Behavrourul
Processes,
105
14 (1987) 105-122
Elsevier COMPUTER SIMULATION OF PIGEONS' PERFORMANCE ON A SPATIAL MEMORY TASK DONALD M. WILKIE and DUNCAN J. KENNEDY Psychology Department, University of British Columbia, Vancouver, Canada V6T lW5. (Accepted 10 November 1987) ABSTRACT
Wilkie, D.M., and Kennedy, D.J. Computer simulation of pigeons' performance on a spatial memory task. Behav. Processes 14: 105-122. Delayed matching of key location is a useful paradigm for the study of pigeons' short-term memory for a spatial location. On each trial a randomly selected key from a matrix of keys is lit briefly as a sample and followed by a retention interval. During the ensuing choice period the sample and one of the non-sample keys are lit; choice of the sample is correct and rewarded whereas choice of the distractor key is not. The computer simulation of performance on this task is based on a simple model: We assume that the pigeon has knowledge of the location of the keys stored in a map-like reference memory. We also assume that short-term memory involves an attention focus or "pointer" that "drifts" on the surface of this map. The pointer migrates from a randomly determined position during the intertrial interval towards the location of the sample when this stimulus is presented. It wanders randomly from its previous position when this cue is no longer present in the retention interval. During the test for retention the bird selects the location (i.e., sample or distractor) closer to the location of the pointer on the map of the matrix. The simulation successfully reproduced several of the phenomena observed in delayed matching of location experiments and provided an account of some hitherto perplexing results. As well, the model successfully predicted some new empirical data.
INTRODUCTION The recent resurgence of interest in how animals remember the spatial location of
such
things as food patches has spawned the development of several
laboratory paradigms to study spatial memory. One of these capitalizes on the fact that certain bird species cache pieces of food, which are subsequently recovered (e.g., Vander Wall, 1982). Other paradigms make use of mazes such as the radial-arm maze (e.g., Olton, 1978). the Morris water maze (e.g., Morris, 1981; Sutherland h Dyck, 1984), and the T-maze (e.g., Olson & Haki, 1983) and require subjects to remember food locations or places of safety. Another paradigm, developed in our laboratory, is a variant of delayed matching to sample. Nine pecking keys, arranged as a three by three matrix, serve as the spatial cues. Trials begin with a "ready" stimulus (a brief
0376-6357/87/$03.50
0 1987 Elsevier Science Publishers B.V.(Biomedical Divmion)
106
operation of the grain feeder or dimming of the houselight); then a randomly selected key is lit briefly as a sample. After a short delay the sample key is lit again along with a non-sample key. A peck at the sample key produces grain reinforcement,whereas a peck at the non-sample key leads only to the intertrial interval. Several studies using the delayed matching of key location procedure have been performed (e.g., Kamil & Balda, in press; Smith, Attwood, h Niedorowski, 1982; Wilkie, 1983a, 1983b, 1984, 1986, in press; Wilkie h Summers, 1982) and several basic facts about pigeons' memory for key location have been established. Consequently, it now seems appropriate to develop some hypotheses about the memory process that results in accurate matching on the delayed matching to key location task. The model that we propose--a "drift" model--was selected primarily because 1) it is conceptually simple and therefore meets the demand of parsimony, and 2) it has an integral spatial component and therefore seems appropriate as a model of spatial memory. The model shares features with one proposed by Roitblat (1984) for pigeons' memory for visual stimuli such as colors in a conventional delayed matching to sample task. The essential components of the model are as follows: 1. The pigeon is assumed to have knowledge of familiar locations--i.e., 1 the key matrix--stored in a stable map-like long-term or reference memory. 2.
The short-term or working memory process is assumed to involve an
attention focus or "pointer" that moves ("drifts") over the surface of the map of the matrix. 3.
This pointer migrates toward the representationof the sample location
during the period of time that the sample key is physically present. 4.
The pointer wanders randomly from its previous location when the
sample is no longer present (i.e., during the retention interval). 5.
During acquisition the pigeons learn the rule to choose the key
location whose representation is closer to the location of the pointer. 6.
During the intertrial interval the pointer is positioned randomly on
the key matrix representation. The model incorporates two widely-acceptednotions of animal memory processes: a) that there is a distinction between long- and short-term (or
lBecause the pigeons directly face the matrix an often neglected problem in most cognitive map theories - how animals transform frontal views of objects in their environment into bird's eye or survey map views - is absent (cf. Strelow, 1985).
107
reference and working) memory, and b) that memory is often prospective rather than retrospective (cf. Honig & Thompson, 1982). A representationof the key matrix is in a map-like reference memory. The pointer location comprises information that must be retained during the trial but discarded before the next trial. The model is prospective in the sense that information concerning the correct choice is carried forward in time rather than being stored and then retrieved at a later point in time.
COMPUTER SIMULATION OF DRIFT MODEL Because the drift process is dynamic, computer simulation seemed ideally 2 suited to evaluate the model. Although computer simulation is a technique widely used in such fields of study as human cognition (e.g., Metcalfe-Eich, 1982) and behavioral ecology (e.g., Chantrey, 1982), with rare exceptions (e.g., Daly 6 Daly, 1984; Hinson 6 Staddon, 1983; Shimp, 1984) this powerful technique has not been used by those studying animal learning and cognition. Accordingly, another rationale for the present project was to demonstrate the utility of computer simulation as a tool in the study of animal memory processes. Simulation of Drift Model An initial step in developing a computer model is to translate the model into a programming flowchart. We have done this in Fig. 1.
In this depiction
time flows from top to bottom. In a given trial the pointer is first moved to a randomly selected spot on the two-dimensionalmap-like reference memory of the key matrix. Once the intertrial interval is over and the sample key is lit the pointer begins to migrate towards the location of the sample key on the reference representation. Once the sample presentation is over, the pointer begins to move randomly from the spot it occupied at sample offset. This wandering continues until the retention interval ends and the choice stimuli appear. It is then assumed that the pigeon compares the distance between the pointer location and the sample key with the distance between the pointer location and the distractor key, and then pecks at the key closer to the pointer. The peck is followed by reward and/or the intertrial interval.
zLehman (1977), Apter (19701, and Dutton G Starbuck (1971) provides good reviews of the theory and practice of computer simulation in the behavioral sciences. An earlier paper by Gregg and Simon (1967) also provides excellent background.
108
6
INTERTRIAL INTERVAL OVER? 4 YES : SAMPLE ‘-t_ PRESENTATION OVER? YES
tNO--j 1 NO+?%i%i+
Processes,
105
14 (1987) 105-122
Elsevier COMPUTER SIMULATION OF PIGEONS' PERFORMANCE ON A SPATIAL MEMORY TASK DONALD M. WILKIE and DUNCAN J. KENNEDY Psychology Department, University of British Columbia, Vancouver, Canada V6T lW5. (Accepted 10 November 1987) ABSTRACT
Wilkie, D.M., and Kennedy, D.J. Computer simulation of pigeons' performance on a spatial memory task. Behav. Processes 14: 105-122. Delayed matching of key location is a useful paradigm for the study of pigeons' short-term memory for a spatial location. On each trial a randomly selected key from a matrix of keys is lit briefly as a sample and followed by a retention interval. During the ensuing choice period the sample and one of the non-sample keys are lit; choice of the sample is correct and rewarded whereas choice of the distractor key is not. The computer simulation of performance on this task is based on a simple model: We assume that the pigeon has knowledge of the location of the keys stored in a map-like reference memory. We also assume that short-term memory involves an attention focus or "pointer" that "drifts" on the surface of this map. The pointer migrates from a randomly determined position during the intertrial interval towards the location of the sample when this stimulus is presented. It wanders randomly from its previous position when this cue is no longer present in the retention interval. During the test for retention the bird selects the location (i.e., sample or distractor) closer to the location of the pointer on the map of the matrix. The simulation successfully reproduced several of the phenomena observed in delayed matching of location experiments and provided an account of some hitherto perplexing results. As well, the model successfully predicted some new empirical data.
INTRODUCTION The recent resurgence of interest in how animals remember the spatial location of
such
things as food patches has spawned the development of several
laboratory paradigms to study spatial memory. One of these capitalizes on the fact that certain bird species cache pieces of food, which are subsequently recovered (e.g., Vander Wall, 1982). Other paradigms make use of mazes such as the radial-arm maze (e.g., Olton, 1978). the Morris water maze (e.g., Morris, 1981; Sutherland h Dyck, 1984), and the T-maze (e.g., Olson & Haki, 1983) and require subjects to remember food locations or places of safety. Another paradigm, developed in our laboratory, is a variant of delayed matching to sample. Nine pecking keys, arranged as a three by three matrix, serve as the spatial cues. Trials begin with a "ready" stimulus (a brief
0376-6357/87/$03.50
0 1987 Elsevier Science Publishers B.V.(Biomedical Divmion)
106
operation of the grain feeder or dimming of the houselight); then a randomly selected key is lit briefly as a sample. After a short delay the sample key is lit again along with a non-sample key. A peck at the sample key produces grain reinforcement,whereas a peck at the non-sample key leads only to the intertrial interval. Several studies using the delayed matching of key location procedure have been performed (e.g., Kamil & Balda, in press; Smith, Attwood, h Niedorowski, 1982; Wilkie, 1983a, 1983b, 1984, 1986, in press; Wilkie h Summers, 1982) and several basic facts about pigeons' memory for key location have been established. Consequently, it now seems appropriate to develop some hypotheses about the memory process that results in accurate matching on the delayed matching to key location task. The model that we propose--a "drift" model--was selected primarily because 1) it is conceptually simple and therefore meets the demand of parsimony, and 2) it has an integral spatial component and therefore seems appropriate as a model of spatial memory. The model shares features with one proposed by Roitblat (1984) for pigeons' memory for visual stimuli such as colors in a conventional delayed matching to sample task. The essential components of the model are as follows: 1. The pigeon is assumed to have knowledge of familiar locations--i.e., 1 the key matrix--stored in a stable map-like long-term or reference memory. 2.
The short-term or working memory process is assumed to involve an
attention focus or "pointer" that moves ("drifts") over the surface of the map of the matrix. 3.
This pointer migrates toward the representationof the sample location
during the period of time that the sample key is physically present. 4.
The pointer wanders randomly from its previous location when the
sample is no longer present (i.e., during the retention interval). 5.
During acquisition the pigeons learn the rule to choose the key
location whose representation is closer to the location of the pointer. 6.
During the intertrial interval the pointer is positioned randomly on
the key matrix representation. The model incorporates two widely-acceptednotions of animal memory processes: a) that there is a distinction between long- and short-term (or
lBecause the pigeons directly face the matrix an often neglected problem in most cognitive map theories - how animals transform frontal views of objects in their environment into bird's eye or survey map views - is absent (cf. Strelow, 1985).
107
reference and working) memory, and b) that memory is often prospective rather than retrospective (cf. Honig & Thompson, 1982). A representationof the key matrix is in a map-like reference memory. The pointer location comprises information that must be retained during the trial but discarded before the next trial. The model is prospective in the sense that information concerning the correct choice is carried forward in time rather than being stored and then retrieved at a later point in time.
COMPUTER SIMULATION OF DRIFT MODEL Because the drift process is dynamic, computer simulation seemed ideally 2 suited to evaluate the model. Although computer simulation is a technique widely used in such fields of study as human cognition (e.g., Metcalfe-Eich, 1982) and behavioral ecology (e.g., Chantrey, 1982), with rare exceptions (e.g., Daly 6 Daly, 1984; Hinson 6 Staddon, 1983; Shimp, 1984) this powerful technique has not been used by those studying animal learning and cognition. Accordingly, another rationale for the present project was to demonstrate the utility of computer simulation as a tool in the study of animal memory processes. Simulation of Drift Model An initial step in developing a computer model is to translate the model into a programming flowchart. We have done this in Fig. 1.
In this depiction
time flows from top to bottom. In a given trial the pointer is first moved to a randomly selected spot on the two-dimensionalmap-like reference memory of the key matrix. Once the intertrial interval is over and the sample key is lit the pointer begins to migrate towards the location of the sample key on the reference representation. Once the sample presentation is over, the pointer begins to move randomly from the spot it occupied at sample offset. This wandering continues until the retention interval ends and the choice stimuli appear. It is then assumed that the pigeon compares the distance between the pointer location and the sample key with the distance between the pointer location and the distractor key, and then pecks at the key closer to the pointer. The peck is followed by reward and/or the intertrial interval.
zLehman (1977), Apter (19701, and Dutton G Starbuck (1971) provides good reviews of the theory and practice of computer simulation in the behavioral sciences. An earlier paper by Gregg and Simon (1967) also provides excellent background.
108
6
INTERTRIAL INTERVAL OVER? 4 YES : SAMPLE ‘-t_ PRESENTATION OVER? YES
tNO--j 1 NO+?%i%i+
