Landmark learning by honey bees

Landmark learning by honey bees

Anita. Behav., 1987, 35, 26-34 Landmark learning by honey bees JAMES L. G O U L D Department of Biology, Princeton University, Princeton, New Jersey...

690KB Sizes 0 Downloads 63 Views

Anita. Behav., 1987, 35, 26-34

Landmark learning by honey bees JAMES L. G O U L D

Department of Biology, Princeton University, Princeton, New Jersey 08544, U.S.A.

Abstract. There are two published models for landmark learning by honey bees: (1) bees remember only the presence or absence of landmarks in each of several sectors around a food source, or (2) bees store a visual image of the landmarks. Experiments reported here indicate that bees are able to remember landmarks pictorially, and store their location, shape and colour. The angular resolution of the memory near the visual horizon is about 3.1 ~ horizontally and about 5.5 ~ vertically; this suggests that landmark memory has a higher resolution than the 8-10 ~ measured for flower memory.

Honey bees can use landmarks to locate an inconspicuous food source (review in von Frisch 1967; Hoefer & Lindauer 1976; Anderson 1977a; review in Wehner 1981; Cartwright & Collett 1982, 1983; review in Gould 1984). This ability, which appears analogous to the ability of digger wasps to use landmarks to locate an inconspicuous burrow entrance (Tinbergen & Kruyt 1938), has been explained in two ways, based on data from two different experimental designs. Anderson (1977a) found that when bees were trained to find food at a site surrounded by (usually) eight landmarks (Fig. la) and then tested with some of the landmarks removed, the returning bees searched too close to the remaining landmarks (Fig. lb). He concluded that the bees were not matching against a stored picture since the configuration seen from the new search point would not have matched a visual image. He proposed instead that the bees simply remembered whether or not each of perhaps four 90 ~ sectors in their visual world was occupied by a landmark at the food source. Hence the removal of some of the landmarks would require bees to move closer to the remaining landmarks in order to fill the now-empty sectors. The precise number of sectors was not determined. The sector hypothesis is similar in some ways to the parameter hypothesis of flower learning (Anderson 1972, 1977b; Cruse 1972; Schnetter 1972; Ronacher 1979; review in Gould 1984); the parameter hypothesis imagines that information about flower shape and pattern is stored, not as a picture, but as a list of defining characteristics; specifically, bees were said to store spatial frequency (roughly, the ratio of edge to area of the flower), relative colour areas, line angles and so on. Both the sector and parameter hypotheses have the advantage that they would require considerably

less memory-storage capacity than the visual memory (picture) hypothesis. To use a human analogy, the information content of an advertisement specifying the important parameters of, say, an automobile (make, model, year, body type, colour, accessories, mileage, etc.) is far lower than the information content of an actual picture of the item for sale; the list of important parameters is clearly an adequate and inexpensive way to store (or communicate) information in certain contexts. Gould (1985, 1986) recently tested the parameter hypothesis, and found that bees in fact remember flowers as pictures. Gould's experiments were based on a critical difference between the general predictions of parameter hypotheses and picture hypotheses: only a pictorial memory necessarily preserves the spatial relationship between elements in a pattern, at least within the limits of the resolution of the memory-storage system. For example, suppose a flower has four petals, two yellow and two blue, and is seen against a green background. A parameter system might store the spatial frequency of the flower (a consequence of petal shape, size and spacing), colour areas of yellow and blue, amount of various edge-colour angles (e.g. 20 ~ of visual angle of horizontal edge with blue above and green below, 15~ of vertical edge with yellow left and green right, etc.) and so on. None of this necessarily preserves the spatial information except incidentally; hence, if all the petals are, say, perfect circles and are non-contiguous (and so each is entirely surrounded by green) the parameter list does not specify whether the petals are arranged as a +-shaped or x-shaped cross, or whether the blue petals are across from one another or adjacent. When Gould (1985, 1986) trained bees to one shape of such a pair, bees readily learned the distinction; if they learned only 26

Gould. Landmark learning by honey bees

9

~

9

[:cm

a--

- - b

Q

[] 9

[50cm LO

Figure I. Anderson trained bees to find food in the centre of a square array of eight landmarks (a), and then recorded where returning bees searched when some of the landmarks were removed (hatched area in b). Cartwright & Collett, on the other hand, generally used only three landmarks positioned to one side (c). Alterations of the array usually left the landmarks in such an arrangement that returning bees could find a location from which the landmarks were all in the same direction as during training (d).

parameters, they should have confused the two patterns and thus have been unable to choose reliably. By increasing the number of elements, G o u l d was also able to measure the resolution of the memory/matching system; it is of the order of 8-10 ~ of visual angle. This is well below the 1-2 ~ resolution of real-time bee vision. By analogy, Anderson's data could also be explained by a picture hypothesis, assuming the m e m o r y is of very low resolution (Gould 1984). Alternatively, it could be that bees have a highresolution landmark memory, and Anderson's bees were matching some of the remaining landmarks to some of the missing ones in their picture, resulting in a 'best match' at the location at which they searched. Cartwright & Collett (1982, 1983) performed a series of experiments in which bees were usually trained with three landmarks (Fig. lc) and then tested with an altered set. Unlike Anderson's usual arrangement, however, there was normally a location in the array from which the landmarks would all appear to be at the same azimuth angles seen

27

during training, and it was at this location that the bees searched (Fig. ld). Cartwright & Collett concluded that the bees had stored an eidetic image of the landmarks seen during training and then had formed a best match during their search on subsequent trips. The nature and resolution of the storage was not investigated. Their use of the term 'eidetic' in this context does not correspond to the usual definition in experimental psychology, from which the term was borrowed. An 'eidetic image' is, technically, a high-resolution mental picture corresponding in accuracy to real-time visual resolution; the ability to form true eidetic images is considered rare even in humans. Cartwright & Collett (1982, 1983) probably mean only that they believe bees store picture-like visual images; certainly, the pictorial flower memory described by Gould (1985, 1986) is not eidetic given its lowerthan-real-time visual resolution. Cartwright & Collett's results might also be explained by Anderson's hypothesis if the number of visual sectors 'scored' by the bees is eight, rather than the four Anderson concluded were adequate to explain his data. The goals of the present research were to deter]nine whether bees remember landmarks by means of sector-occupancy or photograph-like images, to attempt to reconcile the apparent contradiction between Anderson's results and those of Cartwright & Collett, to determine the resolution of the pictures or sectors, to determine whether colour and/or shape can be remembered and to determine whether, as all previous investigators have assumed, flower memory and landmark memory are separate processes. (For example, landmarks might be remembered as part of a p a n o r a m a that includes the flower.)

GENERAL METHODS Individually marked honey bees, Apis mellifera ligustica, were trained to a testing apparatus 25 m from the hive. The apparatus consisted of a square sheet of Plexiglas 12 m m thick and 125 cm on each side. A 15 x 15 array of holes was drilled in the Plexiglas; the holes were spaced 7.5 cm apart; each hole was 3 mm in diameter and 9 m m deep. The sheet was painted flat white. A drop of 1-5 M unscented sucrose solution was placed in one of the 225 holes, and landmarks were set out nearby. After the bee had discovered and fed from the hole,

28

Animal Behaviour, 35, 1

a drop of food was placed in a new, randomly chosen hole and the landmarks were moved and placed so they would be in the same relative location. It was important to move the hole each time or the bee would begin to search only in a particular sector of the sheet. The hole used previously was then rinsed several times with distilled water and emptied with a syringe. Only one bee was used at a time, since when there are two or more bees searching, they clearly interfere with each other. Testing consisted of setting out a new Plexiglas sheet, with a fresh set of landmarks in a new location, with the array altered in some way. The first five landings of the returning bee on the alternative holes (a subset of holes defined below) was then scored, after which the training sheet was again offered with food and the set of landmarks used in training. The testing apparatus never contained food, and alternative holes were never reused. Statistical measures, which will be discussed later, demonstrated that the first five landings can be treated as independent. A landing was defined as a full touchdown with wings still, followed by a probe into the hole with the proboscis. The landmarks usually consisted of 'wooden blocks 2.2 cm or some multiple on a side; the most common landmark, for example, was 2.2 x 2.2 x 4-4 cm. The landmarks were painted with enamel; the usual colour was yellow. The statistical tests used involved a conservative application of the binomial distribution or, where prediction of the null distribution of landings seemed to require making assumptions about bee behaviour, by actually measuring the chance distribution. Spontaneous preferences that might bias the results were generally controlled for by repeating each experiment with the role of training and testing landmarks reversed. The independence of landings was tested in several ways, all adapted from the tests devised by Gould (1985, 1986). First, one of the experiments, the first shape test, which involved columns versus T-shaped landmarks, was repeated using a protocol that allowed returning bees to land only once during each test; the results were indistinguishable from the five-landing-protocol data. The choice patterns for individual bees over all the experiments were then compared within and between the blocks of five landings to see wtiether sequential choices of the same hole within a block were either

more or less common than would be expected on the basis of the overall choice ratio; there was no difference. The data were also examined to see whether the first choice in each set of five landings was more or less likely to be correct than the second, third, fourth, or fifth; again, there was no difference. Finally, the data were analysed to determine whether within-block variance was greater or less than between-block variance; again, there was no difference. Hence, the data appear to be as previous investigators have assumed, independent, and so are combined for the statistical tests used in this report.

M E T H O D S AND R E S U L T S

Shape Perhaps the most direct way to pit the picture hypothesis against the sector hypothesis is to find out if bees can remember the shapes of landmarks; the picture hypothesis requires that shape be remembered, at least within the limits of the resolution of the memory, whereas the sector hypothesis, at least in its orginal form, requires that shape be ignored. A bee was trained during 10 visits to find food between two identical landmarks of type A, each 15 cm from the hole with food, and then tested with arrays containing a second, type B, landmark. The arrays used were of the form A A BB, B B A A and B A A B. The alternatives were the three holes midway between the landmarks, one between each pair. The data from these tests can be considered in two ways: (1) the three alternatives may have equal probabilities of being chosen if the bee is unable to learn the discrimination, or (2) the hybrid alternatives (A B) are more (or less) likely to be chosen by chance and so should be excluded, leaving only A A versus B B. Both statistics are reported here. The shapes tested consisted of vertical yellow columns against T-shaped yellow landmarks of equal area, and the reverse; yellow triangles against yellow squares of equal area, and the reverse; and T-shaped yellow landmarks against yellow columns surmounted by yellow triangles with an equal total area, and the reverse (Fig. 2). (Equal visual areas as seen from the feeding hole were used to control for the possibility that if bees use a sectoroccupancy memory, they might remember 'degree of occupancy' rather than just whether or not the

Gould." Landmark learning by honey bees A

B

C AAAB

P
o_

Pattern

D AABB

P.< 0.001 =92

o

}0 I00

P
P< 0.001

o

P
p.< o.001

L i00

P
29

P~ 0.ooI

,ooo] lZ t

A l t h o u g h consistent with the picture hypothesis, the observation t h a t bees can r e m e m b e r shapes does not rule out a n ad h o c extension of the sector hypothesis: by analogy with the p a r a m e t e r hypothesis of flower learning, bees might be wired to r e m e m b e r more t h a n just sector occupancy, including, perhaps, the p a r a m e t e r s associated with each sector such as spatial frequency, colour a n d line angle. As discussed above, only pictures necessarily preserve the relations between segments of a flower. If bees r e m e m b e r l a n d m a r k s sector by sector according to a set o f measured parameters, then they should confuse properly designed landmarks with the same area, colour, spatial frequency and distribution of line angles. To examine this possibility, bees were trained as before and tested o n cross-shaped l a n d m a r k s with a high crossbar versus a low crossbar, a n d the reverse, a n d on S-shaped l a n d m a r k s versus their mirror images, a n d the reverse (Fig. 3). Each of A

B

C

D

AA AB

A A EIB

I00

P<0.001 I

=89

I

=81

o

E

"5

-~I

o

0 ,

Figure 2. To determine whether honey bees can learn the shape of landmarks, bees were trained to find food between a pair of yellow landmarks of the shape shown in the A column and then tested in arrays offering a pair of As, a pair of Bs, and a hybrid pair (AB), or a pair of As and two hybrid pairs. The B landmarks had the same colour and visual area as the A landmarks. The data in column C indicate the percentage of choices for the training pair (AA) versus the two alternatives (AB and BB), while the data in column D compare only AA versus BB. The sample sizes in column C are always 150 landings; the sample sizes in column D, where AB landings are excluded, vary, and are given on each bar graph. The probabilities are calculated from the binomial theorem. sector c o n t a i n e d a landmark.) Each bee was tested once o n each array, each test consisting of five choices, for a total of 15 landings. Ten different bees were used in each comparison, yielding a total of 150 landings to be used in the three-alternative comparisons, of which 100 landings can be used in the two-alternative comparisons. The results are presented in Fig. 2. Bees can r e m e m b e r shape.

o

2N

~

Figure 3. To determine whether the shape of landmarks is remembered photographically, bees were trained to find food between a pair of yellow landmarks of the shape shown in the A column and then tested in arrays offering either a pair of As, a pair of Bs and a hybrid pair (A B), or a pair of As and two hybrid pairs. The B landmarks had the same colour, visual area, spatial frequency and distribution of line angles as the A landmarks; only the shape was different. Labelling and statistics as in Fig. 2.

Animal Behaviour, 35, 1

30

these landmark pairs has identical colour, area and distibution of line angles (lengths of vertical and horizontal lines), but the spatial arrangement of the parts of each landmark differs from the other member of its pair. Again, 10 bees were tested on three variants of the array (A A B B, B B A A and B A A B, where A represents the training shape) and the landings on the holes halfway between each pair of landmarks in the array were recorded. The results, which are presented in Fig. 3, suggest that bees can remember shape as a picture.

Colour To see whether or not bees are able to remember the colour of landmarks, bees were trained to yellow versus blue landmarks and the reverse. The colours were those used previously (Gould 1985, 1986) because they induced the same rate of spontaneous landings; whether spontaneous landings on flower-like targets is the best measure for sensory equivalency in the context of landmark learning is debatable, and so a conservative interpretation of these experiments is that they test whether the shading, or darkness, of a landmark can be remembered. Again, 10 bees were allowed to land five times on each of three arrays. The results

A

B

C AA

AB BB

P< O.OOI

E

el

Bi

~

g

D AA

BB

P < 0.001

o

~, Ioo

P< o.ool

P-:: O.001

0

Figure 4. To determine whether honey bees can learn the colour of landmarks, bees were trained to find food between a pair of yellow or blue landmarks of the shape shown in the A column and then tested in arrays offering either a pair of As, a pair orBs and a hybrid pair (AB), or a pair of As and two hybrid pairs. The B landmarks had the same shape and visual area as the A landmarks, but were blue instead of yellow; the hatched shapes in columns A and B indicate yellow landmarks, while the black shapes indicate blue landmarks. Labelling and statistics are the same as in Fig. 2.

are presented in Fig. 4. Bees can remember colour or shade.

Horizontal Resolution Determining the resolution of the picture that bees store of landmarks is not necessarily trivial: a searching bee can fly as close to a nearby landmark as it likes. But if the apparent size of two or more landmarks, which must be seen simultaneously in order to 'triangulate' the food source, is slowly reduced, there should come a point at which they are too small to be seen from the spot from which they must both be seen for the bee to learn to locate the food. The original strategy of this experiment, then, was to increase the distance to the landmarks (and hence reduce the apparent visual size) until bees could no longer locate the food at a level better than chance. However, it would be possible to interpret such experiments as merely a test of realtime visual resolution, since the bees might simply remember the first training step, landmarks with large visual size, and then try to match to that original picture on later visits. As a result, two bees were trained on a decreasing series, two on an increasing series (so that the landmarks appeared to get larger for each round of training and testing) and two on a random series. Bees on the decreasing series could not initially resolve even the training shapes from the feeding hole, though they could still fly up and inspect the shapes. At some point the landmark separation in this series must become small enough so both landmarks can be resolved in real time simultaneously at the feeding hole; at this point they would appear to be 1 ~ ~ across. But if landmarks are remembered pictorially in anything but an eidetic fashion, they will need to be closer (that is, visually larger) before they can be remembered and matched from the triangulation point. Because the results of all three approaches (increasing, decreasing and random series) were indistinguishable, the data were combined. Based on early observations, the vertical and horizontal resolution were investigated separately. Bees were trained individually to find food halfway between two landmarks, and then tested by scoring the first five landings; then the separation and/or the size of the two landmarks was altered as appropriate and training was begun anew. As before, the array was moved between each training visit and between each testing. Each

Gould." Landmark learning by honey bees bee was tested five times, for a total of 25 landings each a n d 150 landings in all from the six bees together. After some preliminary experimentation, the alternative holes to be scored were defined as those lying within half of the distance from the training hole to each landmark; hence, w h e n each l a n d m a r k was 50 cm from the training hole, the training hole a n d the three holes o n each side (that is, those lying within 25 cm) on the line connecting the two l a n d m a r k s were scored. Hence, a landing on the training hole was c o u n t e d as a 'correct' choice, a landing o n one o f the alternatives within h a l f the distance to the l a n d m a r k s was counted as 'incorrect', a n d any other landings were ignored. This procedure was chosen empirically, a n d was used for b o t h the control m e a s u r e m e n t s and the resolution tests; hence any u n i n t e n t i o n a l (and unobserved) bias it m i g h t be imagined to have i n t r o d u c e d is controlled for by c o m p a r i n g the various measurements with one other. T h e first step was to measure the p e r f o r m a n c e o f bees w h e n the visual size of the l a n d m a r k remained unchanged. This involved training the first pair o f bees to find food between two l a n d m a r k s 2.2 cm wide, 2.2 cm deep a n d 4-4 cm high, each positioned 15 cm o n each side of the hole with food, a n d then testing with the same array; this was followed by training a n d testing with 3-3 • 3.3 x 6-6 cm landm a r k s at 22.5 cm, and so on; as a result, the visual size of the l a n d m a r k s as seen from the hole with food remained constant. The experiment was then repeated beginning with the m a x i m u m separation a n d decreasing it for each r o u n d o f training and testing. Finally, a r a n d o m series was used. The resulting choice behaviour of the bees is s h o w n in the top, Control curve in Fig. 5. (The value for 'correct' choices at 7"5 cm is t a k e n to be 100% since at that distance there are no alternative holes. W h e n an alternative hole fell exactly halfway between the training hole and a l a n d m a r k , as when the l a n d m a r k s were 15 cm from the training hole, half o f the landings at these b o u n d a r y holes were counted as 'correct' a n d half as 'incorrect'. The likely validity o f this procedure is indicated by the relatively linear nature of the resulting curve, and is not crucial in any case, because the same criterion was used in the other m e a s u r e m e n t s with which these data were compared.) T h e next step was to determine the level o f chance performance. This can be calculated assuming a n equal probability of finding any of the holes

31 Distance (cm)

ioo

7.5 ".

15

22.5

30

37.5

45

60

'

'

'

'

'

' 'x

75

90

IZO

'

'

'

8o-

6o-

"', 4o

g_

~ Horizontal ~ Resolut~orl

99 ''./3 Chance

20

~o ' 9 .13

~ "~176

9

,

,,

.

9

"o..;---.~

Angular Size (o)

Figure 5. The upper curve, labelled Control, represents the combined data from six bees trained and tested by three different techniques (described in the text) to landmarks whose apparent size was constant at all distances. The sample size is 150 landings at each point. The upper horizontal axis shows the distance from the testing hole to each landmark; the vertical axis indicates the percentage of landings on the testing hole. The bottom curve, labelled Chance, represents the probability of a bee locating the correct hole under the conditions of the test (see text) in the absence of any landmarks. Only two bees were tested, and the sample size is 50 landings at each point (squares). The middle curve, labelled Horizontal Resolution, represents the combined data frmn six bees tested by three different techniques to landmarks whose apparent width as seen from the testing hole decreased with increasing distance; the bottom axis shows the corresponding visual angle. The sample size is 150 landings at each point. The performance of these bees appears to fall from control to chance levels at about 3~.

in the array; the chance o f landing o n the training hole is therefore the reciprocal of the total n u m b e r of alternative holes in the array. This calculation yields the lower curve in Fig. 5. To see if this calculation is correct, the chance b e h a v i o u r o f bees was measured after training to l a n d m a r k s separated by various distances. These m e a s u r e m e n t s were m a d e in conjunction with the constant-angle experiment already described. The testing consisted of removing the l a n d m a r k s altogether, arbitrarily defining a new 'correct' hole as t h o u g h the array had, as usual, been moved, a n d then scoring landings. Because these tests took a very long time (the bees were hesitant to land and rarely alighted on the array of alternative holes), only two bees were used a n d 50 total landings scored at each separation. As the data points near the b o t t o m curve in Fig. 5 indicate, the behaviour o f the bees unable to see the l a n d m a r k s a p p r o x i m a t e d the chance calculation.

32

Animal Behaviour, 35, 1

Presumably the behavigur o f the bees in the test of the resolution of l a n d m a r k m e m o r y should approximate the upper, Control curve of Fig. 5 (constant visual angle) when the l a n d m a r k s are of a size that can be readily stored in, or matched to, a bee's l a n d m a r k memory; o n the o t h e r h a n d , performance should be at or near the chance level when the a p p a r e n t size is below the resolution of the storage system. T o measure the h o r i z o n t a l resolution, the same procedure outlined for the Control measurements was followed except t h a t the width of the l a n d m a r k s remained 2.2 cm t h r o u g h o u t . (The height increased with increasing separation so that the vertical visual angle r e m a i n e d c o n s t a n t as the horizontal visual angle declined.) The middle, Horizontal Resolution curve o f Fig. 5 is fitted by eye to the data f r o m these tests. Based o n the point at which this curve of the changing visual angle declines to a p o i n t halfway between the upper, Control curve, a n d the lower, C h a n c e curve, the horizontal resolution o f l a n d m a r k m e m o r y is a b o u t 3.1 ~. O t h e r criteria would give slightly different values.

Vertical Resolution The same set o f procedures was used to measure the vertical resolution of l a n d m a r k memory, except that the initial l a n d m a r k was 4.4 cm wide, 2.2 cm deep and 2.2 cm high, only four bees were used, the r a n d o m - s e p a r a t i o n series was n o t r u n and the width was increased with increasing separation, so that the a p p a r e n t size of the l a n d m a r k as seen from the training hole h a d a changing vertical angle a n d a c o n s t a n t horizontal angle. The sample size was 100 landings at each separation. The data are summarized in Fig. 6, Based on the criterion used on the Horizontal Resolution data, the vertical resolution of l a n d m a r k m e m o r y is a b o u t 5-5 ~.

Anderson's Results Given t h a t bees a p p e a r to have relatively highresolution pictorial m e m o r y for landmarks, why did the bees in A n d e r s o n ' s experiments n o t search accurately? In a n effort to u n d e r s t a n d this behaviour, bees were trained as before to a hole on the Plexiglas sheet s u r r o u n d e d by eight yellow landmarks, each 2.2 x 2.2 x 4.4 cm a n d either 15 or (for those located diagonally) 21.2 cm away (Fig. 7). As before, the array was m o v e d after each training visit and testing. After training, h a l f of the land-

Distance

I00

7,5

15

9

80-

22.5

9

(cm)

30

",

45

6o

x

75

90

120

Control

609~

~ 99

40-

~

"%

Chance

Verticat Resolution

9

~ 1 7 6 1 7 "6~

20-

~176 " 9 g~.-g . q _ ~..

ir -~'.-~4.--.~

I

I

i

I

I

,

,

o

16

IZ

8

6

4

3

2

~.5

1.0

A n g u l a r S i z e (o)

Figure 6. The upper curve, labelled Control, represents the combined data from four bees trained and tested by two different techniques (see text) to landmarks whose apparent size was constant at all distances. The sample size is 100 landings at each point. The upper horizontal axis shows the distance from the training hole to the landmark, while the vertical axis shows the percentage of correct landings. The bottom curve, labelled Chance, represents data from two bees (50 landings at each point) tested in the absence of landmarks according to the criteria used in the other tests (see text). The middle curve, labelled Vertical Resolution, represents the combined data from four bees trained and tested by two different techniques to landmarks whose apparent vertical size as seen from the testing hole decreased with distance from the testing hole; the lower horizontal axis indicates the apparent visual size. The sample size is 100 landings for each point. The performance of these bees appears to decline from control to chance levels at about 5-5~.

marks were removed a n d the first five landings of each bee were scored. Each of five bees was tested twice. T h e results are plotted as a density diagram in Fig. 7B. The results of a similar test p e r f o r m e d when all o f the l a n d m a r k s were left in the array are shown in Fig. 7A. As A n d e r s o n reported, the bees do seem to search closer to the remaining landmarks. To see if this b e h a v i o u r might be based o n some sort of best-fit pictorial matching, the training array was altered so that the diagonal l a n d m a r k s were blue (Fig. 7C,D); since the tests reported above suggest t h a t bees can r e m e m b e r the colour of landmarks, they might be less likely to 'confuse' remaining l a n d m a r k s with the missing ones o f a different colour during testing. Again, a density-oflandings plot was made for the intact array (Fig. 7C) a n d the disturbed array (Fig. 7D). A l t h o u g h there m a y still be some tendency to search closer to the remaining landmarks, it appears to be m u c h

Gould; Landmark learning by honey bees []

4~

[] 1

3

3

1

1

34

1

[]

2 []

[]

2

2 []

[]

[]

[]

3

11

2

26

3

[]

2

[]

A

9

[]

9

[]

B

9

2

2

1

3

38

1

1

1

1

[]

[]

9

C

[]

[]

9

1

2

4

29

6

1

[]

33

flower m e m o r y t h a t it is p r o b a b l y stored separately (as, indeed, previous investigators have assumed), rather t h a n as p a r t of some sort of all-purpose p a n o r a m a t h a t includes the flower a n d the surr o u n d i n g landmarks. T h e value m e a s u r e d here is at least consistent with the r o u g h m e a s u r e m e n t o f 3 4 ~ made by v o n Frisch & L i n d a u e r (1954). W h e t h e r this l a n d m a r k picture is stored as part of the larger set of i n f o r m a t i o n specific to each food source (Bogdany 1978; G o u l d 1984) is yet to be determined. It also remains to be seen whether the ability of bees to r e m e m b e r a n d use l a n d m a r k s en route (for a review see yon Frisch 1967; W e h n e r 1981; G o u l d 1984) is based on or is p a r t of a single system t h a t includes the learning studied in these experiments, or is itself specialized for its own task.

ACKNOWLEDGMENTS

D

Figure 7. When bees were trained to find food in the middle of an array of eight identical yellow landmarks and tested on the same array (A), they searched in the middle of the array; the landmarks are indicated by squares, and the number of landings at each location is indicated. When tested with four of the landmarks removed, bees searched closer to the remaining landmarks (B). Bees trained to find food in the middle of a hybrid set of landmarks (four yellow and four blue, with the blue ones in the corners of the array) searched accurately when tested either with an unaltered array (C) or one with only four landmarks (D).

reduced; this suggests that bees faced with an array of m a n y identical l a n d m a r k s m a y sometimes confuse t h e m w h e n attempting to m a t c h w h a t they see in real time with the picture they have stored in their m e m o r y .

DISCUSSION T a k e n together, the experiments reported here indicate t h a t honey bees r e m e m b e r the l a n d m a r k s s u r r o u n d i n g a food source as a colour p h o t o g r a p h with a horizontal resolution of a b o u t 3 ~ a n d a vertical resolution of 5-6 ~. The better horizontal resc~lution is not surprising: Seidl & Kaiser (1982) report a similar asymmetry in the optics of the c o m p o u n d eye. The l a n d m a r k resolution is so m u c h better t h a n the 8-10 ~ resolution of h o n e y bee

I t h a n k Kate M o n a h a n , F r a n k Wojcik a n d A1 Weisenberger for technical help. This research was supported by N S F g r a n t BNS 85-06798.

REFERENCES Anderson, A. M. 1972, The ability of honey bees to generalize visual stimuli. In: InJormation Processing in the Visual System of Arthropods (Ed. by R. Wehner), pp. 207-212. Berlin: Springer Verlag. Anderson, A. M. 1977a. A model for landmark learning in the honey bee. J. comp. Physiol., 114, 335-355. Anderson, A. M. 1977b. Shape perception in the honey bee. Anim. Behav., 25, 67-69. Bogdany, F. J. 1978. Linking of learning signals in honey bee orientation. Behav. Ecol. Sociobiol., 3, 323-336. Cartwright, B. A. & Collett, T. S. 1982. How honey bees use landmarks to guide their return to a food source. Nature, Lond., 295, 56(~564. Cartwright, B. A. & Collett, T. S. 1983. Landmark learning in bees. J. eomp. Physiol., 151, 521-543. Cruse, H. 1972. A qualitative model for pattern discrimination in the honey bee. In: Information Processing in the Visual System of Arthropods (Ed. by R. Wehner), pp. 201 206. Berlin: Springer Verlag. yon Frisch, K. 1967. The Dance Language and Orientation of Bees. Cambridge, Massachusetts: Harvard University Press. von Frisch, K. & Lindauer, M. 1954. Himmel und Erde in Konkurrenz bei der Orientierung der Bienen. Naturwissensehaften, 41,245-253. Gould, J. L. 1984, The natural history of honey bee learning. In: The Biology of Learning (Ed. by P. Marler & H. S. Terrace), pp. 149 180. Berlin: Springer Verlag. Gould, J. L. 1985. How bees remember flower shapes. Science, N.Y., 227, 149~1494.

34

Animal Behaviour, 35, 1

Gould, J. L. 1986. Pattern lea~rningby honey bees. Anita. Behav., 34, 990-997. Hoefer, I. & Lindauer, M. 1976. Der Schatten als Hilfsmarke bei der Orientierung der Honigbiene. J. comp. Physiol., 122, 5-18. Ronacher, B. 1979. Beitrag einzelner Parameter zum wahrnehmungsgemfissen Unterschied von zusammengesetzten Reizen bei der Honigbiene. Biol. Cyber., 32, 77 83. Schnetter, B. 1972. Experiments on pattern discrimination in honey bees. In: Information Processing in the Visual System oJ'Arthropods (Ed. by R. Wehner), pp. 195 200. Berlin: Springer Verlag.

Seidl, R. & Kaiser, W. 1982. Visual field size, binocular domain, and the ommatidial array of the compound eyes in worker honey bees. J. comp. Physiol., 143, 17 26. Tinbergen, N. & Kruyt, W. 1938. Uber die Orientierung des Bienenwolfes III. Die Bevorzugung bestimmter Wegmarken. Z. vergl. Physiol., 25, 292-334. Wehner, R. 1981. Spatial vision in arthropods. In: Handbook of Sensory Physiology (Ed. by H. Autrum), pp. 287-616. Berlin: Springer Verlag.

(Received 25 September 1985," revised 16 December 1985," MS. number: A4628)