-725X, [...], AND ANDREAS NIEDER HTTPS://ORCID.ORG/0000-0001-6381-0375 +1 authorsAuthors Info & Affiliations
2 Nov 2022
Vol 8, Issue 44
DOI: 10.1126/sciadv.abq3356


Recursion, the process of embedding structures within similar structures, is often considered a foundation of symbolic competence and a uniquely human capability. To understand its evolution, we can study the recursive aptitudes of nonhuman animals. We adopted the behavioral protocol of a recent study demonstrating that humans and nonhuman primates grasp recursion. We presented sequences of bracket pair stimuli (e.g., [ ] and { }) to crows who were instructed to peck at training lists. They were then tested on their ability to transfer center-embedded structure to never-before-seen pairings of brackets. We reveal that crows have recursive capacities; they perform on par with children and even outperform macaques. The crows continued to produce recursive sequences after extending to longer and thus deeper embeddings. These results demonstrate that recursive capabilities are not limited to the primate genealogy and may have occurred separately from or before human symbolic competence in different animal taxa.


Recursion, the cognitive capacity to embed an element structure within others of the same kind, has been claimed as one of the key features of human symbolic competence (1). It has been put forth as what distinguishes human language from all other forms of animal communication (2) to which counter-arguments have been made [see (3)]. Grammatical rules in language use recursion to expand the variety and complexity of possible sentences that could be produced into what is conceptually infinite. In the Chomsky hierarchy of grammars with increasing generative power, center-embedded recursion is said to sit near the top, below context-sensitive (and Turing-complete) grammars. In this grammar, equivalent procedures are embedded in the middle of a sequence. One of the classic example sentences with such a center-embedded structure is “The mouse (A1) the cat (A2) chased (B2) ran (B1)”. Here, the inner clause “the cat (A2) chased (B2)” is embedded within the outer clause “the mouse (A1) [that] ran (B1)”. Such expressions can be formalized as AnBn, a context-free grammar.
To understand the evolution of symbolic skills, there has been considerable interest in investigating whether nonhuman animals can perceive and produce recursive sequences. Several prominent studies have explored differences in animals’ auditory perception of finite-state grammars, i.e., (AB)n, and context-free grammars, i.e., AnBn (46). However, these results have been controversial given that alternative, nonrecursive strategies for task performance have been put forward [e.g., such as discriminating different numbers of syllables; (78)]. The stimuli used (e.g., different syllable types) are not intrinsically paired, such that that A1-B1 is not differentiated from A2-B2. This lack of inherent relationship between stimuli pairs allows for certain exploits that superficially appear like recursive responses but can be explained with simpler phonetic or numerical strategies (911). Therefore, for the demonstration of any context-free grammar, semantic pairing is of paramount importance. However, separate training on the associations between pairs of arbitrary elements could introduce potential bias to test responses once pairs are combined into longer sequences (1213).
A recent study (14) cleverly addressed these issues. They introduced different pairs of colored brackets that are intrinsically linked with an open and closed direction (e.g., [ ], { }, and < > ; Fig. 1, B and C). Using these innovative stimuli in experiments with U.S. adults, Tsimane’ adults, U.S. children, and rhesus monkeys (Macaca mulatta), functionally equivalent training and testing procedures were performed for all groups so that results could be directly compared. Participants were trained to first touch one of the training sequences, e.g., { ( ) } and { [ ] }, in these specific orders. They were then tested on the ability to spontaneously transfer this recursive, center-embedded structure to the novel pairing of bracket sequences, e.g., ( [ ] ) or [ ( ) ]. All humans were able to successfully complete this test, and the monkeys, with additional training, did so as well. These results demonstrate that primates across age, education, culture, and species could all learn to produce basic recursive sequences with nested pairs of bracket stimuli.
Fig. 1. Bracket task design and performance on transfer trials for crows.
(A) Training procedure: Crows were required to peck at the bracket stimuli in a center-embedded sequence order. After initiating a trial, two pairs of brackets appeared simultaneously in random locations on the touchscreen monitor. The crow was required to peck each stimulus in a determined order (depicted here with arrows and numbers) and was rewarded for correct sequence completion. Otherwise, when an incorrect bracket was selected, the screen flashed, an error tone played, and brief time-out was initiated. (B) Training lists that were presented until criterion was reached. After training, these lists were intermixed within a session along with transfer trials that consisted of the inner bracket pairs from the two training lists. (C) Color-coded main response types to the transfer trials—derived from the training lists—which were rewarded regardless of the selected order. (D) Proportion of response types produced by the crows (in saturated colors) as compared to U.S. adults, Tsimane’ adults, U.S. children, and monkeys [in faded colors; (14)]. Response strategies color-coded as in (C). Inset displays results for each crow. Error bars represent the SEM of the population, “*” represents a significant difference (P < 0.05) between the proportion of center-embedded and crossed responses, and ns represents no significant difference between groups. Bracket stimuli modified from (14) @ The Authors, some rights reserved; exclusive licensee The American Association for the Advancement of Science. Distributed under a CC BY-NC 4.0 license (http://creativecommons.org/licenses/by-nc/4.0/).

Because recursive sequence generation applying this bracket protocol has been tested exclusively in primate species so far, the implied assumption is that the ability to track syntactic relationships between elements over distances may have been inherited from a common primate ancestor before reaching its most elaborate and specific expression in humans (15). Inspired by the recent study (14), we replicate and extend the bracket protocol in crows (Corvus corone) to examine recursive behavior through a comparative evolutionary lens. Crows are corvids, a songbird family that exhibit complex cognition such as elaborate tool use (1618), analogical reasoning (19), rule switching (20), and numerical competency (2122). As songbirds (23), their vocal communication skills show interesting parallels with human speech, such as complex acoustic signals, sensitive learning periods, the need for auditory feedback, elaborate vocal production abilities, and social learning (2426). Combined, these traits make crows a promising candidate to search for an understanding of recursive primitives.