Superior abstract-concept learning by Clark's nutcrackers (Nucifraga columbiana)

Abstract

The ability to learn abstract relational concepts is fundamental to higher level cognition. In contrast to item-specific concepts (e.g. pictures containing trees versus pictures containing cars), abstract relational concepts are not bound to particular stimulus features, but instead involve the relationship between stimuli and therefore may be extrapolated to novel stimuli. Previous research investigating the same/different abstract concept has suggested that primates might be specially adapted to extract relations among items and would require fewer exemplars of a rule to learn an abstract concept than non-primate species. We assessed abstract-concept learning in an avian species, Clark's nutcracker (Nucifraga columbiana), using a small number of exemplars (eight pairs of the same rule, and 56 pairs of the different rule) identical to that previously used to compare rhesus monkeys, capuchin monkeys and pigeons. Nutcrackers as a group (N = 9) showed more novel stimulus transfer than any previous species tested with this small number of exemplars. Two nutcrackers showed full concept learning and four more showed transfer considerably above chance performance, indicating partial concept learning. These results show that the Clark's nutcracker, a corvid species well known for its amazing feats of spatial memory, learns the same/different abstract concept better than any non-human species (including non-human primates) yet tested on this same task.

1. Background

The ability to learn abstract relational concepts provides a fundamental building block for many cognitive functions. In contrast to item-specific, stimulus-bound associative learning, abstract concepts transcend the specific identity of the objects involved, allowing relations among objects to be extrapolated to larger domains of novel objects—a relational abstract concept [1]. Research investigating non-human abstract-concept learning has often focused on the same/different task, because it has long been recognized, at least in humans, that ‘the mind makes continual use of the notion of sameness’ [2]. If the abstract concept of sameness is so fundamental to human cognition, then it should be expected to be a property of cognition generally, possibly with some aspects varying across species in accordance with ecological demands and brain development. To this end, we studied same/different abstract-concept learning in Clark's nutcracker (Nucifraga columbiana), a corvid with strong reliance on spatial memory for food caching and superior object–location relational learning [3–5], but average object–object relational learning (in delayed non-matching to sample tasks), compared with other corvids [5].

We used the popular same/different task where subjects classify pairs of pictures as ‘same’ or ‘different’. The gold standard for abstract-concept learning is transfer to novel stimuli: following learning with a stimulus set, we tested for transfer on 10% of the trials with novel (never seen before, never repeated) stimuli [6]. If subjects were responding on the basis of the abstract relationship (same versus different) between stimuli that make up each pair (as opposed to specific features of the stimuli), then transfer performance to novel stimuli should be statistically equivalent to performance with trained stimuli.

Previous transfer tests in this same/different task with rhesus monkeys [7], capuchin monkeys [8] and pigeons [6] showed no significant transfer, and hence no abstract-concept learning, following training with the initial set of these same eight picture stimuli. Only human participants showed transfer of same/different abstract-concept learning following training with this set of eight stimuli [9]. Although apes have not been tested with these picture stimuli, it has been hypothesized that the apes' same/different concept learning (as a function of training set size expansion) would likely be superior to that for rhesus or capuchin monkeys [10, fig. 20] based on rapid concept learning of matching by chimpanzees with real-world objects [11]. Nonetheless, we show here that an avian species, the Clark's nutcracker, can transfer abstract-concept learning following similar training with this same initial set of eight stimuli.

2. Material and methods

(a). Same/different training

We trained nine Clark's nutcrackers in a 2-item same/different task with the same eight stimuli (yielding eight possible same trials, 56 possible different trials), the same procedures and the same acquisition/testing criteria previously used with monkeys and pigeons [6–8,10]. Briefly, trials began with the presentation of a sample picture that required a fixed number of responses (beginning at 1 and systematically increased up to 20 during acquisition) before the comparison picture was presented. The comparison picture was shown beneath the sample picture along with a white rectangle to the right of the comparison picture. When the sample picture and comparison picture were the same, responses to the picture were rewarded. When the sample picture and comparison picture were different, responses to the white rectangle were rewarded. As with the other species, nutcrackers were trained on 100-trial sessions (50 same, 50 different trials) until they achieved 80% accuracy on three consecutive sessions (average acquisition: 33 sessions, range: 23–48).

(b). Transfer testing

Following acquisition, abstract-concept learning was assessed in six transfer sessions, each session containing 90 baseline (training) trials and 10 novel (five same, five different) transfer trials. The transfer stimuli were never repeated across trials to ensure no item-specific learning could benefit transfer performance. Transfer trials were reinforced identically to baseline trials [6–8,10]. We compared mean per cent correct between baseline and novel transfer trials across the six transfer sessions using individual paired t-tests. If the t-test rejected the hypothesis of equality between the means, we next tested the transfer accuracy for the six sessions against chance using a one-sample t-test.

3. Results

Figure 1 shows mean accuracy for each bird for baseline and transfer trials, grouped by transfer performance. Two nutcrackers (R and S) showed full transfer with no significant difference between baseline and transfer accuracy (both ts(5) ≤ 2.1; ps > 0.09). The seven other nutcrackers showed significantly lower transfer accuracy than baseline accuracy (all ts ≥ 3.89; all ps ≤ 0.01). Nevertheless, four of those seven nutcrackers (L, B, G and H) showed transfer significantly above 50% (chance level) and hence partial abstract-concept learning (all ts ≥ 3.87; ps ≤ 0.01). The remaining three nutcrackers' (K, F and T) transfer accuracies were not different from chance (50%) performance (all ts ≤ 1.9; ps ≥ 0.11), indicating no evidence of partial abstract-concept learning. There were no response biases or significant accuracy trends during transfer testing (see the electronic supplementary material for analyses).

Figure 1.

Mean per cent correct during transfer testing for each nutcracker on baseline and transfer trials. Error bars are 1 s.e. of the mean. The horizontally separated groups mark differences in abstract-concept learning: R and S showed full concept learning; L, B, G and H showed transfer significantly greater than chance; K, F and T showed transfer not greater than chance. (Online version in colour.)

4. Comparison with other species' same/different results

Clark's nutcrackers are the first non-human animal species we have tested that shows any transfer of a same/different abstract concept following training with eight stimuli. Figure 2 compares mean performance of the nine nutcrackers with two monkey species (four rhesus and three capuchin monkeys) and another avian species (nine pigeons). A one-way ANOVA across species confirmed the significant species effect (F_3,21 = 9.81, p < 0.001). The nutcrackers' overall accuracy (67.0%; standard error of the mean, s.e.m. = 3.3%) was greater than that of any other species tested in this same task (mean difference with rhesus: 16%, Tukey HSD adjusted p = 0.008; with capuchins: 15%, p = 0.02; with pigeons: 17%, p < 0.001). Mean transfer accuracy was not significantly different (all ps ≥ 0.99) among rhesus (mean = 51%, s.e.m. = 4%), capuchins (mean = 52%; s.e.m. = 2%) and pigeons (mean = 50%, s.e.m. = 1%). Additional across-species analyses showed no differences during acquisition that might account for these differences in transfer (see electronic supplementary material for analyses).

Figure 2.

Mean per cent correct during transfer testing on transfer trials only for individuals of different species: Clark's nutcracker, rhesus monkey, capuchin monkey and pigeon. Error bars are 1 s.e. of the mean. Solid horizontal bars are group means. Grey dashed lines indicate the acquisition criterion required for transfer testing (80% correct) and the expected per cent correct under chance performance (50% correct). (Online version in colour.)

Although only two nutcrackers fully learned the same/different abstract concept following training with the eight stimuli, these other species required several training set expansions (doublings) before any individual of that species demonstrated full concept learning. The first rhesus monkey, pigeon and capuchin monkey that showed full concept learning did so following training with 32-, 64- and 128-item sets, respectively [10]. Nevertheless, other individuals of these species eventually showed full concept learning following additional training set expansions. It is important to emphasize that abstract-concept learning was a function of increased number of exemplars of the rule, and not simply further training or transfer testing in the same/different task. Pigeon control groups matched for amount of training and/or transfer testing, but without training set expansion (i.e. all with the same initial eight picture set), showed no transfer despite as many as 15 400 trials of training and seven novel-stimulus transfer tests [6].

5. Discussion

We studied Clark's nutcrackers because of their superior spatial memory [12,13] even when compared with other caching corvid species [3,5]. In their natural environment, nutcrackers cache tens of thousands of pine nuts in the autumn of each year and then recover them throughout the winter and spring when they are rearing chicks and alternative food sources are scarce [14,15]. The nutcracker's superior food-caching abilities have been thought to be the basis for their superior learning of objects in a specific hierarchy (transitive inference) or objects in specific locations in laboratory tests [12,16,17].

Notwithstanding superior learning of specific objects in specific locations (item-specific object–location memory), abstract object–object concept learning is different. Object–location memory entails unique cache-site locations, making such learning specific to those locations (i.e. item specific), not abstract. Nonetheless, the results presented here show that Clark's nutcrackers have superior ability in learning an object–object same/different abstract concept because they learn with fewer exemplars of the rule than other species.

Our study and findings may help explain why a previous study did not find superior object–object learning by nutcrackers [5]. That study used only two stimuli, ruling out abstract-concept learning, and further, species comparisons were based on learning rates. In this study, we found no superior learning rate for nutcrackers compared with monkeys or pigeons. Superior learning by nutcrackers was shown to be due to their ability to learn the abstract concept, which can be determined only by transfer to novel stimuli.

Over the past 30 years, we and other researchers have developed techniques to assess the ability of a range of diverse species to learn abstract same/different concepts [6–8,16–19]. Findings from many of these experiments have focused on when during set size expansion a sufficient number of rule exemplars have been learned to produce full concept learning, rather than attempting to identify which species can or cannot learn abstract concepts [20–22]. The critical issue of determining the number of exemplars necessary for learning relational rules and accurately applying them in novel situations is a common goal of all transfer of learning and indeed of education generally [23–25]. The number of training exemplars required for concept learning provides a straightforward, quantitative measure of concept learning useful in comparing species differing in their biological makeup and ecological requirements, as well as individual differences within species.

Ethics statement

All procedures were approved by the University of Manitoba's Committee on Animal Care and were accordance with the Canadian Council on Animal Care.

Data accessibility

The nutcracker data reported in this study are freely available from Openwetware.org: http://openwetware.org/wiki/Beauchamp:CompCog.

Funding statement

This research was supported by a Natural Sciences and Engineering Research Council of Canada (NSERC) grant to D.M.K.

Authors' contributions

J.F.M. programmed the experiments, analysed the data and prepared the first draft of the manuscript. D.M.K. was responsible for data collection. All authors participated in the design of the study, the interpretation of its results and contributed to the final version of the manuscript. All authors gave final approval for publication and agree to be accountable for all aspects of the work.

Competing interests

The authors declare no competing interests.