Behavioural Processes 59 (2002) 185 /191 www.elsevier.com/locate/behavproc
Recognition of incomplete patterns by bumble bees J.P. Thivierge, C.M.S. Plowright *, T. Chan School of Psychology, University of Ottawa, Ottawa, Ont., Canada K1N 6N5 Accepted 10 July 2002
Abstract Bumble bees were trained to discriminate between two visual patterns, one of which was rewarding (S/) and one of which was unrewarding (S/). Subsequently, they were tested for discrimination between two non-rewarding patterns: the top halves of the training patterns, the bottom halves or the side halves. Three conditions were tested: (1) When the S/ was a star and the S/ was a circle, all halves of the star were chosen above chance level, which may reflect an unlearned preference for radial patterns. (2) When the S/ and S/ were reversed, the bottom half and the side half of the circle were chosen above chance level, but not the top half. (3) In the last condition, the S/ was again a circle, but the feeder tube was placed below the training pattern rather than above, and again the bottom halves were discriminated but neither the top nor the side halves were. In learning pattern discriminations, the ventral portion of the pattern is weighted more strongly than the dorsal portion, which enables recognition of incomplete patterns, and the weighting depends little on angle of approach. # 2002 Elsevier Science B.V. All rights reserved. Keywords: Bumble bees; Bombus impatiens ; Pattern learning; Pattern recognition
1. Introduction Bees are proficient learners of floral patterns (for reviews, see Wehner, 1981; Gould, 1988a, 1990; Srinivasan, 1994; Horridge, 1997, 1999). Recent research has been devoted to testing for recognition of transformations of learned patterns, mostly in an effort to understand how patterns are represented in memory and perhaps even mentally transformed. Examples include recognition of rotated patterns (in honey bees: Gould, 1990; in bumble bees: Plowright et al., 2001); recognition of
* Corresponding author. Tel.: /1-613-562-5800x4849; fax: /1-613-562-5147 E-mail address:
[email protected] (C.M.S. Plowright).
mirror image transformations or left/right reversals (in honey bees: Gould, 1988b; in bumble bees: Korneluk and Plowright, 1995; Plowright, 1997), and recognition of abstracted properties such as bi-lateral symmetry (Horridge, 1996) and orientation (van Hateren et al., 1990). In this paper we examine recognition of incomplete patterns by foraging bumble bees. In natural settings, such recognition might be adaptive if in between the time of learning and recognition, part of a flower should become missing or occluded. From a psychological point of view, if all halves (top, bottom and side) are not treated equally, this would suggest a bias in perception, attention or memory for one part of the flower or another. Indeed, a dorso-ventral asymmetry in the visual
0376-6357/02/$ - see front matter # 2002 Elsevier Science B.V. All rights reserved. PII: S 0 3 7 6 - 6 3 5 7 ( 0 2 ) 0 0 0 9 3 - 1
186
J.P. Thivierge et al. / Behavioural Processes 59 (2002) 185 /191
field of the honey bee has been documented (Wehner, 1972; Lehrer, 1997), though the visual field assigned to a recognition task can be expanded to include not only the ventral but also the dorsal portions (Giurfa et al., 1999). In the following study on bumble bees, bees were trained to discriminate between an S/ and an S/ which were presented simultaneously. They were subsequently shown just the bottom halves, top halves or side halves. Both halves were unrewarding. The bees’ choices were examined. To ensure that above chance performance on the choice test could be traced to learning of the original discrimination and not to an unlearned preference, the training stimuli used for the S/ and S/ were interchanged in between two conditions. We hypothesized that bumble bees would recognize at least the bottom halves of learned patterns, as do honey bees. We also examined whether other halves (top half, side half) might be discriminated as well. Finally, if learning occurs just before landing (Gould, 1988a) and if what is learned depends on the view taken during approach, then the position of the feeder tube ought to affect half-pattern recognition. Reference to the possibility that the way in which a bee approaches a flower might, or might not, play a role in pattern recognition can be found in the literature (Giurfa et al., 1999, p. 322), but we could find no empirical investigation on the matter. We tested the hypothesis that bees should be more likely to recognize the top half of a pattern if the feeder was placed right above the training pattern rather than below it.
cm) which was connected to a flight cage by a 32 cm long walkway. For one condition (Circle S// Feeder below), a screened flight cage (175 /175 / 175 cm) was used, and for the others, to make it easier to catch the bees, a smaller clear plexiglass cage was used (63 /63/63 cm). A vertical gate in the walkway allowed the experimenter control over which workers entered and exited the colony. Workers were individually marked with colored numbered plastic disks glued to the thorax. To stimulate foraging, the colonies were deprived of nectar until the surface honey pots were empty. Pollen was available ad libitum. Fluorescent bulbs and one incandescent bulb provided lighting. 2.2. Stimuli The patterns were a star of David and a circle (3.5 cm in diameter) made of yellow construction paper covered in vinyl. A pattern was vertically mounted on a circular plastic disk (10 cm in diameter) fixed on a vertical stem. A glass feeding tube protruded either 1.5 cm above (for the two ‘Feeder above’ conditions) or below the pattern (for the one ‘Feeder below’ condition). The tube was filled either with sugar solution (2:1 sugar and water by volume) or water (S/). The patterns were placed at the back of the cage at approximately the same level as the nest entrance. In the test conditions, the training patterns were not taken and cut in half, but rather new half patterns were constructed: top halves, bottom halves and side halves (only the left half was used in the ‘Feeder below’ condition, but both halves were used for the first two conditions). The feeder tubes remained but dispensed no reward.
2. Methods 2.3. Training 2.1. Subjects Four colonies of Bombus impatiens with no prior foraging experience were used. A new colony was used for each condition tested successively, and one condition (Circle S//Feeder above) required two colonies after the first died before testing was completed. The three experimental conditions are summarized in Fig. 1. The colonies were each housed in a wooden box (30 /15/15
During training the colony was given unrestricted access to the flight cage in which were placed 1/4 pairs of patterns: one S/ and one S/ in each pair. The positions of the S/ and S/ were switched every 15 min to encourage attention to the patterns rather than the spatial positions of the patterns. The stimuli were wiped clean to minimize scents. Training of 5 h/day occurred over 3 /7 days.
J.P. Thivierge et al. / Behavioural Processes 59 (2002) 185 /191
187
Fig. 1. Patterns used in the three conditions. The small circles (filled if reward was dispensed, empty if not) marks the position of the feeder.
2.4. Testing On each day of testing the bees were first given a ‘refresher’ of unrestricted access to the S/ and S/ for 1/1.5 h. Only bees that were observed at least once (and usually more often) landing on at least one of the stimuli during the original training or a refresher training period were subsequently tested with the half patterns. Before testing with the half patterns a check that the bees could make the discrimination with the whole patterns was undertaken with empty flowers. Half patterns were then used in a random order for the two ‘Feeder above’ conditions (so for example, day 1 could be bottom halves, day 2 left halves, day 3 first bottom then top halves, etc.) until a minimum of seven bees were tested (each making 1 /25 choices) and a minimum of 35 landings recorded, which took about a week. For the ‘Feeder below’ condition, the bees foraged so actively that each half pattern could be completed in 1 day: top halves on day 1, bottom halves on day 2 and left halves on day 3. A landing on a flower was recorded as a choice, after which the bee was returned to the colony. 2.5. Statistical analysis As in our previous work on pattern discriminations (Korneluk and Plowright, 1995; Plowright, 1997; Plowright et al., 2001) a replicated goodnessof-fit test with the G-statistic was used (Sokal and Rohlf, 1981) because the data were counts with replication within individual bees. For each con-
dition, two G -values were obtained: GH which tests for heterogeneity, or individual differences, and GP which tests whether the pooled data (the data as a whole) deviate from an expected proportion, which is a chance level of 50:50 in all conditions. In the tests of significance, the G test statistics were compared to a chi-square value. The GP values were the ones of primary interest, because they indicate a group discrimination or not. A significant GH, which would indicate a significant individual deviation from the group proportion could in principle arise for a variety of reasons, including: (1) if there were differences in the extent of original learning of the S/ versus S/ discrimination (some bees being more active than others); and (2) if a short experience in one test condition carried over to another (some bees appeared in more than one condition), though no evidence was found for such carry over in between short testing conditions in similar experiments (van Hateren et al., 1990; Plowright, 1997; Korneluk and Plowright, 1995; Plowright et al., 2001). The analysis above compared choice proportions to a theoretical value but did not compare two proportions between conditions. To do this we fit a logistic model to the data because it specifies a binomial error term. Generalized linear interactive modeling (GLIM) statistical software was used (Francis et al., 1993). The tests of significance are x2 tests (but note that these are not x2 tests of independence). The variation due to individual bees was taken into account prior to testing for the effects of interest.
188
J.P. Thivierge et al. / Behavioural Processes 59 (2002) 185 /191
3. Results
3.1. Star S//Feeder above Fig. 2 shows the proportion of flower visits for the whole pattern and the four half-pattern tests. The total number of landings and number of bees in each condition are given in brackets above each bar. The frequencies of choices for the complete star and for each of the half patterns were all significantly above the chance proportion of 0.5 (with 1 df for each GP test: GP /6.25, P B/0.025 for the whole patterns; GP /8.40, P B/0.005 for the bottom halves; GP /10.46, P B/0.005 for the top halves; GP /46.10, P B/0.001 for the right halves and GP /29.45, P B/0.001 for the left halves). No individual differences were found: GH was always non-significant, and so possible individual differences in the extent of learning of the patterns during training are of little concern. The gradual increase from left to right in Fig. 2 does not reflect learning within the experiment because the conditions were interspersed over about a week and did not proceed in the sequence on the graph.
Fig. 2. Choice proportions for the Star S//Feeder above condition. The number of bees in the test and the number of landings on which the choice proportion is based are given in brackets above each bar. The choice proportions marked by an asterisk were significantly different from chance.
The choice proportions for the whole patterns, bottom halves and top halves all seem comparable in Fig. 2, but the choice proportions for the left and right halves are particularly high. A comparison between the choice proportions for the whole and the left half patterns revealed a significant difference (x2(1) /5.52, P B/0.025). If anything the difference between the whole patterns and right halves was larger. 3.2. Circle S//Feeder above When the S/ and S/ were reversed from the condition shown in Fig. 2, not all pattern halves were recognized (Fig. 3). The frequency of choices for the complete circle was significantly above chance (GP with 1 df/35.70, P B/0.001). With respect to the half patterns, only the bottom half was recognized (GP with 1 df /9.55, P B/0.005). The choice proportions of the top half, left half and right half did not deviate significantly from chance (all the GP values were non-significant). No individual differences were found (GH was non-significant), except for the right-half pattern where GH was significant (GH with 14 df/41.76, P B/0.001). Partitioning the total G value into
Fig. 3. Choice proportions for the Circle S//Feeder above condition. The number of bees in the test and the number of landings on which the choice proportion is based are given in brackets above each bar. The choice proportions marked by an asterisk were significantly different from chance.
J.P. Thivierge et al. / Behavioural Processes 59 (2002) 185 /191
contributions from the 15 bees tested on the right half pattern showed that four of their choice proportions deviated significantly from the group proportion: Three of them in favor of the S/ and one in favor of the S/. 3.3. Circle S//Feeder below When the feeders were placed below the patterns used in the condition shown in Fig. 3, rather than above, the same general pattern of results was obtained (Fig. 4). The frequency of choices for the complete circle was significantly above chance (GP with 1 df /7.95, P B/0.005). The bottom half was chosen above chance level (GP with 1 df /4.19, P B/0.05), as was the left half (GP with 1 df /5.00, P B/0.05). The top half was not recognized, however, (GP with 1 df /1.57, NS). No individual differences were found: GH was always nonsignificant. A comparison of the last two conditions shows that for the top halves, the choice proportion was slightly above 0.5 in Fig. 3 and below 0.5 in Fig. 4. Although these proportions did not differ significantly from chance, there might still be a main effect of feeder position if the two conditions were
Fig. 4. Choice proportions for the Circle S//Feeder below condition. The number of bees in the test and the number of landings on which the choice proportion is based are given in brackets above each bar. The choice proportions marked by an asterisk were significantly different from chance.
189
compared to each other. Such was not the case. The logistic analysis revealed only a main effect of pattern half (top vs. bottom) with a higher choice proportion for the bottom half (x2(1) /8.21, P B/ 0.005). The effect of feeder tube placement was not significant (x2(1) /0.84) and neither was its interaction with pattern half (x2(1) /0.35). In Figs. 3 and 4, the choice proportions for the bottom halves seem comparable to those for the whole patterns. Indeed, no significant difference was detected (x2(1) /0.75, NS).
4. Discussion In all three conditions, bumble bees were trained to make a pattern discrimination and then tested for discrimination of various halves of the patterns. In the condition where the S/ was a star, and the S/ was a circle, however, the results do not justify the conclusion that the bees recognized the half patterns as being incomplete versions of the whole patterns that were learned. The choice proportions in all the tests were high (between 72 and 97%); in some cases even higher than for the whole patterns (Fig. 2). While this may be due to the fact that bees were given ‘refresher’ training sessions each day, the same was not observed in the other two conditions, where performance on the half patterns was the same or below that on the whole patterns. Our results more likely reflect an unlearned preference for radial patterns in bumble bees (Simonds and Plowright, unpublished data), as has been shown for honey bees (Lehrer et al., 1995): the bees might have chosen the star halves even if they had not been rewarded on the complete stars during training. If so, then the proper conditions for examining half-pattern recognition and the possible role of feeder position are the other two conditions, where the training situation (Circle S/ and Star S/) was biased against any unlearned preferences. Before turning to the two other conditions, however, one unpredicted aspect of the data shown in Fig. 2 requires an explanation: the particularly high choice proportions for the side halves of the stars. One possible post-hoc account is as follows. In the whole patterns, the top halves
190
J.P. Thivierge et al. / Behavioural Processes 59 (2002) 185 /191
and the bottom halves, there is symmetry about the vertical axis, and previous research (Lehrer et al., 1995) has shown an unlearned preference by bees for vertically symmetrical stimuli. For choices between whole patterns, between bottom halves and between top halves, the star is preferred because it is a radial pattern, yet the whole and half circles are not completely aversive because at least they have symmetry about the vertical axis */ in other words the vertical symmetry serves to equalize the attractiveness of the stimuli. The side half patterns, however, are not vertically symmetrical and so the side half circle has even less to recommend it (from the bees’ point of view) than either the full circle or the other circle halves. We suggest that the very high choice proportions for the side halves of the star are not so much the result of the attractiveness of those stimuli as the unattractiveness of the alternatives. This account is in line with our suggestion that the choice proportions in the Star S/ condition reflect unlearned preferences. The two ‘Circle S/’ conditions allowed us to examine what, if anything, the bees had learned during the training phase. We expected, that bumble bees, like honey bees, would recognize the bottom half of the pattern, and this expectation was confirmed for both conditions (Figs. 3 and 4). The choice proportions for the bottom halves were no different from the choice proportions for the whole patterns: generalization was complete. The top halves of the patterns were not discriminated, though evidence for such a discrimination has been obtained for honey bees (Giurfa et al., 1999). A variety of differences in the procedures between our study and that of Giurfa et al. preclude any direct comparisons: for example, their stimuli were considerably larger than ours, and their bees were tested in a Y-maze. Their procedure might have forced encoding of a larger view of the whole patterns during training. One aspect of the methods that is comparable to their study, however, is that the S/ was presented not alone but in the presence of a comparison S/; and so at least we may conclude that simultaneous presentation of an S/ and S/ does not necessarily lead to top half pattern recognition.
No support was found for the hypothesis that manipulating feeder position would influence pattern recognition. In addition to the obtained bottom-half recognition, top-half recognition had been expected when the feeder was placed just above the training pattern rather than just below it, if for no other reason than the top half of the pattern was closer to the bee than the bottom half just before landing and during feeding. Nonetheless, the bees did not discriminate between the top halves of the star and circle in either case: the effect of feeder tube position was in the expected direction but it was not significant. To the list of the abilities of bumble bees in pattern recognition, we may add that incomplete patterns are recognized, but that not all incomplete patterns are equal. The bottom half of the pattern is privileged over the top half (Figs. 3 and 4). One caveat is that our conclusions are confined to behaviour: further research would be required to determine at what point in the information processing pathway the differential weighting of pattern halves occurred. The side half was recognized inconsistently: above chance performance was found in the ‘Circle S//Feeder below’ but not the ‘Circle S//Feeder above’ condition. The way in which bees approach a pattern is not an important factor, and research efforts to explain incomplete pattern recognition might be better placed elsewhere.
Acknowledgements This research was supported by a research grant to C.M.S.P. and a summer research scholarship to J.P.T. from the Natural Sciences and Engineering Research Council of Canada.
References Francis, B., Green, M., Payne, C., 1993. The GLIM System, Release 4 Manual. Clarendon Press, Oxford. Giurfa, M., Hammer, M., Stach, S., Stollhoff, N., MullerDeisig, N., Mizyrycki, C., 1999. Pattern learning by honeybees: conditioning procedure and recognition strategy. Anim. Behav. 57, 315 /324.
J.P. Thivierge et al. / Behavioural Processes 59 (2002) 185 /191 Gould, J.L., 1988a. Resolution of pattern learning by honey bees. J. Insect Behav. 1, 225 /233. Gould, J.L., 1988b. A mirror-image ‘ambiguity’ in honey bee pattern matching. Anim. Behav. 36, 487 /492. Gould, J.L., 1990. Honey bee cognition. Cognition 37, 83 /103. Horridge, G.A., 1996. The honeybee (Apis mellifera ) detects bilateral symmetry and discriminates its axis. J. Insect Physiol. 42, 755 /764. Horridge, G.A., 1997. Spatial and non-spatial coding of patterns by the honey-bee. In: Srinivasan, M.V., Venkatesh, S. (Eds.), From Living Eyes to Seeing Machines. Oxford University Press, Oxford, pp. 52 /79. Horridge, G.A., 1999. Two-dimensional pattern discrimination by the honeybee. Physiol. Entomol. 24, 197 /212. Korneluk, Y.G., Plowright, C.M.S., 1995. Mirror image pattern matching by bumble bees. Behavior 132, 87 /93. Lehrer, M., 1997. Honeybee’s use of spatial parameters for flower discrimination. In: Dafni, A., Giurfa, M., Menzel, R. (Eds.), Insect Vision and Flower Recognition. Laser, Jerusalem, pp. 157 /168. Lehrer, M., Horridge, G.A., Zhang, S.W., Gadagkar, R., 1995. Pattern vision in bees: preference for radial patterns. Phil. Trans. Roy. Soc., Lond. B 337, 49 /57.
191
Plowright, C.M.S., 1997. Function and mechanism of mirror image ambiguity in bumble bees. Anim. Behav. 53, 1295 / 1303. Plowright, C.M.S., Landry, F., Church, D., Heyding, J., Dupuis-Roy, N., Thivierge, J.P., Simonds, V., 2001. A change in orientation: recognition of rotated patterns by bumble bees. J. Insect Behav. 14, 113 /127. Sokal, R.R., Rohlf, F.J., 1981. Biometry, 2nd edition. W.H.Freeman, New York. Srinivasan, M.V., 1994. Pattern recognition in the honeybee: recent progress. J. Insect Physiol. 40, 183 /194. van Hateren, H.J., Srinivasan, M.V., Wait, P.B., 1990. Pattern recognition in bees: orientation discrimination. J. Comp. Physiol. A 167, 649 /654. Wehner, R., 1972. Dorsoventral asymmetry in the visual field of the bee, Apis mellifera . J. Comp. Physiol. 77, 256 /277. Wehner, R., 1981. Spatial vision in arthropods. In: Augrum, H. (Ed.), Handbook of Sensory Physiology, vol. VII/6C, Comparative Physiology and Evolution of Vision in Invertebrates, C: Invertebrate Visual Centers and Behavior II. Springer-Verlag, Heidelberg, pp. 287 /616.
