Herbert S. Terrace - US grants

How are behavioral sequences learned and integrated? The proposed research will investigate non-verbal mechanisms underlying serially organized behavior by studying its cognitive development in monkeys. This strategy eliminates two factors that complicate the study of serially organized behavior in human subjects: language and experience with serial tasks. The investigation of serial learning animals also provides a comparative perspective of mechanisms of serially organized behavior. How behavioral sequences are learned and integrated will be studied with two recently developed techniques for training monkeys to produce and recall lists of arbitrary and numerical stimuli. In the simultaneous chaining paradigm, all of the stimuli and opportunities to respond are available simultaneously, a key feature of serial tasks used in verbal learning experiments on human subjects. Since their configuration is changed from trail to trial subjects must rely on a representation of their ordinal position while executing the required sequence. Short-term serial memory will be studied with a delayed-matching-to-successive-samples paradigm in which the subject is required to reproduce the list displayed as the sample as a simultaneous chain. Some lists will provide real-time measures of mechanisms underlying sequential behavior. Another unusual feature of the proposed research is the extensive experience of many of the subjects in performing serial tasks with arbitrary and numerical stimuli. The experiments will be conducted with touch-sensitive video monitors that allow the experimenter to select list items from thousands of digitized. The nature of representations that mediate sequence production will be studied by: (1) determining how serial expertise develops; (2) assessing a monkey's ability to "chunk" sequences it recognizes and/or produces; (3) assessing subjects' knowledge of the ordinal position of items in a list; (4) measuring the subjective organization of list items in short-term serial memory; and (5) comparing numerical and symbolic distance effects. The results of these experiments will provide non-verbal models of serial learning. The proposed research will also have two important interdisciplinary ramifications: (1) it can provide preparations for studying the neural control of serially organized behavior; and (2) the non-verbal serial tasks that will be used with monkeys could also be used with pre-verbal children in ways that would reveal the contributions that language adds to basic serial skills.

DESCRIPTION (provided by applicant): Ordinal relationships exert a pervasive influence on behavior. In the laboratory, ordinal knowledge has been studied in a variety of experiments on both humans and animals. Examples include transitive inference (TI) based on reward contingency or social dominance, knowledge of ordinal relationships between items trained by the simultaneous chaining (SC) paradigm, and extrapolations of ascending and descending rules in the case of numerical stimuli. RT and accuracy data from these studies reveal very similar distance and magnitude effects in humans and animals, implying that they use similar nonverbal strategies. The results of these experiments have opened a new line of research on serial learning that has the potential for defining non-linguistic primitives that are shared by animals and humans. The goal of the proposed research is to explore the concept of an ordinal comparator that represents and integrates ordinal information from different ordinal continua. Two basic features of the ordinal comparator are (1) it does not make any assumptions about linguistic ability and (2) it integrates existing data on ordinal knowledge. Our experiments, which will be performed on rhesus macaques (Macaca mulatta) and adult humans, will use a variety of proven methods (simultaneous chaining, transitive inference, numerical, psychophysical, and spatial discrimination tasks) for training monkeys to execute lists composed of arbitrary stimuli and by stimuli that can be characterized as inherently ordered. On an SC, arbitrary list items (usually photographs) are presented simultaneously on a touch sensitive video monitor. The spatial positions of items within an SC are randomized to minimize spatial and motor cues, except in the case of spatial tasks, in which they are intended to convey ordinal information. On an SC, subjects must learn to generate a representation of the entire sequence and their current position in that sequence, as they move from item to item. Under the TI paradigm, the subject is rewarded each time it chooses an arbitrarily defined correct item from pairs of arbitrarily selected stimuli. Subjects are then tested on their ability to infer the correct item from pairs of non-adjacent items. Numerical stimuli will be trained by conceptual and symbolic matching-to- sample paradigms. Our major hypothesis is that positive transfer will occur between two ordinal tasks, whether spatial, numerical, psychophysical, serial or transitive, to the extent that the representations that mediate each task are isomorphic. Toward that end, we propose 4 lines of comparative research on the non-verbal representation of ordinal distance and numerical magnitude. The proposed research has important interdisciplinary ramifications for studies of neuroscience (neural correlates of ordinal memory), child development (nonverbal tests for autistic children), and mental health (cognitive tests of ordinal memory during electroconvulsive therapy (ECT) and magnetic seizure therapy (MST)).The proposed research will advance understanding of the representation of order in humans and monkeys. This line of research is already being applied directly to studies of anterograde and retrograde amnesia induced by ECT and TMS. In addition we expect the results to have implications for the development of new diagnostic tools for autism spectrum disorder.

DESCRIPTION (provided by applicant): We propose a new approach to the study of inferential learning by investigating how monkeys and humans infer implicit serial relationships during training on a Transitive Inference (TI) paradigm. TI implies the ability to conclude that A C if A > B and B > C. This logic can be extended to any number of items as long as their relationships obey transitivity. TI has been shown to exist in species as diverse as pigeons, monkeys, and humans and is thought to be essential for understanding complex social relationships such as dominance hierarchies. TI is also critical for understanding ordinal relationships, which, by definition, obey transitivity. Linear spatial relationships also obey transitivity (if A is to the left of B, and B is to the left of C, then A is to the left of C). Hene, it has been proposed that TI may be related to spatial representations that inhabit a virtual workspace. The idea is that one can imagine adjacent items in an ordered list as occupying neighboring positions on an imaginary line. Thus, ordinal relationships that seem abstract may in fact be mapped onto existing spatial representations. To test this, we plan to investigate the learning and representation of ordinal relationships among novel stimuli in regions of parietal and prefrontal cortex that are believed to be involved in representing spatial information, especially relative spatial position. These are the first experiments to investigate the acquisition of implict inference at the behavioral level that is synchronized to simultaneous measurement of the activity of individual brain cells throughout TI learning (including acquisition). Ours are signifiant because they provide the first neurological investigation of implicit serial learning in a non-human primate that is not confounded by spatial or temporal cues. From a physiologist's perspective, areas LIP and SEF have been shown to encode both spatial and abstract qualities of visual stimuli. There is, however, no theoretical framework that integrates these different representations. We propose to test the idea that a virtual workspace may account for both spatial and non-spatial coding in LIP and SEF. Health Relatedness: These experiments are relevant to Schizophrenia, Autism, Alzheimer's disease, and other conditions whose patient populations have deficits in their performance of TI problems.

Project Summary One of the oldest questions in psychology and neuroscience is whether associations of stimuli and responses are sufficient to explain learning. Or, in addition, are there conditions that require knowledge of rules and representations? We propose a new approach to the study of serial learning by investigating how humans and monkeys infer ordinal knowledge implicitly during training on a Transitive Inference (TI) task. In its simplest form, TI is the ability to conclude that A > C, if A > B and B > C, but here we extend the same logic to longer series composed of 7 items. TI has been shown to exist in species as diverse as pigeons, monkeys, and humans and has been used to explain complex social relationships such as dominance hierarchies. TI is critical for understanding ordinal knowledge, which, by definition, obeys transitivity, and which is believed to give rise to an internal representation of serial order. To investigate this theory, we plan to study learning and representation of ordinal knowledge during and following TI training in monkeys and human subjects. The logic of our experiments is to show (1) how manipulations of expected value do not alter the representation of ordinal knowledge in studies on overtraining of particular pairs during TI acquisition and in studies in which there is a reversal of reward magnitude during TI training, (2) overtraining of a particular stimulus-response contingency does not impair learning, and (3) the inability of association theory to account for accurate performance on derived lists on which knowledge of associations learned on the original list are irrelevant. Our monkey experiments are the first to investigate implicit inference at the behavioral level that is synchronized to simultaneous measurement of the activity of individual neurons in prefrontal cortex (PFC) and posterior parietal cortex (PPC) throughout TI learning (including acquisition). Our experiments aim to show that 1) TI training leads to a representation of serial order of novel stimulus pairs and 2) ordinal position and symbolic distance are represented in PFC and LIP and that those representations arise de novo each time an animal learns a new list. Health Relatedness: These experiments are relevant to Schizophrenia, Autism, Alzheimer?s disease, and other conditions whose patient populations have deficits in learning and reasoning that manifest in the performance of TI problems.