The authors have declared that no competing interests exist.
Conceived and designed the experiments: JMF JLQ. Performed the experiments: JMF EFC LMA. Analyzed the data: JMF SH. Wrote the paper: JMF JLQ SH EFC LMA.
Associative learning is essential for resource acquisition, predator avoidance and reproduction in a wide diversity of species, and is therefore a key target for evolutionary and comparative cognition research. Automated operant devices can greatly enhance the study of associative learning and yet their use has been mainly restricted to laboratory conditions. We developed a portable, weatherproof, battery-operated operant device and conducted the first fully automated colour-associative learning experiment using free-ranging individuals in the wild. We used the device to run a colour discrimination task in a monitored population of tits (Paridae). Over two winter months, 80 individuals from four species recorded a total of 5,128 trials. Great tits (
Cognition–the processes by which animals collect, store and use information from their environment [
Associative learning—a behavioural modification following reinforcement, based on associations between stimuli, or between stimuli and responses–has been found in all bilateral animals tested to date [
Several devices and protocols for studying learning and memory have been used in the field, including artificial flowers [
Here we report on a new portable operant research device that can be used in both field and captive conditions. The device functions as an “inverted”, free operant conditioning chamber, in which subjects that visit voluntarily can proceed to learning trials via an automated process. This approach builds upon earlier work on learning in flocks of free-ranging pigeons [
This presentation of a free-access cognitive testing device in natural settings could however lead to sampling bias, if some individuals are preferentially visiting and using the device [
Empirical evidence for the existence of individual differences in cognitive performance in various cognitive domains has accumulated in the past years [
Here we examine the link between problem-solving performance at a novel foraging task in captivity [
Our main goals were therefore to; (1) develop and test an automated operant device and protocol in field conditions; (2) determine any species, sex, age class, or exploration biases in the population of individuals that landed at the devices and recorded trials (i.e. “sampling biases”); (3) compare learning rates of successful and unsuccessful problem-solvers; and (4) to examine individual determinants (sex, age, and exploration score) of learning rate in the wild. We used automated data outputs to examine the sequence of choices made by marked individuals in a simple colour association test. The learning test consisted of a simultaneous discrimination procedure, in which pecking the red key once was rewarded 100% of the time (fixed ratio 1 schedule), and pecking the yellow or green keys was never rewarded. The position of colour cues was varied pseudo-randomly between trials to ensure that birds learnt colour rather than spatial position as a predictor of reward.
The study took place in Wytham Woods, Oxfordshire, UK, from 5 January to 3 March 2012. In the context of another study targeting mixed-species flocks of tits in Wytham, great, blue (
Species, sex and age composition of the overall population was extracted from a grid of 65 sunflower feeders fitted with RFID antennae that collected spatio-temporal flocking data throughout Wytham woods from December 2011 to March 2012 [
We used five identical operant devices for this experiment (
The weatherproof external casing is made of opaque Perspex, while the three response keys and the section of the front panel comprising the feeding hole are transparent. The perch (bottom of the front panel) is equipped with an antenna that relays information about the unique PIT-tag combination of marked individuals to the printed circuit board located within the device.
To elicit pecking at the keys without having to shape this response through successive steps (e.g. [
The identity of PIT-tagged individuals landing at the device was relayed to the PCB by an antenna located in the perch (Dorset Identification, NL). This perch was a horizontal plane of 10cm x 5cm x 1cm and was small enough to hold only one bird at once, but large enough to give access to all three keys on the front panel. The whole system for each device was powered by a 12V sealed lead-acid battery; a 12Ah battery was enough to power a device for at least three days in winter conditions. When the device was powered on (i.e. almost continuously over the experimental period for each location), all keys were illuminated and the timing of visits and PIT combination, as well as the timing and colour choice for each peck, were recorded automatically on an output file stored on a SD card. A visit recorded the time when a PIT-tagged bird landed on the device, whether it pecked at the keys or not (“visit”). A “trial” was defined as a key peck when the keys would be illuminated with colours and a choice had thus been made (see
Operant devices were deployed at four locations across Wytham Woods. Sites featured an open area with nearby tree cover from which the birds could come and perch on the devices. We first baited each location for several days with a regular multi-access feeder containing sunflower seeds, before installing either two (N = 2 locations) or three operant devices (N = 2 locations) 1m from tree cover, 3 to 8 m from each other, and placing a few mealworms on the roof of each device. We used two to three devices per location to reduce monopolisation of devices by dominant individuals. We then activated the operant devices (see below) and allowed the sunflower feeder to be depleted, refilling it bi-weekly until birds started to peck regularly at the operant devices. Devices were left in place for 26, 18, 21 and 11 days, respectively at each location; devices were removed when depletion rate became high, indicating that no new birds were likely to be given the chance to learn. During this period we changed batteries every 3 days and refilled rewards within devices when they were depleted (no more than twice per day).
We examined species, sex, age class, problem-solving performance and exploratory personality biases among birds that landed at the devices at least once, or registered at least one trial (key peck). The proportion of different species, sexes, age classes, and of solvers (vs. non-solvers) visiting the devices and recording trials were compared with the proportion of birds not visiting the devices or recording trials in the wider Wytham woods population using a Chi-square test with Yates’s continuity correction. Mean exploration score in these groups was compared using two-group t-tests.
We assessed colour-based associative learning by testing for improvement in colour choice accuracy over successive trials by individual birds. Because the number of trials was different for all birds, we scaled trial number by the highest number of trials recorded by any individual (n = 700 trials). We used generalised linear mixed models (GLMM, function “glmer”) with correct (red = 1) vs. incorrect (yellow, green = 0) choice as the response variable, and individual identity as a random intercept. A positive and significant fixed term for trial number would be evidence for learning of the colour-based association. A significant interaction between the random term for ID and trial number (i.e. significant random slope) would be evidence for individual differences in learning rates, and was assessed using the log-likelihood ratio test (LRT) [
Work was subject to review by the Department of Zoology ethical committee, University of Oxford. All work adhered to UK standard requirements and was carried out under Natural England licence 20114732. Field work took place in Wytham Woods (lat 51°46’N, long 1°20’W), private land that belongs to the University of Oxford; for permission contact the Conservator, Nigel Fisher. No endangered or protected species were involved in the study.
We recorded 7,480 visits to the operant devices by 144 PIT-tagged birds (Table A in
Statistics are shown for individuals that visited devices at least once (Visited), and for those that had pecked at least one key (Pecked).
Blue tit | 1534 (54.2%) | 43 (29.9%) | 8 (10.0%) |
Coal tit | 103 (3.6%) | 3 (2.1%) | 1 (1.3%) |
Great tit | 1061 (37.5%) | 88 (61.1%) | 67 (83.8%) |
Marsh tit | 134 (4.7%) | 10 (6.9%) | 4 (5.0%) |
Totals | 2832 (100%) | 144 (100%) | 80 (100%) |
The sex ratio of great tits having recorded at least one visit and at least one trial was not significantly different from the sex ratio in the wintering population (
Statistics are shown for individuals that visited devices at least once (Visited), and for those that had pecked at least one key (Pecked). Individuals that had been marked as chicks and not sexed
Female | 478 (50.4%) | 43 (51.2%) | 34 (51.5%) |
Male | 471 (49.6%) | 41 (48.8%) | 32 (48.5%) |
Totals | 949 (100%) | 84 (100%) | 66 (100%) |
Statistics are shown for individuals that visited devices at least once (Visited), and for those that had pecked at least one key (Pecked). Individuals not recorded on the regular grid of feeders (N = 1; see
Juvenile | 501 (49.4%) | 58 (66.7%) | 46 (68.5%) |
Adult | 514 (50.6%) | 29 (33.3%) | 21 (31.3%) |
Totals | 1014 (100%) | 87 (100%) | 67 (100%) |
Individual great tits (N = 67) improved their colour choice accuracy over successive trials in the field (trial number: Χ2 = 563.2, d.f. = 1, p < 0.001; see
One panel per individual (N = 16 birds; only individuals having recorded 50 trials or more are shown).
(a) non-solvers vs. solvers; (b) adults vs. juveniles; (c) females vs. males; and (d) fast explorers vs. slow explorers. Note that exploration score was analysed as a continuous variable but was split into 50% slower birds vs. 50% fastest explorers for illustration purposes.
We developed a portable, automated operant device and tested a research protocol for the measurement of associative learning in free-ranging songbirds. Wild parids landed on the free-operant device, and great tits could learn to obtain food rewards by pecking at the red-lit key, but not at the green or yellow-lit keys. Our devices were therefore successful in administering learning trials without the need to shape or train individuals before test trials, or requiring an experimenter to be present during the trials. This is, to our knowledge, the first time a fully-automated colour-associative learning test recording individual behaviour has been administered to free-ranging animals.
The experimental protocol was successful in attracting a large number of free-ranging individuals to the devices: 144 PIT-tagged birds visited the devices within two months of automated field observations. All species were able to register trials at the devices but only great tits proved to be a useful model species for this type of study at Wytham, with 76% of individuals visiting the devices registering at least one trial. It is possible that species differences arose from differences in body size, and that the ca. 100% larger great tit [
The sex ratio and mean exploration score of individual great tits using the devices did not differ significantly from the rest of the tagged population, suggesting that the experimental sample was not biased in these respects. It might have been expected that fast explorers should be more likely to encounter the devices due to increased activity rate [
Individual great tits improved gradually over trials in their choice accuracy, providing evidence for colour-based discrimination occurring in the wild. Individuals differed significantly in the rate at which they learned this colour-based association in the field, even when controlling for site and inter-trial interval duration. Birds that were successful at a novel foraging problem (tested in isolation, in captivity) learned the colour association faster than non-solvers once released in the field, which could suggest that individual differences in general associative learning abilities expressed in the novel problem-solving situation and in this colour-based learning test can be stable over time and consistent across contexts. Positive correlations were also observed between innovative problem solving performance and efficiency in motor learning [
Juveniles were more likely to use the devices and showed a marginally non-significant tendency to learn faster than adults. Faster improvement over repeated trials in juveniles cannot be explained by potential differences in temporal patterns of device use because all analyses controlled for inter-trial interval duration. Interestingly, this effect is similar to more rapid learning by young versus older baboons (
Because trials were not run on individuals in isolation, birds could potentially have learned to peck at the red key by observing successful conspecifics interacting with the device [
In conclusion, our operant device has the advantages of being portable, affordable (overall unit price around £700), and capable of automatically running and recording large numbers of trials (up to 698 trials daily per location in this experiment) in individual-based learning assays, with relatively little human intervention in the field. However not all species seem to be responsive to this experimental set-up, with great tits being the only species among the four tit species in Wytham to show participation rates >50%. Because the response keys can be replaced by any other mechanical object that can break the optical beam located behind the front panel, experimenters could potentially use the device with a range of avian or non-avian species, making it a useful tool in comparative cognition research between or within species. The device can be customised by users to administer several types of rewards and tests; for instance multiple schedules of reinforcement can be programmed for the different keys, and reversing the contingencies attributed to each colour would assess reversal learning speed (e.g. [
(PDF)
Total number of visits to the devices, proportion of these visits leading to at least one trial (i.e. key-peck), number of individuals with at least one trial recorded in the field, along with range (minimum-maximum), median and mean ± standard error (s.e.) for total number of trials per individual, for each of the four species (Table A in S2 Text). Results of a binomial GLMM for correct (red) vs. incorrect (green, yellow) choices over successive trials, including a random intercept for individual identity and a random slope for identity over scaled trial number (N = 3470 trials by 21 individuals). This model examines differences in learning slopes over trials in successful problem-solvers vs. non-solvers (Table B in S2 Text). Results of a binomial GLMM for correct (red) vs. incorrect (green, yellow) choices over successive trials, including a random intercept for individual identity and a random slope for identity over scaled trial number (N = 3470 trials by 21 individuals). This model examines differences in learning slopes over trials in adults versus juveniles (Table C in S2 Text). Results of a binomial GLMM for correct (red) vs. incorrect (green, yellow) choices over successive trials, including a random intercept for individual identity and a random slope for identity over scaled trial number (n = 3470 trials by 21 individuals). This model examines differences in learning slopes over trials in females vs. males (Table D in S2 Text). Results of a binomial GLMM for correct (red) vs. incorrect (green, yellow) choices over successive trials, including a random intercept for individual identity and a random slope for identity over scaled trial number (n = 3470 trials by 21 individuals). This model examines differences in learning slopes over trials in relation to exploration score Table E in S2 Text).
(PDF)
(PDF)
(WMV)
(WMV)
(WMV)
Great tits present in the population in the winter 2012.
(XLSX)
Trials recorded by great tits on the operant boxes.
(XLSX)
We would like to thank everyone involved in marking birds in Wytham. Special thanks to Adam Hoffman for allowing us to use modified versions of his ELOPTA device for pre-trials, and to Martin Whitaker (Stickman Technologies Inc., Southampton, UK), Eric Trottier, John Hogg and Tony Price (Zoology and Experimental Psychology Workshop, University of Oxford, UK) for help with design and production of the final devices. We would also like to thank Alex Kacelnik for useful discussions, Louis Lefebvre and the reviewers for commenting on an earlier version of the manuscript. This work was funded by a Natural Environment Research Council grant (NE/I017208/1) to JLQ, and EFC was funded by a Leverhulme grant (RPG-265) to JLQ. JMF was funded by a NERC grant (NE/I017208/1) and a Natural Sciences and Engineering Research Council Discovery grant (435596–2013). JMF designed the device and study with input from JLQ. EC provided problem-solving and some personality data. LMA provided personality data. JMF and SH conducted all of the analyses. JMF wrote the paper and all authors commented on the paper. We are grateful for infrastructure support from a grant from the ERC (AdG 250164) to Ben Sheldon.