«9 The Role of Associative Processes in Spatial, Temporal, and Causal Cognition Aaron P. Blaisdell University of California, Los Angeles, United ...»
9 The Role of Associative Processes
in Spatial, Temporal, and Causal
Aaron P. Blaisdell
University of California, Los Angeles, United States
Associative processes build the structural-representational framework upon which
cognitive processes of computation and inference can act. I review evidence I have
collected showing how associative processes are involved in building spatial, temporal, and causal maps. Evidence comes from studies on simple associative acquisition such as Pavlovian and instrumental conditioning, higher-order conditioning procedures such as sensory preconditioning and conditioned inhibition, and from cue-competition studies. Parallels are drawn between acquisition and integration of information in conventional associative paradigms on the one hand and cognitive paradigms on the other.
1 Introduction One of the great psychological debates of the twentieth century involved an exchange between Tolman and Guthrie. Tolman, originally a behaviorist himself, conducted experiments with rats that lead him to develop a nascent cognitive framework during the first half of the twentieth century—a period dominated by the S-R behaviorist ideology. He suggested that rats held expectations about impending events, rats could learn without explicit (i.e., food) reinforcement, and that rats formed cognitive maps while navigating a maze. Tolman was ridiculed by many of his colleagues for these heretical notions. Guthrie, one of behaviorism’s chief proponents, even accused Tolman of “leaving the rat buried in thought” (Guthrie, 1935, p. 172). Tolman planted the seed, however, that led to the cognitive revolution of the 1960s and 1970s. The cognitive framework finally found acceptance and hypotheses about mental states and cognitive processes in both humans and animals became commonplace. Today, the cognitive framework is the dominant ideological stance, though currents of behaviorism still exist. This is not to say that one framework, such as the cognitive, provides a more accurate depiction than the other, such as behaviorism. Rather, both frameworks continue to have heuristic value, and the tension between them epitomizes the Aaron P. Blaisdell Hegelian dialectic—a process that can lead to a synthesis of new ideas and enlightenment.
In fact, the more we understand about the neural underpinnings of learning and behavior, the more difficult it is to distinguish between behaviorist and cognitivist explanations of psychological processes.
Despite the acceptance of cognitive explanations of behavior in general, learning theorists have been slow to adopt a truly rich cognitive framework. Even now, the dominant conventional view is that only the strength of the CS-US association is encoded during Pavlovian conditioning. Likewise, theories of instrumental conditioning are largely focused on the strengths of S-R and R-O associations with only a minority focus on the quality of these associations. Evidence has been accumulating over the past two decades, however, suggesting that subjects acquire a much richer representation of their experiences during associative learning. These representations include information about time, space, and qualitative attributes of a CS and US (e.g., Blaisdell, Denniston, and Miller, 1997). Moreover, there is evidence that both humans and animals learn cause-effect relationships during even the most simple of Pavlovian procedures. In this chapter, I review this accumulating evidence presenting key examples from my own laboratory and from my work with Dr. Ralph Miller.
2 Spatial Cognition
Since its introduction by Tolman (1948), the cognitive map has gained widespread use as a conceptual tool for understanding spatial memory and cognition. Spatial abilities are fundamentally important for navigating the world in order to migrate, avoid dangers such as predators, and to locate biological necessities such as food, shelter, and mates. The concept of a cognitive map usefully describes many aspects of spatial behavior and continues to facilitate the discovery of new behavioral phenomena and processes (Gallistel, 1990; Healy, 1998;
Shettleworth, 1998). The cognitive map has also been useful in understanding processes of timing in associative memory (Honig, 1981, see Section 3). One important feature of cognitive maps is that they can be used to compute novel routes between separate spatial locations (Tolman, 1948). Despite some criticisms (Bennett, 1996; Gibson and Kamil, 2001; Shettleworth, 1998), I have found the cognitive map to be a useful concept in describing the results of recent experiments from my and others’ laboratories.
For example, the demonstration that pigeons learn to use visual landmarks on a touchscreen or in an open field to find a hidden goal supports the interpretation that they learned the spatial vector between the landmark and the goal (Blaisdell and Cook, 2005; Cheng, 1994; Cheng and Spetch, 1995, 1998; Kamil and Cheng, 2001; Sawa, Leising, and Blaisdell, 2005; Spetch, Cheng, and MacDonald, 1996; Spetch et al., 1997; Spetch, Cheng, and Mondloch, 1992; Spetch and Mondloch, 1993). A vector is a metric encoding both distance and direction between two points in space occupied by specific objects. The direction is coded in reference to a larger framework of orientation, such as the sides of the touchscreen monitor or the walls of the room containing the open field. A vector is easily conceptualized as an allocentric spatial map between two objects: A and B. Object A may be a junk object located in the open field and Object B may be a food goal buried under sand on the floor of the
The Role of Associative Processes in Spatial, Temporal, and Causal Cognition
open field. Alternatively, Object A may be a colored shape presented on the surface of the monitor and Object B may be a spatial location on the monitor at a fixed distance and direction relative to Object A. Work carried out in my lab has investigated the role of associative processes in the acquisition and expression of spatial maps. The fundamental issue is whether learning the spatial relationship between objects, such as landmarks and goals, obeys the same principles as learning and expression of associations in conventional Pavlovian and instrumental conditioning procedures. We adopted a method used by others of testing specific functional parallels between spatial learning and conventional associative learning. The existence of functional parallels strongly suggests a common process underlying learning in both domains.
2.1 Acquisition of Spatial Maps
Figure 1 (inset) shows an example of a procedure used in our lab to train pigeons to use a visual cue presented on a touchscreen in an operant box as a landmark to a hidden goal also located on the touchscreen. Pigeons were reinforced with mixed gain from a food hopper below the touchscreen for pecking at the goal location which could be any one of 56 dots in an 8 x 7 grid. Each dot was centered within a 2-cm2 response area that served as a possible goal location. The goal was initially marked with the 2-cm2 white square that was gradually faded out until only the dot was visible (see Sawa et al., 2005 for details). The training landmark (denoted by the “T” in Figure 1) bore a fixed spatial location relative to the goal. By the completion of fading out the goal marker, subjects were able to locate the hidden goal based solely on its spatial relationship with the training landmark. Because the location of the goal was randomly determined on each trial, subjects were not able to predict its location on a given trial other than to attend to the landmark. The main panel of Figure 1 shows peck location data collected on 30-s nonreinforced probe trials with LM T. The strong spatial control over pecks by LM T shows that pigeons had encoded the T➔Goal spatial vector.
The benefit of studying allocentric representations of space on the touchscreen is that it is virtually free of confounding spatial processes that are engaged when navigating threedimensional space, such as optic flow, dead reckoning, and motion parallax. Nevertheless, it is important to verify the validity of results from touchscreen experiments by conducting similar tests in a more ecologically valid setting such as the open field. To this end, some of the experiments reported here include replications in an operant open-field procedure called ARENA recently developed in our lab (Badelt and Blaisdell, 2008; Leising, Garlick, Parenteau, and Blaisdell, in press).
2.2 Integration of Spatial Maps
If a Landmark-Goal spatial map is encoded during first-order associative conditioning in which two events are directly paired, then multiple spatial maps between landmarks and goals could be integrated during higher-order associative conditioning such as second-order conditioning and sensory preconditioning. In second-order conditioning, a CS1-US association is learned prior to a CS2-CS1 association. In sensory preconditioning, the CS2-CS1
Aaron P. Blaisdell
Figure 1. Inset: An example of two trials during initial training of the landmark-based search task.
The dots mark the center of each unit of the response grid. The white square marked the goal location.
“T” is the training landmark. Main panel: Total number of pecks to the screen on nonreinforced probe trials with LM T. “G” marks the screen location (relative to LM T) where pecks were reinforced on training trials. This location was unmarked during nonreinforced test trials. The arrow (not visible during the trial) indicates the LM T➔Goal spatial map. Distribution of pecks shown separately for X and Y screen axes. Data were pooled across trials so that the goal was zeroed to location (0, 0).
Figure 2. Left panel: LM A and LM B are paired without food in Phase 1 of sensory preconditioning.
Middle panel: LM A signals the location of the hidden goal (“G”) in Phase 2. Right panel: Hypothetical maps at test (not to same scale as other two panels.)
association is learned prior to the CS1-US association. According to this associative integration hypothesis of cognitive map formation, complex spatial representations can be built by linking together simpler representations that share common elements. The simplest conceivable spatial association encodes the spatial relationship between two events: A and B, such as two landmarks, or a landmark and a food goal (see Section 2.1). A spatial representation containing three events (A, B, and C) can be built in one of two ways. On the one hand, all three events could be presented simultaneously, in which case the subject could construct a spatial or configural representation containing all three elements. For example, presenting LMs A and B together with a food goal could establish a spatial map containing all three elements. On the other hand, the same three-element spatial map could be constructed in a piecemeal fashion by joining together two simpler representations, each containing two of the three elements. This process would allow subjects to construct the same three-element representation without experiencing all three elements at the same time. An integrated map allows the subject to extrapolate novel relationships beyond its direct experience.
Consider the example in Figure 2 (from Sawa et al., 2005). Pigeons received pairings between two visual landmarks (A and B; actual landmarks were colored geometric shapes) in Phase 1 of sensory preconditioning. The screen location of the pair of landmarks varied across trials, but they always bore the same spatial relationship to each other. Pigeons then received first-order conditioning in Phase 2 consisting of A➔Goal pairings. The screen location of the goal was randomly determined from trial to trial. LM A maintained a stable spatial relationship to the goal, thereby signaling the goal location. After pigeons were reliably finding the goal in the presence of LM A, pigeons received nonreinforced test trials with LM B alone which terminated after 30 s without food. We recorded the screen location of all pecks during these test trials. If pigeons had acquired the B➔A map during Phase 1 and the A➔Goal map during Phase 2, then they should search the location one grid unit to the left of LM B (right panel of Figure 2; arrows represent spatial maps acquired between the landmarks and the goal; events enclosed in quotes indicate memories retrieved by way of the associations between events).
Figure 3 shows the results of LM B test trials. The response density peak was located one grid unit to the left of LM B. Thus, through an inference-like process the pigeon arrived at the three-item representation without the concurrent presentation of all three elements (A, B, and goal). That is, associative integration allowed subjects to compute a novel B-Goal spatial relationship despite the fact that B had never been directly paired with the goal. These results suggest that associative learning may serve as a mechanism for the acquisition and expression of spatial behavior guided by both simple and complex maps. According to the associative integration hypothesis, only maps that share linking, common elements are bound together. In our example, the B➔A and A➔Goal associations were integrated into a B➔A➔Goal map through the common element LM A. Unpaired controls (not shown) revealed that associative integration depended on the Phase 1 and Phase 2 pairings (see Sawa et al., 2005). Similar findings have been reported by Blaisdell and Cook (2005) and Chamizo, Roderigo, and Mackintosh (2006).
Aaron P. Blaisdell
Figure 3. Test trial data for LM B plotted both as a density plot of screen locations pecked and as separate frequency histograms for the X and Y screen axes.
The location of LM B in relation to the predicted location of the goal (0, 0) based on integration of the B➔A and A➔Goal spatial maps.
Sawa, Leising, and Blaisdell, 2005. Sensory preconditioning in spatial learning using a touch screen task in pigeons. Journal of Experimental Psychology: Animal Behavior Processes 31, 368-375, APA, adapted with permission.
Figure 4. Top panel depicts the overshadowing procedure separately for AX+ and Y+ trials.
Bottom panel shows test trial types separately for A-, AX-, X-, and Y- trials.