Concurrent recall of serially learned visual discrimination problems in dwarf goats (Capra hircus)

Behavioural Processes 79 (2008) 156–164

Contents lists available at ScienceDirect

Behavioural Processes journal homepage: www.elsevier.com/locate/behavproc

Concurrent recall of serially learned visual discrimination problems in dwarf goats (Capra hircus) J. Langbein a,∗ , K. Siebert a , G. Nuernberg b a b

Research Unit Behavioural Physiology, Research Institute for the Biology of Farm Animals, D-18196 Dummerstorf, Germany Research Unit Genetics & Biometry, Research Institute for the Biology of Farm Animals, D-18196 Dummerstorf, Germany

a r t i c l e

i n f o

Article history: Received 28 February 2008 Received in revised form 7 July 2008 Accepted 10 July 2008 Keywords: Dwarf goat Intermediate feeder Visual discrimination Serial learning Concurrent recall Shape similarity

a b s t r a c t Studies of cognitive ability in farm animals are valuable, not only because they provide indicators of the commonality of comparative influence, but understanding farm animal cognition may also aid in management and treatment procedures. Here, eight dwarf goats (Capra hircus) learned a series of 10 visual four-choice discriminations using an automated device that allowed individual ad lib. access to the test setup while staying in a familiar environment and normal social setting. The animals were trained on each problem for 5 days, followed by concurrent testing of the current against the previous problem. Once all 10 problems had been learned, they were tested concurrently over the course of 9 days. In initial training, all goats achieved criterion learning levels on nearly all problems within 2 days and under 200 trials. Concurrently presenting the problems trained in adjacent sessions did not impair performance on either problem relative to single-problem learning. Upon concurrent presentation of all 10 previously learned problems, at least half were well-remembered immediately. Although this test revealed a recency effect (later problems were better remembered), many early-learned problems were also well-retained, and 10-item relearning was quite quick. These results show that dwarf goats can retain multiple-problem information proficiently and can do so over periods of several weeks. From an ecological point of view, the ability to form numerous associations between visual cues offered by specific plants and food quality is an important pre-grazing mechanism that helps goats exploit variation in vegetation and graze selectively. © 2008 Elsevier B.V. All rights reserved.

1. Introduction Mammalian species may be expected to have basic learning and problem-solving methods in common, either as a result of shared evolutionary history or as a result of common adaptive processes to the structure of nature (Macphail and Bolhuis, 2001). Although a variety of experimental tasks have been developed to assess learning and memory function in animals, comparable interspecies tests have not been easy to devise. Multiple-problem discrimination procedures like serial exposure to successive problems have provided valuable approaches for investigating basic learning principles (e.g., the ‘learning to learn’ phenomenon) from a comparative perspective (Thomas, 1986). Furthermore, learning performance in multiple-problem discrimination can serve as a measure for acquisition and retention in comparative stud-

∗ Corresponding author at: Forschungsinstitut für die Biologie landwirtschaftlicher Nutztiere, Forschungsbereich Verhaltensphysiologie, WilhelmStahl-Allee 2, D-18196 Dummerstorf, Germany. Tel.: +49 38208 68814; fax: +49 38208 68802. E-mail address: [email protected] (J. Langbein). 0376-6357/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.beproc.2008.07.004

ies (Santucci and Treichler, 1990; Treichler et al., 1977). In this approach, which has been referred to as ‘concurrent discrimination’ (Hayes et al., 1953), the trials on a number of different problems are intermixed, either during problem acquisition or during recall of previously learned problems (Thomas, 1996). Applying this approach, animals’ capacity has been tested for retention of information about a large number of object-discriminations over longer time periods (Bakner and Treichler, 1993; Nakagawa, 1992; Treichler, 1984). In farm animals, most tests of performance in serial (one after the other) or concurrent (at the same time) multiple-problem discrimination have been carried out on horses (Murphy and Arkins, 2007; Nicol, 2002). Various studies reported a significant reduction in the number of trials horses needed to reach a given learning criterion in successive problems, indicating that horses have the ability to use previously learned information to facilitate subsequent learning (Fiske and Potter, 1979; Sappington and Goldman, 1994). Recently, similar results were achieved in dwarf goats (Langbein et al., 2007a). Furthermore, horses, like other Equidae, performed quite well in concurrent discriminations (Thomas, 1986). Conducting similar studies on the cognitive capacity of other domestic animal species is important for various reasons. Cognitive

J. Langbein et al. / Behavioural Processes 79 (2008) 156–164

processes like learning and memory allow animals to display complex adaptive behavior in dynamic environments (Toates, 2004). The behavior of grazing animals is greatly influenced by the way they perceive, process, and memorize information from their environment. Observation of domestic goats under laboratory conditions clearly indicates that they possess excellent vision and readily respond to visual stimuli (Baldwin, 1979; Blakeman and Friend, 1986). Good vision greatly increases wild goats’ chance of survival as they are preyed upon by carnivores and birds of prey in the wild. Furthermore, habitats preferred by feral goats include mountain areas with frequent rocky outcrops and abundant scrubs to use as food and protection (Bullock, 1985). Additionally, goats live in large, complex social groups with a strong linear hierarchy (Langbein and Puppe, 2004), so good visual perception and learning skills would be expected as they are prerequisites for social recognition, as has already been shown in sheep (Kendrick et al., 2001). Feral and wild goats show a highly selective feeding behavior compared to other domestic ungulates (Aldezabal and Garin, 2000). They are able to learn very quickly about spatial and temporal variation of preferred plant species (Provenza et al., 1994). A broader understanding of visual learning abilities, and the memory capacity and stability of goats can help explain the development and stability of learned food preferences in this species. In farm animal species, a lack of information about learning abilities and memory, or misconceptions about how they learn and adapt their behavior, can result in mismanagement and mistreatment (Held et al., 2002). Another reason for studying learning and memory in farm animals is to evaluate theories relevant to ethical concerns about animal welfare (Croney et al., 2004). Furthermore, understanding the learning flexibility and memory capacity of farm animals is a prerequisite for the future design of speciesappropriate devices for cognitive enrichment of housing facilities. As already applied in zoos and recently discussed for farm animals (Bloomsmith et al., 2007; Carlstead and Shepherdson, 2000; Melfi and Thomas, 2005; Swaisgood et al., 2001; Watson et al., 1999), these devices provide long lasting positive effects by combining mental stimulation and a rewarding outcome to facilitate successful coping (Langbein et al., 2004; Puppe et al., 2007). To keep cognitive tasks challenging, they have to be modified and updated regularly, e.g., by varying the rewarded cue (Meehan and Mench, 2007), wherefore knowledge about memory capacity and learning flexibility is essential. Until now, the vast majority of learning studies have routinely involved training single individuals to perform a limited number of trials per day, organized in separate sessions while separated from their social group and normal housing. However, both separationrelated stress and changing the context between the periods of acquisition and retention can impair learning performance (Mendl, 1999; Sondergaard and Ladewig, 2004; Thomas et al., 1985). With this study on serial as well as concurrent multiple-problem discrimination in group-housed dwarf goats, we wanted to broaden the insight into cognitive abilities of animals such as learning flexibility (replacing former cues by new cues), memory capacity (number of cues which can be stored at any one time), and retention time (for how long several cues can be concurrently recalled) under non-laboratory and more naturalistic conditions. We used an experimental approach where the animals could learn individually while remaining in their familiar environment and normal social settings. This was achieved by integrating a fully automated learning device into the animals’ home pen. With this experimental design, we wanted to overcome the restrictions of previous learning studies as described above. As the learning device was accessible all day, the animals could decide themselves when to learn and for how long at all stages of the experiment. There was no restriction

157

on individuals with regard to the number of visits to or trials at the learning device or with regard to the overall amount of reward an animal could gain. Furthermore, with such a learning device, we were able to test the individual learning behavior of a large number of animals simultaneously. 2. Animals, materials and methods 2.1. Animals and housing The subjects of this study were eight female Nigerian dwarf goats (Capra hircus, mean age 132 days at the start of the experiment), from a line bred at our institute for over 10 years. The animals were group-housed in an indoor pen (12 m2 ), which contained straw litter, a wooden two-floor climbing rack, a hayrack (hay ad lib.), and a round feeder to deliver concentrate (300 g per day/animal). The learning device was installed in a separate compartment inside the pen. The device was accessible to all animals 24 h a day, but only one animal could enter the compartment at any one time. Drinking water was available only as a reward for a correct trial at the learning device (see Section 2.6). All animals wore a collar with a responder for individual recognition (Urban, Germany). 2.2. Learning device The computer-controlled learning device (Fig. 1) has been described in more detail in previous studies (Langbein et al., 2004, 2006). Applying a four-choice design, we presented the goats with different sets of black shapes (discrimination problems), each with one S+ (positive/rewarded) and three different S− (negative/unrewarded) stimuli, on a white 38 cm thin film transistor screen (TFT). The screen was protected by a transparent acrylic pane on which four press buttons were mounted. To get a water reward (30 ml), the goats had to discriminate the shape that was predetermined to be the S+ and press the associated button. The size of single shapes was 7 cm2 on the screen with a VGA resolution of 640 × 480 pixels. All shapes used were symbols taken from the symbol gallery in COREL 8.0. Each trial was followed by an inter-trial interval (ITI) of 6 s of a black screen, before the shapes were shown rearranged in the next trial. The arrangement of the shapes in consecutive trials followed a pseudorandom series. This series consisted of two different subsets of all 24 possible pattern combinations. By using this series, we ensured that the S+ , as well as the three unrewarded shapes (S−1 , S−2 , and S−3 ), were normally distributed in the four

Fig. 1. Sketch and measures (in mm) of the compartment containing the learning device: (1) antenna for individual identification, (2) head gate, (3) water bowl for reward delivery, (4) computer screen for shape presentation (protected by an acrylic pane), and (5) press buttons to choose a shape (mounted on the acrylic pane). The compartment is closed surrounded by opaque walls to avoid observational learning.

158

J. Langbein et al. / Behavioural Processes 79 (2008) 156–164

positions and that S+ was not in the identical position in consecutive trials. Different discrimination problems were presented following different pseudorandom series. The controlling software of the learning device was designed to counteract any side or position preferences that might begin to develop (Langbein et al., 2007b). Apart from the physical restriction that allowed only one goat to enter the compartment at a time, all goats had unlimited access to the device and were allowed to perform an unlimited numbers of trials. Throughout all sections of the study, for all individual visits to the learning device and all button-presses, the following data were recorded: (a) individual, (b) date and time of the day, (c) current state of the screen (shapes or ITI) and pattern combination, (d) chosen shape (S+ or S−1 , S−2 , and S−3 ), and (e) currently involved subroutine (standard plan or plan for correction of side preferences). 2.3. Pre-training Over a period of 8 weeks, starting from weaning at an age of 6 weeks, the goats were pre-trained stepwise at the learning device to press different buttons in order to get a portion of drinking water. Subsequently, their first visual discrimination problem (P00) was with one S+ (a circle) and three identical S− (blank sector). When all goats had learned to discriminate S+ reliably, the goats were given two further discrimination problems consecutively (P01 and P02, see Fig. 2) each with one S+ and three different S− . Each of the three problems was run for 14 days. From a previous study, we knew that dwarf goats gradually improve daily learning performance on consecutive discrimination problems (Langbein et al., 2007a). Analysis of the learning curves of P00 to P02 demonstrated the same effect in this study. All animals fulfilled the learning criterion (see Section 2.5) on day 9 in P00, and on day 4 in P02.

2.4. Training and recall of discrimination problems The goats were trained on 10 different, consecutive, four-choice discrimination problems (P1 to P10, see Fig. 2). Reflecting the results from pre-training, each problem ran for 5 days during the training phase. New problems were always started at 10 a.m. To check for preferences for single shapes, for each problem, we conducted a 1-day pre-test with all trials rewarded equally. P1 was directly followed by P2. Subsequent to P2, we tested performance during concurrent recall of the current problem (P2) and the previous problem (P1) when presented in a mixed series over a period of 24 h. This series was composed of two sets of all 24 pattern combinations of each problem. The sets were mixed following a pseudorandom order different from those used during initial training (see Section 2.2). Next, P3 was offered to the animals in the same way as P1 and P2, and subsequent to P3, performance during concurrent recall of the last two problems was tested for 24 h (P3 vs. P2). The same procedure was repeated for all following problems (P4 to P10). Finally, all 10 discrimination problems (P1 to P10) were presented together in a mixed series for 9 days. This series consisted of four different pattern combinations of each problem such that for each problem, S+ was presented once in each possible position on the screen. The complete experimental time schedule is given in Fig. 3. 2.5. Data analyses and statistical methods Because single goats were separated from the rest of the group and undisturbed while acting at the learning device, and because the compartment with the device was surrounded by opaque walls to avoid any form of social learning, we treated individuals within the group as independent replicates for statistical purposes (Langbein et al., 2006).

Fig. 2. Sets of 2D-shapes presented as four-choice discrimination problems during pre-training (P01 and P02) and during the learning experiment (P1 to P10). The rewarded stimulus (S+ ) within each task is placed in the upper left corner. The negative symbols are placed as follows: S−1 : upper right; S−2 : lower left; S−3 : lower right.

J. Langbein et al. / Behavioural Processes 79 (2008) 156–164

159

Fig. 3. Time schedule of the different sections of the learning experiment. New discrimination problems and all mixed problem series were presented successively. New problems were at first presented for 1 day with all trials rewarded to test for shape preferences. Thereafter, training of the problem ran over 5 days. Mixed series including two problems (the current and the previous one) were presented for 1 day while the mixed series including all 10 problems was presented for 9 days.

We analyzed the similarity between the shapes within the different discrimination problems. To do so, we calculated the structural similarity index (SSIM; Wang et al., 2004) for each pair of shapes using a computer algorithm included in the ‘Perceptual Feature Toolbox’ (PFT; Cooke, 2006). The PFT is primarily intended for psychologists and cognitive neuroscientists who wish to analyze a set of experimental stimuli using standard computer vision measures. The PFT is a compilation of algorithms for analyzing similarities between images/objects applying features from computer vision and multidimensional scaling techniques. Features operate on either 2D images or 3D objects. In particular, the features in the PFT can be divided into three types: (1) computer algorithms that take pairwise distances between images by directly using raw image data in the distance calculation; (2) computer algorithms that first operate separately on each image and these intermediary values are then compared pairwise using a given distance function to generate a matrix of proximities; (3) computer algorithms that take image pairs as primary input and output a scalar measure of proximity between them (e.g., SSIM; Kannengiesser and Cook, 2006). Calculation of the SSIM is based on the assumption that the human visual system is highly adapted to extract structural information from the viewing field. To simplify matters, we will assume the same for the goats’ visual system here. SSIM is originally intended as a method for assessing perceptual image quality, but it can also be used as an objective method to compare different images for their level of identity. The SSIM compares local patterns of pixel intensities that have been normalized for luminance and contrast. The system dissects the task of similarity evaluation into three comparisons: luminance, contrast, and structure. Finally, the three components are combined to give an overall similarity measure. All algorithms of picture pre-processing and calculation of the SSIM are described in detail in Wang et al. (2004). The SSIM as calculated within the PFT has decimal values between 0 and 1, where 1 would indicate zero correlation between the two objects compared, and 0 would indicate that the two are the exact same object. The occurrence of spontaneous preferences were tested for S+ within the single discrimination problems by analyzing choice behavior during the 1-day pre-test applying the LOGISTICprocedure of SAS (SAS 9.1, SAS Institute Inc., Cary, NC).

Mean daily learning success was calculated as the percentage of correct trials per day, and was calculated separately for each individual discrimination problem during training. Furthermore, as an index of the absolute learning effort, we computed the number of trials the animals needed to fulfill the pre-defined learning criterion (trials to criterion, TtC). Because of the four-choice design, the criterion used to define a successful level of learning was 46% correct responses in at least two consecutive blocks of 20 trials (Hanggi, 1999) (p < .05; according to the binomial test with p0 = 0.25 and n = 20). The same criterion (46% of correct choices per day) was applied to daily learning success as the number of trials per individual and per day was always higher than 20. Daily learning success was calculated separately for each problem in the mixed series of the current and the previous problem. Furthermore, learning success was calculated separately for each problem over the first 10 trials per problem as well as over trials 30–40 in the mixed series of all 10 problems. For statistical analyses, we applied various models using an analysis of variance (ANOVA) within the MIXED-procedure of SAS. To determine the effect of the discrimination problem on TtC during training, we used a model with ‘problem’ as a fixed factor and ‘animal within each problem’ as a repeated factor. To test for the effect of the current and the previous problem on daily learning success in the mixed series, we applied a different model with ‘series’ and ‘discrimination problem within the series’ as fixed factors, ‘animal’ as a repeated factor and corresponding interaction terms. Finally, to verify the effect of the problem on learning success within the mixed series of all 10 problems, we again used a model with ‘problem’ as a fixed factor and ‘animal’ as a repeated factor. Post hoc tests of subclasses with a Tukey-honestly significant difference correction (to ensure a multiple test risk of first kind ≤.05) were computed in all models where significant effects were detected. We conducted all statistical analyses with SAS software (SAS 9.1, SAS Institute Inc., Cary, NC). 2.6. Ethical note Drinking water was solely delivered as a reward at the learning device during all sections of the study. However, this did not mean any limitation of water consumption to the animals when learning

160

J. Langbein et al. / Behavioural Processes 79 (2008) 156–164

success was low, as individuals were not restricted regarding the number of trials at the device. To balance for low learning success, animals just had to increase the number of button presses to get enough water, and we know from previous studies that they do this (Langbein et al., 2004, 2006). At no time in this study did we find any signs of serious distress because of insufficient water supply. All procedures involving animal handling and treatment in this study were approved by the Committee for Animal Use and Care of the Ministry of Agriculture of Mecklenburg-Vorpommern, Germany (file reference: LVL M-V/310-4/7221.3-1.1-010/03). 3. Results 3.1. Shape similarity The mean SSIM between pairs of shapes over all discrimination problems was 0.502 (±0.046) as analyzed by pairwise comparison of shapes within each problem. The lowest mean SSIMs were found in P3 (0.443) and in P8 (0.444), indicating a higher level of similarity between the shapes in those two problems. The lowest SSIMs were found between S+ and S−1 (0.390) in P3 and between S+ and S−3 (0.375) in P8. The highest mean SSIM was found in P9 (0.583), indicating a low level of similarity between the shapes in this problem. Except for P8, there was a significant effect of shape on the distribution of choices during the 1-day pre-test in all problems (p < .0001; all). While we detected spontaneous preferences for one or two specific S− in different problems, the animals never showed a preference for S+ . 3.2. Learning performance during training Learning curves based on mean daily learning success during training are shown separately for the 10 discrimination problems (P1 to P10) in Fig. 4 (top panel). All dwarf goats reached the learning criterion (46% of correct choice per day) in all problems except for P3. In this problem, only four animals reached the criterion. With the exception of P1 and P3, the animals passed the learning criterion on day 2 of training. As an indicator of absolute learning effort, Fig. 4 (bottom panel) presents box plots of TtC separately for the 10 problems. We calculated a hypothetical value of TtC for those animals who failed to reach the learning criterion in P3, based on their learning performance at the end of that problem. We found a significant effect of discrimination problem on TtC during training (F9,63 = 17.87, p < .001). Post hoc tests revealed a higher value of TtC in P1 and P3 compared to all other problems (p < .01, all). Furthermore, TtC was higher in P3 compared to P1 (p < .01). No differences in TtC were found between any other problems.

Fig. 4. Learning success of a group of eight dwarf goats in 10 discrimination problems trained consecutively. Top panel: Time related learning curves based on mean daily learning success (±S.D.). The dotted line marks the learning criterion. Bottom panel: The number of trials the animals needed to reach the learning criterion (TtC). Box plots show medians, 25th and 75th percentiles and data range, except outliers.

3.3. Short-term retention performance Learning success during concurrent recall of current vs. previous problems when presented in a mixed series is given separately for each problem in Fig. 5. We found a significant effect of the mixed series (F8,119 = 16.00, p < .001) and the discrimination problem within the series (current vs. previous; F1,119 = 15.43, p < .001) on learning success, the latter revealing that learning success in the current problem was higher compared to learning success in the previous problem. The interaction effect of series × problem was significant as well (F8,119 = 6.58, p < .001). Nevertheless, the success rate in the previous problem was clearly above the learning criterion in all mixed series, with the exception of P8, as the previous problem in the mixed series together with P9. Post hoc tests revealed a significant difference between the current and the previous problem only in series P1/P2 (p < .05) and P8/P9 (p < .001).

Fig. 5. Learning success in the current and in the previous discrimination problem (seen last time 6 days before) when presented concurrently in a mixed series over 1 day. Box plots show medians, 25th and 75th percentiles and data range, except outliers. The dotted line at 46% represents the learning criterion.

J. Langbein et al. / Behavioural Processes 79 (2008) 156–164

161

To break down these results to the individual level, Table 1 summarizes the number of problems recalled above criterion by each goat as well as the number of goats who remembered single problems during concurrent recall of P1 to P10. All goats recalled P9 and P10, a majority recalled P3 to P6 and P8, and at least one animal remembered even P1. Two goats concurrently recalled seven problems and nearly all animals recalled at least five problems. When we calculated learning success over trials 30–40 per individual and problem (Table 2), both the number of problems recalled above criterion by each goat and the number of goats who recalled single problems was much higher, indicating fast relearning during the concurrent presentation of the 10 problems. 4. Discussion

Fig. 6. Learning success during concurrent recall of 10 previously trained discrimination problems when presented in a mixed series. Learning success was calculated over the first 10 trials per problem. The time between initial learning and concurrent recall differed between problems (see Fig. 3). Box plots show medians, 25th and 75th percentiles and data range. Outliers are indicated by the letter O. The dotted line at 46% represents the learning criterion.

Learning success when P3 was the current problem was significantly smaller than that for the current problems in all other mixed series (p < .001) except for P9. As with the previous problem, learning success in P1 was significantly smaller than in P4, P5, P7, and P9 (p < .001, all), and learning success in P8 was significantly smaller (p < .001, all) than in all other problems except for P1 as the previous and P3 as the current problem. 3.4. Long-term retention performance Fig. 6 shows learning success over the first 10 trials separately for each problem during concurrent recall of all 10 discrimination problems (P1 to P10) presented together in one mixed series. To minimize the possibility of fast relearning, first learning success was only analyzed over the first 10 trials per animal per problem. We found a strong influence of discrimination problem on learning success (F9,63 = 17.58, p < .001). As expected, the highest learning success was reached in P9 and P10, which were seen last by the animals. However, the success rate was also above the learning criterion in a number of other problems (P3, P4, P5, P6, and P8). At this time, P8 had not been seen by the animals for 7 days while P3 had not been seen for 42 days. In addition to P1 and P2, the success rate in P7 was clearly below the learning criterion.

The aim of the study was to broaden the knowledge of cognitive and learning abilities of animals by studying a non-showpiece species in learning research. Our experimental setup enabled us to investigate individual learning performances under normal housing conditions and social settings. This setup should sufficiently mimic a naturalistic learning situation in domestic animals to satisfy the criticisms of learning research expressed by cognitive ethologists (Ristau, 2001). Dwarf goats were trained in 10 multipleobject discrimination problems one at a time (or serially). We did this because we wanted to test their ability to recall different problems concurrently after short as well as long intervals without practice with the problem. In this respect, our training procedure was in contrast to studies with horses where single problems were presented in a concurrent or serial/concurrent fashion (Dixon, 1970; Hanggi, 1999; Voith, 1975). However, investigations in rats have shown that these animals were equally skilful in concurrent recall of multiple problems, whether or not single problems were initially trained concurrently or serially (Santucci and Treichler, 1990). The goats performed well during initial training. After intensive pre-training, beginning with P2, the goats learned subsequent problems faster and reduced absolute learning effort (TtC) to reach the pre-defined learning criterion. This may be interpreted as evidence that they were starting to develop a learning set (Harlow, 1949). These results mirror a recently published study on the ‘learning to learn’ phenomenon in dwarf goats (Langbein et al., 2007a). There was one exception, P3, which had an unexpectedly higher value of TtC and for which not all goats reached the learning criterion within the 5 training days. Besides an initial preference for S−1 in this problem, as revealed in the pre-test, we computed a low value of SSIM between S+ and S−1 , indicating high similarity between specific shapes in P3 compared to other discrimination problems. Furthermore, some details of S+ and S−1 , like the axe handle and the kangaroo tail, look and behave similarly (have a similar direction). According to the theory of recognition-by-components

Table 1 Learning success of individual dwarf goats during concurrent recall of 10 previously trained discrimination problems when presented in a mixed series Animal

Learning success (>46%) P1

1398 1406 1409 1413 1417 1424 1430 1444 Animals per problem

P2

P3

P4

P5

P6

70 50

70 60 60 90 70 60

60

80

80 70 80 70

90 70

80 7

70 6

80 5

70

1

0

Problems per animal

60 80 50 5

P7

P8

P9

P10

60 60

100 100 90 80 80 100 90 90 8

90 60 60 80 60 80 90 90 8

80

80

0

70 4

7 5 5 6 5 6 3 7

Learning success was calculated over the first 10 trials per problem. We indicated learning success only for those animals that passed the learning criterion. The number of animals who successfully recalled single problems as well as the number of problems recalled per animal is summarized.

162

J. Langbein et al. / Behavioural Processes 79 (2008) 156–164

Table 2 Learning success of individual dwarf goats during concurrent recall of 10 previously trained discrimination problems when presented in a mixed series Animal

Learning success (>46%) P1

1398 1406 1409 1413 1417 1424 1430 1444 Animals per problem

50 60 50 3

P2

Problems per animal

P3

P4

80

70

80 100

100 50

60 50

60 50

5

5

90 70 70 100 70

100 6

P5 50 50 100 70 70 90 50 7

P6

P7

P8

P9

P10

90

60

90 100 60 60

60

60 90 78 70 60

100 90 90 100 90 100 100 100 8

90 80 90 90 70 80 90 80 8

70 6

50 90

4

70 50 7

9 5 8 8 7 8 7 7

Learning success was calculated over 10 trials per problem between trials 30 and 40. We indicated learning success only for those animals who passed the learning criterion. The number of animals who successfully recalled single problems as well as the number of problems recalled per animal is summarized.

(RBC; Biedermann, 1987), a higher level of similarity between two objects will occur if some elementary parts, such as cylinders, bricks, or cones (called geons), are identical in both. Thereby, some geons may be especially distinctive, while others may be incapable of supporting accurate recognition on their own. The RBC theory has so far been applied successfully to explain various aspects of visual pattern recognition in humans and pigeons (Kirkpatrick, 2001). We suggest, in the current study, the main reasons for retarded learning performance in P3 to be a combination of an initial preference for S−1 , a high level of similarity between S+ and S−1 , and identical geons of S+ and S−1 . Likewise, higher similarity between the shapes in P8 probably caused the slightly impaired learning performance in this problem during initial training as well. However, in P8, we did not find an additional preference for any S− as a further handicap and all goats learned to discriminate S+ during training, although with a higher error rate. Throughout the entire training, the experimental design allowed for the evaluation of two different learning processes in an alternating manner: serial learning of new discrimination problems and concurrent recall of different problems. Concurrent presentation of mixed series of the currently trained problem and a second one trained up to 6 days before, and not seen since, did not cause a serious decrease of learning success in either problem compared to the training phase. On the contrary, the dwarf goats clearly exceeded the learning criterion in both problems in all but one mixed series (P8/P9), demonstrating that they were capable of proficient retention of serially learned four-choice discrimination problems over a period of 6 days and could concurrently recall two different fourchoice problems. One would expect learning success to be higher in the current problem compared to the previous one. However, we did not find such differences in any mixed series except for P1/P2 and P8/P9. After training on five discrimination problems under serial learning conditions, the goats were probably surprised by the unexpected interspersion of the previously trained problem in P1/P2, resulting in a higher error rate for P1. We assume that the decreased learning success in P8 when presented as the previous problem, like in P3 when presented as the current problem, could be explained by higher shape similarity in these two problems as discussed above. Hanggi (1999), in a study of visual categorization in horses, argued that lower learning success with specific shapes is related to higher shape complexity. However, as one can see from the wide range of successfully discriminated shapes here, this was probably not a major reason for impaired learning success in specific problems for the dwarf goats. Accordingly, previous studies have demonstrated that goats are capable of discriminating a wide variety of shapes of different complexity (Baldwin, 1979; Blakeman and Friend, 1986). When presented with all 10 previously trained discrimination problems in one mixed series, the dwarf goats were able to

concurrently recall between five and seven different problems. While, as expected, recently trained problems were recalled most successfully, animals also remembered previously trained problems above criterion. Again, recalling specific problems (P3 and P8) seemed to some extent impaired, probably because of shape similarity as already discussed. Surprisingly, all goats were unable to discriminate P7. Learning success was high in this problem during initial training and it was successfully discriminated when presented in a mixed series concurrently with P6 or P8. Furthermore, we did not detect increased similarity between any pairs of shapes when comparing S+ of P7 with the negative stimuli of all other discriminations. Nevertheless, we suppose that some kind of hidden likeness may have been the reason for the uniform failure of all goats to discriminate S+ in P7 in the mixed series. Beside concurrent recall of multiple problems, dwarf goats showed proficient retention of a number of serially trained discriminations over quite long periods (up to 56 days). As postulated by Santucci and Treichler (1990), concurrent discrimination of multiple objects could be a potential evaluator of memory characteristics for comparison between different species. Thereby, they postulated that the concurrent task’s requirement may provide information on reference or long-term memory capacity rather than on working memory. At first glance, dwarf goats seemed to rank low in comparison to other species in terms of this ability. Macaque monkeys can retain 16 problems concurrently (Bakner and Treichler, 1993), rats and mice manage 7–8 problems (Rensch, 1976), and horses (Dixon, 1970) and elephants (Rensch, 1957) can memorize up to 20 discriminations concurrently. Furthermore, horses have been shown to retain multiple-problem discrimination over a period of 6 months (Dixon, 1970). Studies in food-storing birds under controlled laboratory conditions have provided convincing evidence of long-term retention of a very large number of learned relationships between visual cues and stored food as this kind of learning is of particular significance to these animals (Balda and Kamil, 1992; Vaughan and Greene, 1984). Regarding performance of the dwarf goats in this study, one has to bear in mind that our experimental design differed fundamentally in a number of aspects compared to the above cited investigations. The dwarf goats acted autonomously under naturalistic conditions without human supervision of any kind at all stages of the experiment. In operant learning studies with intensive human–animal interaction, it is often difficult to control for unintended additional reinforcement given by the supervisor himself (Sebeoke, 1970). Furthermore, we trained the goats by applying an open approach, meaning the animals had free access to the learning device all day and an unlimited number of trials. Thereby, the experimental design enabled the animals to obtain the reward via persistent responding without learning the cueing stimulus. Nevertheless, the learning success of the dwarf goats over all 10 problems in the mixed series was above 50% within the first 10

J. Langbein et al. / Behavioural Processes 79 (2008) 156–164

trials per problem, meaning every second trial was rewarded. The motivation to respond correctly each time might be lower with our setup compared to the more conventional design of learning experiments, in which the number of trials, and therefore, the amount of available reward, is limited (Franz and Roitberg, 2001). Because of this drawback, we worked with a four-choice design which reduced the chance of getting a reward and compelled the goats to discriminate between shapes and not arbitrarily press all buttons at a high frequency, which can be an alternative strategy with our open design (Langbein et al., 2006). Finally, although the goats presumably only memorized the cueing stimulus in each discrimination problem, the use of a four-choice design forced the goats to invest a higher discrimination effort compared to conventional two-choice tasks. Continuous presentation of all 10 discrimination problems in a mixed series resulted in increased learning success because of fast relearning even of problems trained long ago. As discussed previously, dwarf goats, like other species, are skillful in concurrent recall of multiple-problem discrimination, no matter if trained concurrently or serially (Santucci and Treichler, 1990). This observation indicates that goats can probably learn to simultaneously retain and concurrently recall a much larger number of discrimination problems than are shown in this study. From an ecological point of view, the ability to form numerous associations between visual cues offered by specific plants and food quality is an important pre-grazing mechanism that helps goats to exploit variations in vegetation and to graze selectively. According to the classification of ungulate species by their diets and feeding habits, Hofmann (1989) derived a browser-grazer continuum, which proceeds from grazers to intermediate feeders to pure browsers (or concentrate selectors). He argues that specialization in the consumption of a particular diet is reflected in adaptations in the selection, processing, and digestion of food. According to this theory, goats are classified as ‘intermediate feeders’ that feed on a mixture of shrubs/herbs/forbs and grass, often switching seasonally (Stuth, 1991). Their strategy to select for high quality food, if available, necessitates a greater degree of ability in diet selection and plant discrimination compared to plain grazers like cattle. Goats are able to associate certain plants with their toxic effects in terms of post-ingestive consequences and to avoid those plants in the future. Provenza et al. (1994) demonstrated that goats learned to avoid current season’s growth of the shrub Blackbrush (Coleogyne ramosissima), which contains higher concentrations of condensed tannins than older growth. The mechanism of how animals select a specific diet is not well known. It is probably influenced by genetic predisposition and by learning. Provenza et al. (1992) argue that ruminants learn about foods through two interrelated systems: the affective and the cognitive. The affective system integrates the taste of food and its post-ingestive feedback, while the cognitive system integrates the odor, appearance, and feel of food with its taste. Appearance, in the context of the affective system, has not yet been thoroughly discussed. Until now, the taste and odor of different plants have been considered the main variables that cause goats to distinguish between novel foods that differ in post-ingestive consequences (Duncan and Young, 2002; Provenza et al., 1992). This is probably true when the animals have already approached the food source; however, at medium range, the capacity to visually discriminate different foods and to judge the value (reward) of these foods may arguably be at a premium. Grazing animals use the sense of sight in the selection of their diet, as already shown in sheep (Arnold, 1966; Edwards et al., 1997). Therefore, in addition to a keen visual sense and the ability to discriminate palatable herbs and shrubs, it makes sense for goats to have a well-developed visual memory to store information about a large number of previously learned food items.

163

Acknowledgements The authors would like to thank D. Sehland for excellent technical care, data collection, and primary processing of raw data. Our very special thanks go to F. Robert Treichler, two anonymous referees, and J.S. Katz for their constructive revision of previous versions of this manuscript. Further thanks go to Christian Wallraven for his assistance with the Perceptual Feature Toolbox. Finally, we are greatly indebted to H. Erhard, D.L. Thomson, and Christine Baes for correcting the English style of the manuscript. References Aldezabal, A., Garin, I., 2000. Browsing preference of feral goats (Capra hircus L.) in a Mediterranean mountain scrubland. J. Arid Environ. 44, 133–142. Arnold, G.W., 1966. The special senses in grazing animals. I. Sight and dietary habits in sheep. Aust. J. Agric. Res. 17, 521–529. Bakner, L., Treichler, F.R., 1993. Age-differences in the acquisition and retention of concurrent discriminations by monkeys. Anim. Learn. Behav. 21, 51–58. Balda, R.P., Kamil, A.C., 1992. Long-term spatial memory in clark nutcracker, Nucifraga columbiana. Anim. Behav. 44, 761–769. Baldwin, B.A., 1979. Operant studies on shape discrimination in goats. Physiol. Behav. 23, 455–459. Biedermann, I., 1987. Recognition-by-components: a theory of human image understanding. Psychol. Rev. 94, 115–147. Blakeman, N.E., Friend, T.H., 1986. Visual discrimination at varying distances in Spanish goat. Appl. Anim. Behav. Sci. 16, 279–283. Bloomsmith, M.A., Marr, M.J., Maple, T.L., 2007. Addressing nonhuman primate behavioral problems through the application of operant conditioning: is the human treatment approach a useful model? Appl. Anim. Behav. Sci. 102, 205–222. Bullock, D.J., 1985. Annual diets of hill sheep and feral goats in Southern Scotland. J. Appl. Ecol. 22, 423–433. Carlstead, K.C., Shepherdson, D.J., 2000. Alleviating stress in zoo animals with environmental enrichment. In: Moberg, G.P., Mench, J.A. (Eds.), The Biology of Animal Stress. CABI Publishing, Oxon, NY, pp. 337–354. Cooke, T., 2006. Perceptual similarities amongst novel, 3D objects. Doctoral Thesis. Eberhard-Karls-University, Tübingen, Germany. Croney, C.C., Gardner, B., Baggot, S., 2004. Beyond animal husbandry: the study of farm animal cognition and ensuing ethical issues. Essays Philos. 5, 1–9. Dixon, J., 1970. The horse: a dumb animal?.neigh! The Thoroughbred Record 192, 1654–1657. Duncan, A.J., Young, S.A., 2002. Can goats learn about foods through conditioned food aversions and preferences when multiple food options are simultaneously available? J. Anim. Sci. 80, 2091–2098. Edwards, G.R., Newman, J.A., Parsons, A.J., Krebs, J.R., 1997. Use of cues by grazing animals to locate food patches: an example with sheep. Appl. Anim. Behav. Sci. 51, 59–68. Fiske, J.C., Potter, G.D., 1979. Discrimination reversal learning in yearling horses. J. Anim. Sci. 49, 583–588. Franz, H., Roitberg, E., 2001. A comparison of learning performance of dwarf goats in visual discrimination tasks with two or four simultaneously offered stimuli. Arch. Tierzucht 44, 661–669. Hanggi, E.B., 1999. Categorization learning in horses (Equus caballus). J. Comp. Psychol. 113, 243–252. Harlow, H.F., 1949. The formation of learning sets. Psychol. Rev. 56, 51–65. Hayes, K.J., Thompson, R., Hayes, C., 1953. Discrimination learning set in chimpanzees. J. Comp. Physiol. Psychol. 46, 99–104. Held, S., Mendl, M., Laughlin, K., Byrne, R.W., 2002. Cognition studies with pigs: livestock cognition and its implication for production. J. Anim. Sci. 80 (E.Suppl.), E10–E17. Hofmann, R.R., 1989. Evolutionary steps of ecophysiological adaptation and diversification of ruminants—a comparative view of their digestive-system. Oecologia 78, 443–457. Kannengiesser, S., Cook, T., 2006. Perceptual feature toolbox. In: Documentation and Technical Reference. Max Planck Institute for Biological Cybernetics, Tuebingen, Germany. Kendrick, K.M., da Costa, A.P., Leigh, A.E., Hinton, M.R., Peirce, J.W., 2001. Sheep don’t forget a face. Nature 414, 165–166. Kirkpatrick, K., 2001. Object recognition. In: Cook, R.G. (Ed.), Avian Visual Cognition [On-line]. Available: www.pigeon.psy.tufts.edu/avc/kirkpatrick/. Langbein, J., Puppe, B., 2004. Analysing dominance relationships by sociometric methods—a plea for more standardised and precise approach in farm animals. Appl. Anim. Behav. Sci. 87, 293–315. Langbein, J., Nürnberg, G., Manteuffel, G., 2004. Visual discrimination learning in dwarf goats and associated changes in heart rate and heart rate variability. Physiol. Behav. 82, 601–609. Langbein, J., Nürnberg, G., Puppe, B., Manteuffel, G., 2006. Self-controlled visual discrimination learning of group-housed dwarf goats (Capra hircus): behavioral strategies and effects of relocation on learning and memory. J. Comp. Psychol. 120, 58–66.

164

J. Langbein et al. / Behavioural Processes 79 (2008) 156–164

Langbein, J., Siebert, K., Nürnberg, G., Manteuffel, G., 2007a. Learning to learn during visual discrimination in group housed Dwarf Goats (Capra hircus). J. Comp. Psychol. 121, 447–456. Langbein, J., Siebert, K., Nürnberg, G., Manteuffel, G., 2007b. The impact of acoustical secondary reinforcement during shape discrimination learning of dwarf goats (Capra hircus). Appl. Anim. Behav. Sci. 103, 35–44. Macphail, E.M., Bolhuis, J.J., 2001. The evolution of intelligence: adaptive specializations versus general process. Biol. Rev. 76, 341–364. Meehan, C.L., Mench, J.A., 2007. The challenge of challenge: can problem solving opportunities enhance animal welfare? Appl. Anim. Behav. Sci. 102, 246–261. Melfi, V.A., Thomas, S., 2005. Can training zoo-housed primates compromise their conservation? A case study using Abyssinian colobus monkeys (Colobus guereza). Anthrozoos 18, 304–317. Mendl, M., 1999. Performing under pressure: stress and cognitive function. Appl. Anim. Behav. Sci. 65, 221–244. Murphy, J., Arkins, S., 2007. Equine learning behaviour. Behav. Proc. 76, 1–13. Nakagawa, E., 1992. Effects of overtraining on reversal-learning by rats in concurrent and single discriminations. Q. J. Exp. Psychol. B 44B, 37–56. Nicol, C.J., 2002. Equine learning: progress and suggestions for future research. Appl. Anim. Behav. Sci. 78, 193–208. Provenza, F.D., Pfister, J.A., Cheney, C.D., 1992. Mechanisms of learning in diet selection with reference to phytotoxicosis in herbivores. J. Range Manage. 45, 36–45. Provenza, F.D., Lynch, J.J., Burritt, E.A., Scott, C.B., 1994. How goats learn to distinguish between novel food that differ in post-ingestive consequences. J. Chem. Ecol. 20, 489–822. Puppe, B., Ernst, K., Schön, P.C., Manteuffel, G., 2007. Cognitive enrichment affects behavioural reactivity in domestic pigs. Appl. Anim. Behav. Sci. 105, 75–86. Rensch, B., 1957. The intelligence of elephants. Sci. Am. 196, 44–49. Rensch, B., 1976. The evolution of brain achievements. Evol. Biol. 1, 26–68. Ristau, C., 2001. Cognitive ethology. In: Wilson, R., Keil, F. (Eds.), The MIT Encyclopedia of Cognitive Sciences. MIT Press, Cambridge, Mass, pp. 131–133. Santucci, A.C., Treichler, F.R., 1990. Concurrent object-discrimination learning in rats. Anim. Learn. Behav. 18, 295–301. Sappington, B.K.F., Goldman, L., 1994. Discrimination learning and concept formation in the Arabian Horse. J. Anim. Sci. 72, 3080–3087.

Sebeoke, T.A., 1970. Clever Hans Phenomenon: Communication with Horses, Whales, and People. NY Academy of Sciences, New York. Sondergaard, E., Ladewig, J., 2004. Group housing exerts a positive effect on the behaviour of young horses during training. Appl. Anim. Behav. Sci. 87, 105–118. Stuth, J.W., 1991. Chapter 3—foraging behavior. In: Heitschmidt, R.K., Stuth, J.W. (Eds.), Grazing Management: An Ecological Perspective. Timber Press, Portland Oreg, pp. 259–272. Swaisgood, R.R., White, A.M., Zhou, X.P., Zhang, H.M., Zhang, G.Q., Wei, R.P., Hare, V.J., Tepper, E.M., Lindburg, D.G., 2001. A quantitative assessment of the efficacy of an environmental enrichment programme for giant pandas. Anim. Behav. 61, 447–457. Thomas, D.R., Mckelvie, A.R., Mah, W.L., 1985. Context as a conditional cue in operant discrimination reversal-larning. J. Exp. Psychol. Anim. B 11, 317–330. Thomas, R.K., 1986. Vertebrate intelligence: a review of the laboratory research. In: Hoage, R.J., Goldman, L. (Eds.), Animal Intelligence. Smithsonian Institution Press, Washington, D.C./London, pp. 37–55. Thomas, R.K., 1996. Investigating cognitive abilities in animals: unrealized potential. Cogn. Brain. Res. 3, 157–166. Toates, F., 2004. Cognition, motivation, emotion and action: a dynamic and vulnerable interdependence. Appl. Anim. Behav. Sci. 86, 173–204. Treichler, F.R., 1984. Long term retention of concurrent discriminations by monkeys. Physiol. Psychol. 12, 92–96. Treichler, F.R., Wetsel, W.C., Lesner, S.A., 1977. Some characteristics of concurrent discrimination and retention by monkeys. Learn. Motiv. 8, 213–228. Vaughan, W., Greene, S.L., 1984. Pigeon visual memory capacity. J. Exp. Psychol. Anim. B 10, 256–271. Voith, V.L., 1975. Pattern discrimination, learning set formation, memory retention, spatial and visual reversal learning by the horse. Master’s Thesis. Ohio State University, Columbus, OH. Wang, Z., Bovik, A.C., Sheikh, H.R., Simoncelli, E.P., 2004. Image quality assessment: from error visibility to structural similarity. IEEE Trans. Image Process. 13, 601–612. Watson, S.L., Shively, C.A., Voytko, M.L., 1999. Can puzzle feeders be used as cognitive screening instruments? Differential performance of young and aged female monkeys on a puzzle feeder task. Am. J. Primatol. 49, 195–202.