V. The Role of Motion Concepts in Avian Cognition 

In the preceding two sections of this chapter, we have shown that birds, or at any rate all birds that have been tested, have a sophisticated capacity for recognizing motion, and for recognizing objects in motion. It would have been surprising if it had been otherwise, but we needed the demonstration in the laboratory to be able to examine the underlying mechanisms. But the subtext has been that movement might pose special problems for the visual system and for visual cognition. In this final section of the chapter, we return to one of the ideas with which we opened the chapter: that, so far from being a problem, movement might actually aid visual perception to maintain visual stability but, furthermore, also in object recognition.

As we have seen at the beginning of the chapter, movement does appear to help birds, as it helps humans, to segregate figure from ground. That is clearly important in recognizing that there is some object out there, and indeed in recognizing what it is. But can it also help in the analysis of the feature information? We have a speculative argument to advance, which we have put before (see Dittrich & Lea, 1998; Lea & Dittrich, 1999), but can illustrate better here. A key issue in pattern recognition is the extent to which a collection of features is analysed as if they constituted a single object. Biederman (1987) proposes a theory of pattern recognition that depends on what he calls "geons", higher in level than a simple feature like an edge or a corner, but below the level of a whole object, and with colleagues, he has successfully applied this kind of analysis to pigeons' discrimination of static outline drawings (Kirkpatrick-Steger, Wasserman, & Biederman 1996, 1998). But there are some well known cases in which pigeons do not show any apparent integration of features; for example, Cerella (1980) reported that pigeons showed no generalisation decrement when a cartoon of the "Peanuts" character Charlie Brown was replaced with a drawing in which the features were scrambled at random.

Our idea is that in some situations movement may facilitate the integration of a collection of features into a geon. This hypothesis links to a well known phenomenon in human visual perception, which we call 'motion coherence'; this covers the phenomena first described by the Gestalt psychologists as 'common fate' (e.g., Wertheimer, 1923), in which stimulus elements that move in the same way are perceived as belonging to a single whole. We prefer the term "motion coherence" because it includes cases where elements move in different ways that can nonetheless be interpreted by seeing them as belonging to the same object - as happens, for example, in Johansson's point-light "biological motion" displays. The 'motion integrators' that we postulated in section 2 would be an ideal mechanism to bring about such dynamic feature binding.

Our experiments (described in Section IV) demonstrating successful discrimination by birds of point-light stimuli, and generalisation to them from full-detailed moving images, already suggest that birds can integrate moving stimuli into perceived objects. But we have a second line of evidence. A standard experimental test of visual integration, much used in studies of infant perception, uses a stimulus in which a vertical object is occluded by a horizontal bar. Subsequent tests with no occluder present, allow response either to the entire, unoccluded object, or to the separated parts that were visible to either side of the occluder. Two experiments have used this paradigm with birds, and they obtained opposite results. Sekuler, Lee, and Shettleworth (1996) reported generalisation from an occluded bar to its two elements, but Lea, Slater, and Ryan (1996) reported generalisation to the whole object. Comparisons between experiments are always weak, and these two experiments differed in many ways: Sekuler et al. used a discrimination learning situation with adult pigeons, of the sort we have been discussing in this chapter,  while Lea et al. used imprinting with recently hatched chickens. Furthermore Lea et al's stimuli were brightly coloured, and provided a vivid background.  But the point we focus on is that in Sekuler et al.'s (1996) experiment, the occluded object was stationary, while in Lea et al's it was continually moving, and because of the background the movement was highly salient.

These two apparent complications of the situation, making the object move and providing a background, were suggested by our knowledge of what makes it more likely that very young human subjects will perceive an occluded object as an integrated whole. Obviously the hypothesis that movement plays a critical role in feature integration requires more direct experimental test, but it is encouraging that such different lines of evidence support it.

Why would movement aid feature integration? We argue that because movement stimuli inherently require feature integration, in the temporal as well as the spatial domain, the recognition of objects in motion has to be based on multiple-features processing at various levels. Only some aspects of this difficult problem related to the chapter's title can be addressed here. The very nature of movement stimuli has several implications for bird's ability to process such stimuli (see Dittrich et al. 1998). For example, no feature is exactly the same, from moment to moment. As an object moves, it presents different facets to the viewer, and movement either of the object or organism will enhance quite dramatically the chances of stimuli or features being hidden in the course of the movement. This means that multiple features inevitably have to be used in order to discriminate moving stimuli. It is well known that pigeons have difficulties with discriminations that depend on the simultaneous use of multiple features (e.g. Blough 1985, Fersen & Lea 1990); in the terms used in human cognition, one might say that in this respect pigeons have limited attentional resources (cf. Sutton & Roberts, this volume, Blough this volume; cf. Eriksen & Hoffman, 1972; Treisman, Kahneman & Burkell, 1983). It is postulated that motion information triggers some attentional process that makes integration across features easier - because without such a process, it would not be possible to discriminate moving stimuli at all (see Dittrich & Lea, 1998; Dittrich 1999). For to identify movements, it is definitely necessary to integrate feature information from successive views of the stimuli, which then may even occur at different locations with the visual field. Again in humans, we know that attention is strongly modulated by the spatial organization of the stimuli (e.g. Treisman, 1988). Similarly, it has been found that pigeons showed a superior localisation accuracy of target stimuli among distractors whenever dynamic changes of the target area were involved (Cook, Cavoto, Katz & Cavoto, 1997, see also Cook, this volume). Even display durations of only 100 ms were sufficient and resulted only in a moderate drop of discrimination performance despite swiftly changing stimulus conditions. Again these findings seem to support the view proposed here that quick dynamic changes rather enhance than hinder feature binding.

If moving stimuli inevitably involve integration of features, it makes sense to describe their recognition as inherently conceptual. The bird has to recognize the particular perceptual scene before it as an instance of a particular object, which can appear in many guises as it moves. One most important role of such motion concepts is to enable birds to recognise movement patterns and moving natural stimuli under different lightness and colour conditions as well as different viewing angles and distances (see Emmerton, 1986; Dittrich & Lea 1993; Cook & Katz, 1999). The ability to generalize across these different viewing conditions while moving can be seen as one important aspect of the problem of movement perception in birds when we assume that another side of the problem is the kind of retinal stimulation through the birds' own movements. In order to recognize moving patterns birds must be able to generalize between scenes in which the individual features have constantly changing spatial relationships. This necessary ability to use concepts to overcome variations the exact spatial organization of object features seems to challenge some findings about pattern recognition with purely static stimuli in pigeons, e.g. that pigeons' visual discrimination is purely based on piecemeal-type absolute feature processing. Recent findings have indicated that pigeons are particularly sensitive to the configuration of features and changes in the spatial relationship of features led to a sharp decrease in discrimination (e.g. Kirkpatrick-Steger & Wasserman, 1996; Wasserman et al. 1993, Watanabe & Ito, 1991). In most of these studies the spatial organization of the stimuli was changed in the test situation by scrambling the features of a stimulus (see Kirkpatrick, this volume). But scrambling is a relatively crude approach to studying the importance of spatial organization, and perhaps a direct comparisons between scrambling and the kind of variations in spatial organization that occur during motion is inappropriate. Also, attempts to understand and calculate pigeons' ability to process and recognize complex spatial relationships in terms of information theory are highly instructive (Young & Wasserman, this volume). Further lines of evidence for the configurational processing capacities in pigeons can be found in studies on multiple-cue or hierarchical stimulus processing (Sutton & Roberts; Cook, both this volume). Recently, evidence has been accumulating that pigeons are able to perceive both global and local aspects of concept stimuli. Clearly, these findings support a view that pigeons must have attended to more than one particular aspect or feature of the stimuli (Kirkpatrick-Steger et al. 1996, 1998). The future task will be to test experimentally the new predictions and hypotheses based on the ideas of 'motion integrators' and the role of attentional processes in animals' ability to perceive and recognize motion information.

Next section: Conclusions