THE PSYCHOLOGY OF LEARNING AND MOTIVATION Advances in Research and Theory
VOLUME 38
This Page Intentionally Left Blank
THE PSYCHOLOGY OF LEARNING AND MOTIVATION Advances in Research and Theory
EDITED BY DOUGLAS L. MEDIN DEPARTMENT OF PSYCHOLOGY NORTHWESTERN UNIVERSITY EVANSTON, ILLINOIS
Volume 38
ACADEMIC PRESS San Diego London Boston New York Sydney Tokyo Toronto
This book is printed on acid-free paper.
@
Copyright 0 1998 by ACADEMIC PRESS All Rights Reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the Publisher. The appearance of the code at the bottom of the first page of a chapter in this book indicates the Publisher’s consent that copies of the chapter may be made for personal or internal use of specific clients. This consent is given on the condition, however, that the copier pay the stated per copy fee through the Copyright Clearance Center, Inc. (222 Rosewood Drive, Danvers, Massachusetts 01923), for copying beyond that permitted by Sections 107 or 108 of the US. Copyright Law. This consent does not extend to other kinds of copying, such as copying for general distribution, for advertising or promotional purposes, for creating new collective works, or for resale. Copy fees for pre-1998 chapters are as shown on the title pages. If no fee code appears on the title page, the copy fee is the same as for current chapters. 0079-742 1/98 $25.00
Academic Press a division of Harcourt Brace & Company 525 B Street, Suite 1900, San Diego, California 92101-4495, USA http://www.apnet.com
Academic Press Limited 24-28 Oval Road, London NWI 7DX, UK
http://www.hbuk.co.uWap/ International Standard Book Number: 0- 12-543338-7 PRINTED IN THE UNITEDSTATESOF AMERICA 98 99 0 0 0 1 02 0 3 B B 9 8 7 6
5
4
3 2 1
CONTENTS
Contributors
................................................................................
ix
TRANSFER-INAPPROPRIATE PROCESSING NEGATIVE PRIMING AND RELATED PHENOMENA
W. Trammel1 Neil1 and Katherine M . Mathis I . Introduction
.........................................................................
I1. Transfer-Appropriate Processing ..............................................
111. IV . V. VI . VII .
Memory: Abstraction or Instances? .......................................... Negative Priming ................................................................... Related Phenomena ............................................................... Toward a Tip-Tap Theory of Priming ....................................... Summary ............................................................................. References ...........................................................................
1 3 8 10 21 33 36 37
CUE COMPETITION IN THE ABSENCE OF COMPOUND TRAINING ITS RELATION TO PARADIGMS OF INTERFERENCE BETWEEN OUTCOMES
Helena Matute and Oskar Pinefio I . Introduction ......................................................................... I1. Cue Competition in the Absence of Compound Training: The Basic Effect .......................................................................... 111. The Unlearning Explanation ................................................... IV . Theories of Interference between Outcomes .............................. V. Competition between Elementally Trained Cues in Judgmental Tasks .................................................................................. VI . Extending Models of Competition between Outcomes to Account for Competition between Cues ................................................ V
45 47
55 59 66
73
vi
Contents
VII . Conclusions .......................................................................... Acknowledgments ................................................................. References ...........................................................................
76 77 77
SOONER OR LATER THE PSYCHOLOGY OF INTERTEMPORAL CHOICE
Gretchen B . Chapman I . Introduction ......................................................................... I1. Violations of Normative Theory ...............................................
111. Domain Effects
.....................................................................
IV . Agreement among Measures of Time Preferences ....................... V . Conclusions ..........................................................................
Acknowledgment .................................................................. References ...........................................................................
STRATEGY ADA-
83 85 94 101 108 111 111
AND INDIVIDUAL DIFFERENCES
Christian D . Schunn and Lynne M . Reder I. Conceptions of Individual Differences .......................................
115 117
Environment ........................................................................ IV . Individual Differences in a Complex. Dynamic Task .................... V. General Discussion ................................................................ References ...........................................................................
131 139 147 150
I1. Strategy Adaptivity in Arithmetic ............................................. I11. Investigating Individual Differences in a More Controlled
GOING WILD IN THE LABORATORY LEARNING ABOUT SPECIES TYPICAL CUES
Michael Domjan I. Introduction ......................................................................... I1. Use of a Conspecific as a Conditioned Stimulus .......................... I11. Do Species Typical Cues Have Special Properties as Conditioned Stimuli? ............................................................................... IV . Potentiation of Responding to Species Typical Cues .................... V . Conclusion ........................................................................... Acknowledgments ................................................................. References ...........................................................................
155 158 170 175 181 183 183
Contents
vii
EMOTIONAL MEMORY: THE EFFECTS OF STRESS ON “COOL” AND “HOT” MEMORY SYSTEMS
Janet Metcalfe and W. Jake Jacobs I . Hot and Cool Memory Systems ............................................... I1. Stress .................................................................................. 111. Adaptive Significance of the Two Systems ................................. IV . Conclusion ........................................................................... Acknowledgments ................................................................. References ...........................................................................
188 201 213 214 215 215
METACOMPREHENSION OF TEXT: INFLUENCE OF ABSOLUTE CONFIDENCE LEVEL ON BIAS AND ACCURACY
Ruth H . Maki I. Introduction ......................................................................... I1. Effect of Expectancy and Processing on Metacomprehension Measures .............................................................................. 111. Recommendations for Metacomprehension Measures ................... References ...........................................................................
223 228 244 246
LINKING OBJECT CATEGORIZATION AND NAMING EARLY EXPECTATIONS AND THE SHAPING ROLE OF LANGUAGE
Sandra R. Waxman I. Introduction ......................................................................... I1. The Developmental Process: A Dynamic Interplay between the Constraints in the Learner and the Amount and Information Present in the Environment ..................................................... 111. A Vignette ........................................................................... IV . The Relation between Object Categorization and Naming: Background Issues ................................................................. V . Developmental and Crosslinguistic Considerations ...................... VI . Empirical Evidence ................................................................ VII . Integrating the Developmental and Crosslinguistic Evidence ......... Conclusion ........................................................................... References ...........................................................................
249
Index .......................................................................................... Contents of Recent Volumes ............................................................
293 303
250 251 252 255 259 271 283 283
This Page Intentionally Left Blank
CONTRIBUTORS Numbers in parentheses indicate the pages on which the authors’ contributions begin.
Gretchen B. Chapman (83), Psychology Department, Rutgers University, Piscataway, New Jersey 08854 Michael Domjan (155), Department of Psychology, University of Texas at Austin, Austin, Texas 78712 W. Jake Jacobs (187), Department of Psychology, University of Arizona, Tucson, Arizona 85721 Ruth H. Maki (223), Department of Psychology, Texas Tech University, Lubbock, Texas 79409-2051 Katherine M. Mathis (l), Department of Psychology, University at Albany, State University of New York, Albany, New York 12222 Helena Matute ( 4 9 , Departamento de Psicologia, Universidad de Deusto, Apartado 1, 48080 Bilbao, Spain Janet Metcalfe (187), Department of Psychology, Columbia University, New York, New York 10027 W. Trammel1 Neil1 (l), Department of Psychology, University at Albany, State University of New York, Albany, New York 12222 Oskar Pineiio (45), Departamento de Psicologia, Universidad de Deusto, Apartado 1, 48080 Bilbao, Spain Lynne M. Reder (115), Department of Psychology, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213 Christian D. Schunn (115), Department of Psychology, Carnegie Mellon University, Pittsburgh, Pennsylvania 15213 Sandra R. Waxman (249), Department of Psychology, Northwestern University, Evanston, Illinois 60208
This Page Intentionally Left Blank
TRANSFER-INAPPROPRIATE PROCESSING Negative Priming and Related Phenomena W. Trammel1 Neil1 Katherine M. Mathis
I. Introduction
Learning, memory, and cognition are fundamentally dependent on a sensitivity to repetition. Without the ability to detect that a present event is the same as, or similar to, previously experienced events, it would be impossible to learn the regularities of one’s environment or to form coherent conceptual categories in which instances are related by similarity. It is hardly surprising that much of the research on learning, memory, and cognition, both historical and contemporary, has been concerned with effects of repeated stimuli on performance. Performance often benefits from repetition. Most obviously, recall or recognition of a studied item usually improves with repeated study of the same item (e.g., Ebbinghaus, 1885). In reaction-time tasks, a subject responds much faster to a stimulus (e.g., naming it) if he or she has just responded to an identical stimulus (the stimulus-repetition effect) (Bertelson, 1963; Hyman, 1953; Keele, 1969). Beller (1971) introduced the term priming to describe the facilitatory effect of a preexposed stimulus (without an overt response) on responding to a subsequent similar stimulus (see also Neely, 1977; Posner & Snyder, 1975). Repetition priming refers to the facilitatory effect on responding to an identical stimulus. However, facilitatory effects also often occur for stimuli that share only certain THE PSYCHOLOGY OF LEARNING AND MOTIVATION, VOL. 3X
1
Copyright 0 1998 by Academic Press. All rights of reproduction in any form reserved. 0079-7421/YX$25.00
2
W. Trammel1 Neill and Katherine M. Mathis
attributes, for example, perceptual, orthographic, phonological, morphological, syntactic, or semantic similarity. For this reason, the presence or absence of priming for a specific level of similarity is frequently used to index what kind of information about a stimulus has been encoded. If the task does not explicitly require recollection of the prior exposure, stimulusrepetition and priming effects demonstrate implicit memory for the exposure (Richardson-Klavehn & Bjork, 1988; Schacter, 1987). Contrary to the common wisdom, there are also many circumstances in which an identical or similar repetition may hamper performance. The most prominent historical example is the interference that occurs when people learn similar lists of items for a subsequent recall test. Such interference effects were most thoroughly investigated in paired associate learning, in which subjects are required to learn pairings of items (e.g., SHOEPENCIL) and are then tested for recall of the item that was paired with a cue (e.g., SHOE-?). In the “A-B, A-D” condition, subjects learn two lists of paired items in which the cue items are the same, but the target items are changed (e.g., SHOE-PENCIL in List 1, but SHOE-BANANA in List 2). In a control condition (“A-B, C-D”), subjects learn two lists in which the cues, as well as the targets, are different. Subjects who learned both lists with the same cue items recall the List 1 targets more poorly than subjects in the control condition (McGeoch, 1932; Melton & Irwin, 1940;Underwood, 1945). In other words, more recent learning can interfere with recall of earlier material (retroactive interference). Furthermore, the subjects who learned both lists are also less likely to recall the List 2 items (Melton & von Lackum, 1941; Underwood, 1945, 1948). Hence, earlier learning can also interfere with recall of more recent material (proactive interference). Or, stated more generally, repetition of identical stimuli can hurt performance if the repetitions require association with different responses. In the classic studies of proactive interference, a repeated stimulus is studied in two different contexts and then is tested for recall or recognition at a later point in time. Hence, it is ambiguous whether the first occurrence interferes with the “on-line” processing of the second occurrence or whether the interference reflects processes required for retrieval at the later time. The present chapter is concerned with several phenomena in which an initial occurrence interferes with the immediate response to the second occurrence of an identical or similar stimulus. The prototypical case, negative priming, occurs when a stimulus is first deliberately ignored on a prime trial and then reappears as a target requiring a response on a probe trial (Neill, 1977b;Tipper, 1985). Research on the causes of negative priming suggests a general theoretical framework (transfer-inappropriateprocessing) that may in turn apply to other phenomena in which repetition ham-
Transfer-InappropriateProcessing
3
pers performance, including “repetition blindness” (Kanwisher, 1987), the “before-disruption effect” (Smith, Haviland, Reder, Brownell, & Adams, 1976), and “inhibition of return” (Posner & Cohen, 1980, 1984). 11. Transfer-Appropriate Processing
The phrase transfer-appropriateprocessing was introduced into contemporary cognitive psychology by Morris, Bransford, and Franks (1977) as an alternative to the “levels of processing” framework proposed by Craik and Lockhart (1972). According to the levels-of-processing hypothesis, as originally stated, retention improves with the degree of abstraction (or “depth”) to which a stimulus is processed. Hence, if a subject’s attention is oriented toward the meaning of a word, subsequent recall or recognition memory is likely to be better than if the subject’s attention is oriented toward more superficial attributes like the number of letters or presence or absence of a specific letter (e.g., Hyde & Jenkins, 1969, 1973). Morris et al. (1977) argued that memory performance cannot be attributed entirely to how the studied items are encoded in memory; equally important is whether the encoded information is appropriate to the type of test used to measure retention. Standard recall and recognition tests presumably depend most on semantic information (for whatever reason): therefore, they benefit from processing of meaning. However, other tests of retention may well depend on other information, and thereby benefit most from orienting tasks at study that emphasize attributes other than meaning. Accordingly, Morris et al. demonstrated that orienting subjects toward phonological information (rhymes) produced superior recognition memory for phonologically similar words. In a series of experiments, subjects judged whether target words were consistent with a sentence frame. Some frames required judgment of meaning, for example, “The -has feathers: EAGLE.” Other frames required judgment of phonology, for example, “rhymes with legal: EAGLE.” Subjects were afterward given an unexpected recognition memory test for either the studied words (e.g., EAGLE), or for words that rhymed with studied words (e.g., REGAL). Recognition of the studied words was better if subjects had initially been oriented toward the words’ meanings, as would be predicted by the levels-of-processing hypothesis. However, recognition of rhyming words was better if subjects had initially judged other rhymes. In the absence of further elaboration, “transfer-appropriate processing” risks circularity: If study condition A results in better performance on a given test than study condition B, then the processing under condition A is by definition more “transfer-appropriate.’’ It is implicit in the transfer-
4
W.Trammel1 Neill and Katherine M.Mathis
appropriate processing approach that different kinds of memory tests depend on different kinds of information and that different study conditions may or may not effectively encode the subsequently required information. However, this is really just what we mean by “memory”: one does not prepare for a Geography exam by studying one’s Biology notes! The difficulty of independently defining transfer appropriateness is illustrated by an experiment by Neill (1977a): Subjects were shown 18 animal names and 18 inanimate-object names, one at a time and in random order. A random half of each set was presented in uppercase, and the other half in lowercase. Each word was preceded by a randomly selected question word, ANIMAL?, OBJECT?, UPPER?, or LOWER? Subjects first pronounced the word, and then pressed a key for “yes” or “no,” corresponding to whether the word agreed with the question. The exposure times, determined by the subjects’ responses, were slightly longer for case-oriented words (M = 2.19 s) than for meaning-oriented words (2.14 s). Subjects then received an unexpected recognition test in which each word was presented in its original case and its opposite case. Case recognition was above chance for both meaning-oriented words (p = .65) and case-oriented words ( p = .62). Although the correct recognition proportions did not differ significantly between the two study conditions, correct recognition latencies were significantlyfaster to semantically oriented words (M = 3.40 s) than to case-oriented words (3.97 s). In other words, attention to meaning appears to be more “transfer-appropriate” to a case-recognition test than attention to case! The results of Neill (1977a) are deceptive, though, because the memory test did not really measure recognition of letter case per se. Rather, the test required recognizing the conjunction of letter case with specific words. If the case-orienting task had successfully encoded only case information (“upper” or “lower”) and no word-specific information, there would be no way to reconstruct at the time of test which words appeared in which case. Both orienting tasks must have allowed the encoding of some identity information as well as some information about letter case; presumably, the semantic orienting task provided better encoding of the identity information. But, unless one has a more detailed theory of what the informational requirements are of a given task, and what processes underlie the encoding and retrieval of that information, one cannot predict apriori what conditions will be more “transfer-appropriate.” By itself, the construct of transfer appropriateness has no explanatory power. Transfer-appropriate processing has been invoked in more recent years to explain various implicit memory phenomena. As noted earlier, implicit memory is demonstrated when a prior event influences performance in a task that does not explicitly require recollection of that event. For example,
Transfer-Inappropriate Processing
5
in a word-fragment completion task, subjects are shown some of the letters and are asked to fill in the letters that of a word (e.g., -SS-SS--) complete the word. If a subject read the target word earlier in the experiment, he or she is much more likely to complete the fragment. Remarkably, this facilitation occurs even if the subject does not recognize the word as having been encountered in the experiment, and the effect may persist for months between the initial exposure and the test (Sloman, Hayman, Ohta, Law, & Tulving, 1988; Tulving, Schacter, & Stark, 1982). Interest in implicit memory was heightened by findings that suggested a dissociation of variables that influence implicit memory versus explicit memory. Whereas explicit memory (standard recall and recognition memory tests) is typically influenced by “depth” of processing, implicit memory tests often are not (Graf & Mandler, 1984; Graf, Mandler, & Haden, 1982; Jacoby & Dallas, 1981). Indeed, whereas recognition memory is relatively insensitive to changes in surface detail like letter case, letter font, or even sensory modality, implicit memory may be very sensitive to such changes (Blaxton, 1989; Jacoby & Dallas, 1981; Jacoby & Hayman, 1987; Kolers & Ostry, 1974; Roediger & Blaxton, 1987a). Generating a target at time of study typically produces better recall or recognition than merely reading it (Slamecka & Graf, 1978); however, this “generation effect” is often reversed in implicit memory measures (Blaxton, 1989; Jacoby, 1983b; Winnick & Daniel, 1970). Roediger and Blaxton (1987b; Blaxton, 1989; Roediger, Weldon, & Challis, 1989) argued that such dissociations are best understood from a transfer-appropriate processing perspective. Certain implicit memory tasks, like word-fragment completion, appear to depend more on perceptual information (hence, sensitivity to modality and surface detail). Facilitation will result as long as the surface form has been processed; “deeper” processing may not encode any further information that is appropriate to the task. In a generation task, the subject is deprived of the perceptual processing that would benefit fragment completion, and so priming is less than if the subject passively reads the word. In order to assert something more than “what is remembered depends on what is learned,” Roediger and Blaxton adopted the distinction between data-driven and conceptually driven processes (cf. Jacoby, 1983b). Conceptually driven processes are subject initiated-for example, rehearsal, elaboration, organization, reconstruction. In contrast, data-driven processes are initiated by the perceptual encounter with the stimulus. Explicit memory (recall and recognition) is primarily conceptually driven, and thereby benefits from conceptually driven processing at time of study; some implicit memory tests, such as fragment completion, require more data-driven processing, and thereby benefit more from data-driven processing at time of study.
6
W. Trammell Neill and Katherine M. Mathis
In order for the distinction between data-driven versus conceptually driven processes to be useful, it must be shown that attributes of the two types of processes have qualitatively different consequences for different memory tests. For example, if data-driven tasks are not subject initiated, then data-driven implicit memory tests should not benefit from intention to learn. Greene (1986) compared the effects of intentional versus incidental learning on an implicit memory test, word-stem completion, and an explicit memory test, cued recall. At study, subjects read words aloud, ostensibly as part of a distractor task between presentation and recall of lists of digits. Some subjects were forewarned that they would receive a memory test afterwards. Greene found that the forewarned subjects had better cued recall than the naive subjects, but the two groups showed the same degree of priming in the word-stem completion test. Neill, Beck, Bottalico, and Molloy (1990, Experiment 1) found similar results comparing wordfragment completion and recognition memory. The results of Greene (1986) and Neill et al. (1990, Experiment 1) might seem to confirm a qualitative distinction between data-driven implicit memory and conceptually driven explicit memory. However, the subjects who were forewarned of a memory test were not informed as to what kind of memory test they would receive. Neill et al. (1990, Experiment 2) gave some subjects practice on word-fragment completion and warned them that the words on a list to be named aloud would be re-presented for a similar fragment-completion test. Other subjects received a practice recognition memory test and were warned that the study list would be similarly tested. A control group (incidental learning) did not expect a memory test. Half of each group was then tested on word-fragment completion, and the other half on recognition memory. Although subjects who expected a recognition test showed no enhancement of priming on the fragment-completion test relative to the control group, the subjects who expected the fragmentcompletion test performed significantly better on the studied words (Fig. 1).Contrary to the characterization of word-fragment completion as “datadriven,” subjects were able to voluntarily enhance the retrievability of information that would be useful for that test-they just had to know what information would be useful. Again, transfer-appropriate processing appears to be nothing more than “memory” as we generally understand it. If the concept of transfer-appropriate processing has theoretical utility, it is probably to remind us that certain phenomena do not require theories. In particular, processing dissociations do not imply either qualititatively different processes (a la Blaxton, 1989; Roediger & Blaxton, 1987a, 1987b; Roediger et al., 1989) or qualitatively different storage systems ( a la Schacter, 1990;Schacter, Cooper, & Delaney, 1990).Studying one’s Geogra-
Transfer-InappropriateProcessing Expected Test Recognition
0
Fragment
None
I s
3s Exposure Duration
6.5 s
1s
3s Exposure Duration
6.5 s
Fig. 1. Proportions of correct recognitions (top) and word-fragment completions (bottom) as a function of exposure duration and test expectation (Neill, Beck, Bottalico, & Molloy, 1990).
phy notes results in greater “priming” on a Geography exam than on a Biology exam, whereas studying one’s Biology notes results in greater enhancement of performance on the Biology exam. The “dissociation” of effects of the two study conditions does not imply that we have qualitatively
8
W. Trammel1 NeiU and Katherine M. Mathis
different geography-encoding and biology-encoding processes or that we have a Geography Representational System that differs qualitatively from a Biology Representational System. 111. Memory: Abstraction or Instances?
The ancient Greek philosopher Plato (428-348 BC) apparently believed that memories are faint copies of perceptions. To think about “dogs,” for example, one would bring to mind specific instances of dogs encountered in the past. In contrast, Aristotle (384-322 BC) proposed that generalized concepts can be abstracted from multiple perceptual experiences. So, to think of “dogs,” one can bring to mind an image of a dog that retains the common properties of dogs but is not identical to any specific experience (Leahey, 1987; for detailed discussion, see Wedin, 1988). The debate over whether mental representations are instances a la Plato or abstractions a la Aristotle has been a recurrent one in cognitive psychology. For example, why do repetitions of a studied word increase the likelihood that the word will be recalled? One view is that a unitary representation of that word increases in “strength” with repetition (Wickelgren, 1970, 1974; Wickelgren & Norman, 1966). Alternatively, perhaps each repetition results in another instance in memory; the more instances of a given word in memory, the greater the probability that at least one instance will be retrieved (Hintzman, 1976; Hintzman & Block, 1971). Posner and Keele (1968,1970) taught subjects to categorize random-dot patterns that were distortions of several prototype patterns. After training on the distortions, subjects were able to categorize the prototype patterns as accurately as the instances on which they were actually trained. Furthermore, ability to classify the instances declined significantly over a week’s delay, whereas ability to classify the prototypes did not. Posner and Keele concluded that subjects learned an abstract “schema” for each category, which represented the averages of the instances on which they had been trained. Insofar as the instances were random distortions of the prototypes, their averages would most closely resemble the prototypes. Hintzman (1986), however, demonstrated that a memory retrieval model based entirely on instances could account for the results of Posner and Keele perfectly well. According to the model, a pattern to be categorized is compared in parallel to all of the instances in memory. Because the prototype is closer, on average, to all of the instances, it is easier to classify than new distortions. Further, as some instances are forgotten, the prototype remains close to the average of the remaining instances, whereas individual studied instances show forgetting.
Transfer-Inappropriate Processing
9
The abstraction versus instances debate has most recently resurfaced in theories of repetition priming. According to abstractionist views, such as Morton’s (1969, 1979) logogen theory, the initial exposure to an object causes a preexisting mental representation of the object to become more highly activated or more easily retrievable. Similarly, semantic network theories (e.g., Anderson, 1976;Anderson & Bower, 1973; Collins & Loftus, 1975) assume that abstract “nodes” representing concepts become activated and that such activation also “spreads” to associated nodes representing semantically related concepts. In contrast, according to episodic or instance views, exposure to a stimulus results in a highly specific memory trace of that exposure. The subsequent appearance of a similar stimulus prompts the retrieval the related memory traces, which may in turn speed its classification or the selection of a relevant response (e.g., Jacoby, 1983b; Logan, 1990). Logan (1988) proposed that tasks become more “automatic” through the accumulation of instances. According to this theory, response selection can be accomplished through either of two mechanisms: (a) a relatively slow, limited-capacity computational process or “algorithm”; (b) retrieval of past instances in which stimuli that are similar to the current stimulus were processed. This retrieval is presumed to be relatively effortless, but probabilistic, and occurs in parallel with algorithmic processing. Retrieved instances provide information about what response was made to similar stimuli in the past. Therefore, if a relevant instance can be retrieved in less time than that required for algorithmic computation, reaction time will be speeded. (This mechanism is quite similar to that proposed by Bertelson, 1963, and Keele, 1969, to account for the stimulus repetition effect.) As more instances are accumulated in memory, the probability increases that retrieval will succeed sooner than the algorithmic computation. Logan (1988) demonstrated that such a probabilistic mechanism naturally accounts for the ubiquitous “power function’’ that describes the reduction in response latencies over extended task practice. Empirically, the abstractionist and episodic views are distinguished by whether priming depends on specific details that should be irrelevant to an abstract representation-for example, the effect of typeface in word recognition or the context in which a target stimulus is imbedded. Tenpenny (1995) examined the literature on repetition priming in word-naming and lexical-decision (wordhonword judgment) tasks and concluded that repetition priming is often dependent on surface similarity and contextual similarity, contrary to the abstractionist view. In addition, priming effects in some tasks have been found to persist over days (Jacoby, 1983a;Jacoby & Dallas, 1981; Scarborough, Cortese, & Scarborough, 1977), months (Kolers & Ostry, 1974; Sloman et al., 1988), and even a year (Kolers, 1976). On the
10
W.Trammel1 Neill and Katherine M. Mathis
assumption that “activation” of representations should be rather labile, one might expect that activation should dissipate over a much shorter interval of time. That priming can be caused by discrete episodes or instances does not, however, preclude the possibility that activation of abstract representations also can contribute to priming. Other results have, in fact, been interpreted as supporting the involvement of abstract representations in priming. (See Tenpenny, 1995, for review.) As will be discussed in the next section, research on negative priming suggests a contribution of abstract (semantic or lexical) activation in (positive) repetition priming, as well as a contribution of episodic retrieval.
IV. Negative Priming Tipper (1985) introduced the phrase “negative priming” to contrast the inhibitory effects of deliberately ignored stimuli with the typical facilitatory effects (“positive” priming) of directly attended or passively experienced stimuli. Negative priming was first reported by Dalrymple-Alford and Budayr (1966) in a version of the Stroop (1935) color-word task. In this task, subjects are required to name the ink colors in which a list of words are written. The well-known “Stroop effect” is that color naming is considerably slowed if the words denote different colors (e.g., GREEN written in red ink). Dalrymple-Alford and Budayr constructed lists in which each distractor word named the ink color of the next word (e.g., GREEN in red ink, followed by YELLOW in green ink). They found that the total colornaming time for such lists was even longer than that for lists of successively unrelated color-words. Dalrymple-Alford and Budayr concluded that naming the color of each word required suppressing the response to the accompanying distractor word. If the suppressed response was required as the correct response for the next color, naming time would be slowed by the time required to overcome the suppression. Neill (1977b) replicated this “distractor suppression effect” in a randomized, discrete-trials version of the Stroop task. Thus, color naming on a given trial n was slower if the target color matched the distractor word on trial n - 1 than if the target color was unrelated to the previous trial (see also Lowe, 1979). Subsequent research has yielded similar results across a wide variety of materials, including words, pictures, letters, digits, and nonsense shapes, and various tasks, including naming, categorization, matching, counting, and localization (for reviews, see Fox, 1995; May, Kane, & Hasher, 1995; Neill & Valdes, 1996; Neill, Valdes, & Terry, 1995).
Transfer-InappropriateProcessing
11
In many experiments on negative priming, the effect is demonstrated for targets that are physically identical to the preceding ignored distractor or differ from it only on an attribute used to distinguish targets from distractors (e.g., ignoring a picture drawn in green ink, and then naming the same picture drawn in red ink). It is critically important, however, that negative priming does not depend on physical identity between the ignored prime and the target. Indeed, in the Stroop task, it is an ignored word that hampers naming a target color. Tipper (1985) demonstrated that negative priming in a picture-naming task generalizes to semantic associates of an ignored picture. For example, the time to name a picture of a dog (drawn in red ink) is slowed if the subject has just ignored a picture of a cat (drawn in green ink). Yee (1991) found negative priming between semantically related words, and Tipper and Driver (1988) found negative priming between words and pictures. It is also important to note that negative priming is not merely inhibition of a response. Neill, Lissner, and Beck (1990) required subjects to match the second and fourth letters of five-letter strings (e.g., BABAB) as “same” or “different,” ignoring the flanking letters. Negative priming occurred for targets that matched the previous ignored distractors (e.g., CBCBC after BABAB) regardless of whether the prime and probe trials required similar or opposite responses. DeSchepper and Treisman (1996; Treisman & DeSchepper, 1996) also found negative priming in a task requiring matching of nonsense shapes, even though the same responses were made for related and unrelated prime and probe trials. Similarly, Yee (1991) found negative priming in a lexical-decision task, for which word targets would require the same response regardless of whether they were related or unrelated to an ignored word. In other words, it is the relation between the stimuli, not the relation between responses, that determines negative priming. Neill and Terry (1995) found negative priming to affect accuracy of tachistoscopic recognition, with unlimited time to respond, further discounting the role of response processes in negative priming. What, then, causes negative priming? Initial theorizing assumed that some aspect of a central cognitive (i.e., abstract) representation of the ignored stimulus is inhibited. It is tempting to think of inhibition as merely the opposite of activation, that is, deactivation (as naively assumed by Neill, 1979). However, this characterization of inhibition can be quickly dismissed: Numerous studies have found that negative priming depends on the presence of a distracting stimulus on the probe trial, and some of these studies found that negative priming actually reversed to positive priming if no distractor accompanied the target stimulus (Lowe, 1979;Milliken, Joordens, Merikle, & Seiffert, in press; Neill, Kahan, & VerWys, 1996; Tipper & Cranston, 1985). If a cognitive representation of the ignored color-word
12
W.Trammell Neill and Katherine M. Mathis
were simply deactivated, one would expect negative priming of related targets regardless of whether they were accompanied by distractors. Lowe (1979) found that that color naming of letter-string probes (e.g., OMTTV) showed negative priming from an ignored color-word prime only if such prime-probe sequences were intermixed with sequences in which the probe target was a color-word. When sequences with letter-string probes were intermixed with sequences in which the probe consisted of four colored dots, the letter-string probes showed positive priming from the distractor word. In addition, the colored-dot probes showed positive priming, even when such sequences were intermixed with sequences with color-word probes. Thus, negative priming of nonconflict probes appears to depend on whether the stimulus is sufficiently wordlike that the subject anticipates interference. To quote Moore (1994), “where has all the inhibition gone?” Furthermore, the fact that negative priming can reverse to positive priming implies that some source of facilitation persists when the probe conditions for negative priming are removed. Tipper and Cranston (1985) suggested that the cognitive representation of the ignored distractor remains activated, but is actively “blocked” from accessing response mechanisms (see also Neill, 1989; Neill & Westberry, 1987). The blocking is presumed to depend on the subject’s maintenance of a “selection set.” If the subject knows that selection will not be required on the target trial or readily perceives the target as nonconflict, the subject relinquishes the selection set and no longer blocks the previously activated distractor representation from further processing. Positive priming for a related target then emerges from the persisting activation of its cognitive representation. Neill and Valdes (1992; Neill, Valdes, Terry, & Gorfein, 1992) proposed that negative priming is caused by episodic retrieval rather than blocking of an abstract representation. According to this theory, adapted from Logan’s (1988) instance theory of automaticity, a target stimulus cues the retrieval of past processing episodes involving similar stimuli. If the retrieved episode includes information that a similar stimulus was ignored, processing of the current stimulus is slowed. This interference effect might occur because the episode in which the stimulus was ignored interferes with retrieval of other episodes that would contain appropriate response information, which would then force more reliance on slower, algorithmic computation. Alternatively, the retrieved “ignore this” information might directly affect the algorithmic computation. It is assumed that the most recent episode involving a stimulus similar to the target is the one most likely to be retrieved. (Logan, 1988, did not address effects of recency.) In many negative priming experiments, a relatively small set of stimuli are used repeatedly as both targets and dis-
Transfer-Inappropriate Processing
13
tractors (i.e., a “varied mapping” procedure). If remote episodes were as likely to be retrieved as more recent ones, positive priming from episodes in which the stimulus had been attended would presumably negate the effects of episodes in which the stimulus had been ignored. In this regard, Malley and Strayer (1995) failed to find negative priming for words presented only once as prime and target, although words used repeatedly in the experiment did yield negative priming. Although Malley and Strayer argue this result in support of an abstract-inhibition theory over the episodic retrieval theory, its theoretical implications remain unclear: It is possible that more familiar stimuli are encoded more effectively in the processing episode, or that a familiar target serves as a more effective retrieval cue for past episodes. It is also difficult to reconcile the Malley and Strayer result with other studies that have found negative priming for targets that had been presented only once before, as the ignored prime (DeSchepper & Treisman, 1996; Simpson & Kellas, 1989; Treisman & DeSchepper, 1996; Yee, 1991). The episodic retrieval theory of negative priming also assumes that the representation of an ignored stimulus in the processing episode may include perceptual, lexical, or semantic information, depending on the demands of the task (cf. Neill et al., 1995). As noted earlier, negative priming in the Stroop (1935) color-word task occurs between an ignored word and a target color, and several studies have found generalization of negative priming to semantic associates of ignored stimuli. Therefore, negative priming cannot be limited to strictly perceptual representations of ignored stimuli. Episodic retrieval offers two possible explanations for the dependence of negative priming on a distractor accompanying the target stimulus. First, following Logan (1988),it is assumed that task performance can be mediated by either episodic retrieval or algorithmic computation. Because the two processes are assumed to occur in parallel, a retrieved instance affects performance only if it is retrieved prior to completion of the algorithmic computation. Hence, any manipulation that slows the algorithmic computation (such as presence of a distractor) increases the opportunity for priming effects. Second, episodic retrieval is presumed to be influenced by similarity of context between “study” and “test” conditions (Tulving, 1972, 1983). The absence of a distractor on the probe trial may create a sufficiently different context from the prime trial that the probability of retrieving prime-trial information is diminished. Neill (1997, Experiment 1) tested the contextual-similarity hypothesis using a variation of the “flanker” task devised by B. A. Eriksen and C. W. Eriksen (1974). Subjects were required to press a key corresponding to the identity of the middle letter of a triplet (e.g., ABA), ignoring the flanking letters. Negative priming occurs if the prime-trial flanker re-
14
W. Trammell Neil1 and Katherine M. Mathis
appears as the probe-trial target. Because the negative priming paradigm inherently requires a distractor on the prime trial, it is not possible to test the contextual-similarity hypothesis by omitting the distractor. Instead, the onset of the distractor relative to the target was manipulated independently on the prime and probe trials: On some trials, the distractor onset was early (simultaneous with the target); on other trials, it was late (delayed by 400 ms). As shown in Fig. 2, significant negative priming occurred only if the prime-trial and prime-trial distractors were both early or both late. In other words, the presence or absence of a distractor during the first 400 ms establishes a context that determines whether prime-trial information will be retrieved on the probe trial. The results of Neill (1997, Experiment 1) are extremely damaging to the notion that the distractor representation is inhibited during the prime trial, and that negative priming on the probe trial is due to persistence of this inhibition. It is conceivable that the degree of inhibition might be affected by the prime-distractor onset and that its effect on the probe trial might also depend on the onset of the probe-trial distractor; however, there is no ready explanation for why such effects should be manifested only when the prime and probe conditions are similar. In a second experiment, Neill (1997) investigated the effect of distractor onsets on repetition priming, that is, facilitation when the prime and probe targets were identical. As also shown in Fig. 2, priming again depended on similarity of onsets. Hence, repetition priming, like negative priming, must be due at least in part to episodic retrieval. However, unlike the experiment on negative priming, there was significant residual positive priming even when the prime and probe distractor onsets mismatched. This could be attributable to the greater overall magnitude of repetition priming. Within the episodic retrieval theory, this could be explained by more effective encoding of the target than of the distractor in the prime-trial episode. However, it could also reflect a second source of priming, that is, activation of an abstract representation. Other experiments also suggest a dependence of negative priming on contextual similarity, providing further support for episodic retrieval. Neill and Valdes (1996) reported an experiment in which target and distractor letters were presented at varying spatial separations but equidistant from fixation. As expected, distractor letters caused more interference when they were near the target than when they were far (cf. C. W. Eriksen & Hoffman, 1973). Significant negative priming occurred if the prime- and probe-trial distractors were both near. But, as shown in Fig. 3, positive priming was obtained if either the prime-trial or the probe-trial distractor alone was far, and priming reverted to the negative direction when both distractors were far.
Transfer-InappropriateProcessing Probe Distractor Onset Early
Late
.......................
.......................
................
Early Late Prime Distractor Onset
Early Late Prime Distractor Onset
Fig. 2. Negative priming (top) and positive repetition priming (bottom) as a function of distractor onset on prime and probe trials (Neill, 1997).
There has been considerable disagreement over the duration of negative priming, with some studies finding measurable decline over a few seconds (Neill & Valdes, 1992;Neill, Valdes, Terry, & Gorfein, 1992; Neill & Westberry, 1987), some studies failing to find a significant decline over similar intervals (Hasher, Stoltzfus, Zacks, & Rypma, 1991; Hasher, Zacks, Stoltz-
W. Trammel1 Neil1 and Katherine M. Mathis
16
Probe Distractor Proximity Near
Y
40
..................................
30
..................................
.-p 2 0 - . . . . .
..................................
.e E 10
-
................
a
.-f
Far
--..
0-
g-10 - ........... P) z -20 _. ...........
-30
'
I
....
......... .........
...........
.........
...........
....
I
Near Far Prime Distractor Proximity
Fig. 3. Negative priming as a function of target-distractor proximity on prime and probe trials (Neill & Valdes, 1996).
fus, Kane, & Connelly, 1996;Neill & Kahan, 1997;Stoltzfus, Hasher, Zacks, Ulivi, & Goldstein, 1993; Tipper, Weaver, Cameron, Brehaut, & Bastedo, 1991), and some finding very long-term effects showing measurable decline only after one-week and one-month delays (DeSchepper & Treisman, 1996; Treisman & DeSchepper, 1996). The factors that influence the persistence of negative priming remain to be determined. However, the question of whether negative priming is a short-term or long-term phenomenon is orthogonal to the question of whether negative priming is caused by inhibition of a cognitive representation or to retrieval of a processing episode. To whit, Neill and Westberry (1987) interpreted their results as indicating that the cognitive representation recovers rapidly from inhibition, whereas Tipper et al. (1991) and Hasher et al. (1991; Stoltzfus et al., 1993) interpreted their results as indicating a relatively permanent inhibition of the cognitive representation until the stimulus reappeared as a target. Neill et al. (1992) interpreted their results as indicating a strong recency effect on episodic retrieval, whereas DeSchepper and Treisman (1996; Treisman & DeSchepper, 1996) initerpreted their results as evidence for long-term persistence of episodic traces. Whereas effects of the interval between prime and probe may not be diagnostic for the causes of negative priming, the effect of the interval prior to the prime may be. Much research suggests that the likelihood of retrieving an episodic memory depends on its temporal discriminability (Baddeley,
Transfer-InappropriateProcessing
17
1976; Bjork & Whitten, 1974; Glenberg, 1987; Glenberg & Swanson, 1986; Murdock, 1974;Neath & Crowder, 1990). Neill, Valdes, Terry, and Gorfein (1992) suggested that a longer pre-prime interval should increase the temporal discriminability of the prime-trial processing episode, and thereby increase the magnitude of negative priming. Two recent experiments by Neill and Kahan (1997) directly confirmed this prediction, and also again demonstrated effects of contextual similarity. Neill and Kahan (1997) manipulated the delay (more precisely, responsestimulus interval, or RSI) between successive trials in a letter-identification task similar to that used by Neill (1997). Two experiments found statistically equivalent negative priming effects for prime-probe intervals of 500 and 4000 msec. However, both experiments found greater negative priming when the interval preceding the prime was 4000 msec. In addition, the effect of pre-prime interval interacted with prime-probe interval: As shown in Fig. 4, a prime caused more negative priming if the prime-probe interval was identical to the pre-prime interval. (The two experiments yielded essentially the same results; Fig. 4 displays the results of Experiment 1.) Apparently, pre-prime interval (or more likely, psychological variables that covary with interval, such as alertness or impatience) creates a context that influences the retrieval of prime-trial information. Experiments by DeSchepper and Treisman (1996; Treisman & DeSchepper, 1996) provide further support for the episodic retrieval theory. Subjects Pre-Probe Interval 4000ms
500 ms
...................................
................
................
4000 ms 500 ms Pre-Prime Interval
Fig.4. Negative priming as a function of pre-prime interval and pre-probe interval (Neill & Kahan, 1997).
18
W. Trammell NeiU and Katherine M. Mathis
were required to decide whether two novel, meaningless shapes were the same or different. One of the shapes, drawn in green color, was overlapped by an irrelevant shape drawn in red color. Negative priming occurred when the red shape on a prime trial reappeared as a green shape on the probe trial. Remarkably, DeSchepper and Treisman found in several experiments that negative priming persisted across 200 or more trials intervening between the prime and probe. In at least two experiments, negative priming appeared to persist as much as a month! (Compare with similar durations for repetition priming reported by Kolers and Ostry, 1974, and Sloman et al., 1988.) Most of the experiments by DeSchepper and Treisman used a large stimulus ensemble, such that subjects experienced each shape only once before a probe trial. DeSchepper and Treisman argued that negative priming could not have resulted from the inhibition of a preexisting cognitive representation, because the shapes had not been experienced prior to their appearance as distractors in the matching task. Instead, the negative priming must be due to a persisting trace of the specific perceptual encounter, together with information reflecting the fact that it had been ignored rather than attended-in present terms, a processing episode. It might be argued that the distinction between an instance and an abstract representation is ambiguous if a pattern has been experienced only once. However, an experiment by Treisman and DeSchepper (1996, Experiment 4) on the effects of two exposures clearly indicates the persistence of two separate episodes: Some patterns were first presented as to-be-ignored distractors and then on the next trial as targets. Subjects were then tested on old and new patterns after a delay of one day, one week, or one month. After one day, the old patterns produced positive priming relative to the new patterns, suggesting that the attended experience might have erased the effect of the previously ignored experience. However, negative priming showed recovery after a week, and by a month reached the same level as had been found for patterns that had only been ignored. Hence, at the shortest delay (one day) the more recent experience dominated in priming, but at the longer delays priming was dominated by the earlier experience. It may be noted that this is exactly the pattern expected from the classic research on proactive and retroactive interference in the A-B, A-D paradigm: Immediately after acquisition of the second pairing (A-D), retroactive interference is greatest, and there is little proactive interference. However, the effect of the first pairing (A-B) recovers over time, such that there may be more proactive interference than retroactive interference at long delays (Koppenaal, 1963; Postman, Stark, & Fraser, 1968; Underwood, 1948).
Transfer-InappropriateProcessing
19
Both the abstract-inhibition and episodic-retrieval theories, as initially formulated, have assumed that a distractor stimulus must be deliberately ignored during the prime trial in order to produce negative priming. According to the abstract-inhibition theories, the distractor representation is inhibited as a consequence of selective attention to the prime-trial target; according to the episodic-retrieval theory, the processing episode includes information that the distractor was ignored. However, a series of experiments by Milliken and colleagues (Milliken & Joordens, 1996; Milliken et al., in press; Wood & Milliken, in press) demonstrates that it is not necessary to deliberately ignore a stimulus in order for the initial exposure to cause negative priming. In experiments by Milliken et al. (in press), naming of target words showed negative priming from single prime words (in white) that did not require a response, if the target word (in red) was accompanied by a distractor word (in green). Negative priming was obtained if subjects were instructed to ignore the prime word, if subjects received no instructions at all regarding the prime word, or if the prime word was rendered unreportable by a pattern mask. The prime word caused positive priming if subjects were instructed to attend to the prime word, or if the probe target was presented without a distractor. Because Milliken et al. manipulated presence of a distractor between experiments, Neill, Kahan, and VerWys (1996) considered the possibility that subjects may have processed the prime word differently depending on their expectations of whether a distractor would accompany the target. Neill et al. therefore manipulated presence or absence of a distractor randomly over trials, and also counterbalanced the colors of the target and distractor words across subjects. In one experiment, the prime duration was 33 msec, followed by a pattern mask. Whereas practically none of the subjects in Milliken et al. (in press) reported seeing the prime words at this duration, roughly half of the subjects in this experiment reported awareness of the primes. However, awareness of the primes made no difference to the results, and as shown in Fig. 5, the results clearly replicated Milliken et al. The masked prime word caused positive priming when the target word was presented alone, but negative priming when the target was accompanied by a distractor. In a second experiment, the prime duration was 200 msec, at which the prime was clearly visible. Subjects were instructed to name only the word appearing in the designated target color, but were given no explicit instructions regarding the prime. Here, the presence of a distractor failed to reverse positive priming to negative; nonetheless, it significantly reduced the positive priming effect. Recent experiments by Wood and Milliken (in press) demonstrate that even attended stimuli can result in negative priming. In the study phase of
W.Trammell Neill and Katherine M. Mathis
20
Target Condition Distractor 15
No Distractor
I
............................ ...............................
. . . . . . . .
2-15 -20
'
. . . . . .
...
.........
...
................
...
...
..................................
I
I
33 rns
200 m8
Prime Duration
Fig. 5. Negative priming as a function of prime duration and presence or absence of a distractor word (Neill, Kahan, & VerWys, 1996).
one condition, subjects counted the corners of individual nonsense shapes similar to those used by DeSchepper and Treisman (1996; Treisman & DeSchepper, 1996). In the study phase of another condition, subjects wrote down what each shape reminded them of. Subjects then performed a matching task like that in the experiments by DeSchepper and Treisman, on the studied (old) and new patterns. Subjects who counted corners in the study phase responded more slowly to old patterns than new patterns, whereas the subjects in the other study condition showed no significant priming effects. As in other experiments, the negative priming effect depended on presence of a probe-trial distractor: In a control experiment in which the target and comparison stimuli were presented without a distractor, subjects who had counted corners showed no negative priming. That negative priming can be obtained even when the priming stimuli are directly attended is certainly problematic for both the abstract-inhibition and episodic-retrieval theories, as originally formulated. It does not seem possible for even a revised abstract-inhibition theory to account for such results, because there is no reason that the representation of an attended prime should be inhibited. However, the episodic-retrieval theory can be rescued if it is assumed that negative priming depends on whether the initial processing of the prime stimulus was compatible or incompatible with the probe-trial task demands-not just whether it was ignored. It seems likely that the counting-corners study condition induced subjects to
Tramfer-InappropriateProcessing
21
attend to features of the figures, rather than the whole shape. In contrast, judging what the figure reminded them of probably induced the subjects to attend to the whole shape. If the subsequent shape-matching task is also relatively “wholistic,” then retrieved episodes in which the shape was processed more analytically might interfere with matching. This possibility will be developed further in Section VI of this chapter.
V. Related Phenomena
In the prototypical demonstration of negative priming, an attended primetrial target is accompanied by a distractor stimulus to be ignored. The response to a probe target is then slower or less accurate if its identity is the same as, or related to, the identity of the ignored distractor. However, there are a variety of other experimental paradigms in which a relation between two successive stimuli results in impaired responding to the second stimulus. A few of these are briefly reviewed in this section.
A. ENDOGENOUS NEGATIVE PRIMING Neill (1979) examined the effect of receiving an unexpected stimulus in either the same category or a different category as an expected stimulus, in a letter-matching task adapted from Posner and Snyder (1975). On each trial, subjects first saw a warning signal that could be a letter, a digit, or a neutral plus sign. They then received a pair of letters to be judged as “same” or “different.” If the warning signal was a letter or digit, there was an 80%probability that the same character would appear in the match pair. On 10%of such trials, the unexpected stimuli were drawn from the same category (letters or digits) as the prime; on the remaining 10%they were drawn from the opposite category. Responses to expected stimuli were much faster than the control condition in which no letter or digit was primed, and responses to unexpected stimuli were slower, as found by Posner and Snyder (1975). The surprising result was that there was significantly more “cost” to unexpected stimuli from the same category as the prime, than to unexpected stimuli from the opposite category. Neill concluded that members of the same category as the prime were suppressed, in order to circumvent possible confusions. For example, if the prime A activates category member B, then the same-pair AA could be misjudged as AB, or the different-pair AB could be misjudged as BB. Various studies have suggested that polysemous words activate their multiple meanings automatically and in parallel, although the relative access times may be influenced by context and meaning frequency (Neill,
22
W.Trammel1 NeiU and Katherine M. Mathis
Hilliard, & Cooper, 1988; Simpson, 1981, 1984). If one meaning is appropriate to the context, the activation of other meanings should interfere with comprehension, creating a conflict analogous to the Stroop (1935) colorword task and similar selective-attention conditions. Several studies have, in fact, demonstrated a negative priming effect for words that are related to a homograph prime, but inappropriate to the context (Hekkanen, 1981; Marcel, 1980;Simpson & Kellas, 1989). In the absence of a selective context, negative priming occurs for words related to the subordinate meaning of homographs (Simpson & Burgess, 1985); further, this effect appears to be specific to right-visual-field presentations, that is, left-hemisphere processing (Burgess & Simpson, 1988). Neill (1989) suggested the phrase endogenous negative priming to describe negative priming effects that reflect competing internal associations to an attended stimulus, in contrast to exogenous negative priming caused by externally present distractors. Roediger and Neely (1982) suggested the term retrieval blocks to describe similar effects in episodic and semantic memory tests (see also Anderson and Neely, 1996).Several lines of evidence suggest that similar mechanisms underlie exogenous and endogenous negative priming. For example, both types of negative priming are more likely to occur if instructions to subjects stress accuracy: negative priming may reverse to positive if subjects are instructed to sacrifice accuracy for greater speed (Neill, 1979; Neill & Westberry, 1987; Neumann & DeSchepper, 1992). Both types of negative priming also depend on the stimulus-onset asynchrony (SOA) between prime and probe: With a very short SOA, a distractor is more likely to produce positive priming, with negative priming developing at longer SOAs (Burgess & Simpson, 1988; Hekkanen, 1981: Lowe, 1979; Marcel, 1980; Yee, 1991). OF LOCATION B. NEGATIVE PRIMING
In most of the studies cited above, negative priming was specific to the identity of a target stimulus. The selective-attention tasks used in these studies typically required the subject to first select a target by location (or some attribute like color associated with a location), and then to respond according to its identity. Tipper, Brehaut, and Driver (1990) noted that action often depends on selection by identity, and then responding according to location, for example, searching for one’s coffee cup and then reaching for it. They devised a “select-whathespond-where”task in which a target symbol (@) appeared in one of four possible locations, and a distractor (+) appeared in another. Subjects were instructed to press a key corresponding to the location of the target. If a probe-trial target appeared in the same location as the ignored prime-trial distractor, the localization
Transfer-Inappropriate Processing
23
response was slower than if the target appeared in a new location. Hence, negative priming can occur for the location of an ignored object, as well as its identity. (Tipper et al. concluded that the negative priming was really “object-specific,” insofar as negative priming occurred when the ignored object appeared to move to a new location.) Location-specific (or object-specific) negative priming, like identityspecific negative priming, is not simply inhibition of a response. In one experiment by Neill, Valdes, and Terry (1992; see also Neill et al., 1995), subjects were required to press one key for either of two locations, and another key for the other two locations. If a probe target appeared in the same location as the prime-trial distractor, the response was significantly slower than if the probe target appeared in the other location requiring the same response. There has been some controversy, however, over whether locationspecific (or object-specific) negative priming is due to the same mechanism as identity-specific negative priming. In a critical experiment by Park and Kanwisher (1994), subjects were instructed to respond to the location of a target “X”, ignoring a distractor “O”, on a prime trial. The target and distractor definitions were then reversed for the probe trial, such that subjects responded to a target “0”and ignored a distractor “X”. Contrary to what would be expected if the prime-trial distractor location had been inhibited, a probe-trial target appearing in that location resulted in facilitation of responding, that is, positive priming. Furthermore, the response to a probe-trial target appearing in the same location as the prime-trial target was slowed relative to a new location. Park and Kanwisher argued that negative priming of location is really due to an “object-file mismatch”. Following Kahneman, Treisman, and Gibbs (1992), they assumed that the identities of the prime-trial target and distractor are encoded in “object files” associated to their locations. If an object appears in a location for which an object file was created, the object file is retrieved. If the identity of the current object matches the record in the object file, processing is facilitated; however, if the identities mismatch, then the identity information in the object-file may hamper the identification of the current object. The Park and Kanwisher explanation of location-specificnegative priming does not directly account for identity-specificnegative priming, because object-file mismatch typically occurs in both the primed and unprimed conditions. For example, in the flanker task, a new identity appears in the same location as the prime-trial target for both primed trials (e.g., ABA followed by CAC) and unprimed trials (e.g., DBD followed by CAC). On the other hand, it is conceivable that in an identification task, the retrieval of an object file is cued by identity rather than location. Thus, a probe
24
W. Trammel1 Neill and Katherine M. Mathis
target A may cue the retrieval of an object file containing the identity “A” along with the location information “to-be-ignored,” which conflicts with the identification of the target location as “to-be-attended.’’ In this case, an object-file mismatch explanation of identity-specific negative priming is essentially the same as the episodic retrieval theory (Neill & Valdes, 1992). Indeed, there is little difference between the concept of an “object file” and that of a “processing episode” if object-files are permitted to include response information (cf. DeSchepper & Treisman, 1996; Treisman & DeSchepper, 1996). Milliken, Tipper, and Weaver (1994; Tipper, Weaver, & Milliken, 1995) argued that location-specific negative priming cannot be entirely due to object-file mismatch, on the basis of experiments that found negative priming without a mismatch of identity. For example, subjects in one experiment were instructed to respond to the location of the larger of two circles. If the larger circle on a probe trial appeared in the same location as the selected-against circle on the prime trial, negative priming occurred even when the absolute sizes were identical. However, from the perspective of the “transfer-inappropriate processing” approach developed in the next section, negative priming may occur because the processing associated with the location of the prime-trial distractor is incompatible with the processing of a probe-trial target appearing in that location. Whether the conflict is between “identity A” and “identity B” or between “ignore this” and “respond to this” is secondary to the hypothesis that negative priming is episodically based. In this regard, negative priming of location may be affected by contextual similarity between prime and probe trials, like negative priming of identity. Neill, Valdes, and Kennedy (1994) investigated the effects of stimulusresponse compatibility on negative priming in the target-localization task. Previous research (Neill, Valdes, &Terry, 1992)found that negative priming was greater if subjects made spatially incompatible responses, than if they made compatible responses. However, compatibility was manipulated as a between-subjects variable, and prime-trial compatibility was completely confounded with probe-trial compatibility. In the experiment by Neill, Valdes, and Kennedy, target stimuli could be either “S”, requiring a spatially compatible response, or “O”, requiring the spatially opposite keypress; in either case, subjects had to ignore a distractor “X” in another location. As shown in Fig. 6, significant and roughly equivalent negative priming occurred if compatible responses were made on both prime and probe trials, or if incompatible responses were made on both trials. No negative priming occurred if response assignments were switched between trials.
Transfer-Inappropriate Processing
25
Probe S-R Compatibility Compatible
Incompatible
50
Compatible Incompatible Prime S-R Compatibility
Fig. 6. Negative priming as a function of stimulus-response compatibility on prime and probe trials (Neill, Valdes, & Kennedy, 1994).
It may be noted that this result is incompatible with the explanation proposed by Park and Kanwisher (1994). That is, a mismatch of identities occurs at the prime-distractor location, regardless of what S-R mappings are required on the prime and probe trials. Neill, Valdes, and Kennedy (1994) interpreted these results as indicating an inhibition of the specific mapping between the ignored prime-trial distractor and the response associated to it in the response set selected by the prime-trial target identity. However, this interpretation is inconsistent with the finding by Neill and Terry (1995) that negative priming occurs for the accuracy of tachistoscopic recognition, with unlimited time to respond. That is, if negative priming is caused by difficulty of response selection per se, one would expect that subjects should eventually overcome any inhibition and make the correct response. An alternative interpretation of the results of Neill et al. (1994) is that the target cued the appropriate set of S-R mappings, which established a context for retrieving relevant processing episodes in which the same response assignments were required. (Neill, Valdes, and Terry, 1992, found that a change in target identity did not by itself eliminate negative priming.)
C. INHIBITIONOF RETURN If attention is attracted to a spatial location, subsequent identification and responding to a stimulus at that location is typically facilitated (e.g., Eriksen & Hoffman, 1973; Posner, 1980; Posner, Davidson, & Snyder, 1980).
26
W. Trammel1 Neil1 and Katherine M. Mathis
However, if attention is attracted to a location and then withdrawn from that location, responses may be slower to a stimulus at the initially attended location than to a stimulus at a new location (Posner & Cohen, 1980, 1984; Maylor, 1985; Maylor & Hockey, 1985). This phenomenon is called inhibition of return (IOR); that is, it appears that attention is inhibited from returning to a location that it has just left. Posner and Cohen (1980) argued that IOR is an evolutionary adaptation of the visual system to favor sampling of new locations in the visual field, instead of inefficiently resampling old locations. Posner and Cohen (1980) initially demonstrated IOR in an experiment in which subjects responded to the onset of a bright probe dot appearing in either a centrally fixated frame or in one of two frames to the left and right of fixation. At the beginning of a trial, one of the two peripheral frames briefly increased in brightness. The target appeared after a delay of 0, 50, 100, 200, 300, or 500 msec, in either the central frame (80% of trials), the cued frame (10%)or the opposite frame (10%). At the shortest delays, reaction time was faster to a probe appearing in the cued frame than in the opposite frame. But for 300- and 500-msec delays, this effect was reversed: reaction time was actually slower at the cued frame. Presumably, the brightening of a frame first attracted attention to that location. At short delays, target detection at the cued frame benefited from attention to that location. However, subjects withdrew their attention from that location because the target was expected to be at the central frame. By 300 msec, attention was successfully reoriented to the center, and a return to the cued location was inhibited. In the “cue-target’’ procedure used by Posner and Cohen and many other researchers, the cue that attracts attention to a location does not have any overt relevance to the task. However, IOR also occurs in a “target-target” procedure, in which the target stimulus occurs in the same location as another target stimulus to which the subject has just responded (Maylor, 1985; Maylor & Hockey, 1985;Posner, Cohen, Choate, Hockey, & Maylor, 1984). The relation between IOR and negative priming of location is unclear. Negative priming is usually demonstrated for an ignored location, whereas IOR is demonstrated for an attended location. However, it is possible that attention is initially attracted by a prime-trial distractor, and then withdrawn in order to attend selectively to the target. In fact, just as negative priming is object-specific, rather than simply location-specific (Tipper et al., 1990), IOR also appears to be at least in part object-specific. Tipper, Driver, and Weaver (1991) required subjects to make detection responses to a probe dot appearing in one of two frames moving clockwise around fixation. As in Posner and Cohen (1980,1984), one of the frames was briefly brightened
Transfer-InappropriateProcessing
21
before the target appeared. IOR occurred for a target in the cued frame, even though the frame had moved to a new location. Tipper, Weaver, Jerreat, and Burak (1994) obtained similar results, although they also obtained some IOR at the cued location independent of the cued object. A problem for simply attributing negative priming to IOR (or vice versa) is that many experiments on negative priming have found facilitation, rather than inhibition, for probe targets appearing in the same location as the attended prime target (Neill, Valdes, & Terry, 1992;Neill, Terry, & Valdes, 1994; Park & Kanwisher, 1994). Some of these experiments may not have sufficiently encouraged subjects to withdraw attention from the primetarget location; thus, if attention was still oriented to that location, processing of a probe target there would benefit. However, in one experiment, Neill, Valdes, and Terry (1992) investigated location-specificpriming across an intervening trial: On a prime trial, a target 0 and distractor X appeared in two of four possible locations. On the intervening trial, a target 0 appeared alone in one of the two remaining locations. A probe target 0 then appeared alone in any of the four locations. Relative to the previously unoccupied location, a probe at the prime-distractor location caused significant negative priming, whereas a probe at either of the two previous target positions caused significant positive priming. It can be safely assumed that attention was withdrawn from the prime-target location, because subjects would need to orient to the intervening location. Terry, Valdes, and Neill (1994) noted that previous demonstrations of IOR had used either simple onset-detection or localization of a single stimulus onset. In contrast, negative priming experiments required subjects to discriminate between target and distractor identities in order to select the appropriate response. Terry et al. then tested whether IOR occurs in discrimination tasks. In one experiment, subjects simply pressed a key if a target letter (A or B) appeared to either side of fixation. In the detection condition, no letter appeared on nontarget trials; in the discrimination condition, another letter (C or D) appeared instead. In a second experiment, subjects pressed one of two keys corresponding to the location of a target letter. In the detection condition, the opposite location was blank; in the discrimination condition, another letter appeared in the opposite location. Target-target IOR occurred in the detection conditions, but not in the discrimination conditions (see also Tanaka & Shimojo, 1996). Pratt (1994), however, found IOR in a discrimination task, using the “cue-target” procedure. Here, one of two peripheral locations was cued by a 300-msec asterisk, followed by a cue to reorient attention to the center. Subjects were required to make an eye movement toward the location of a specified shape (box or diamond). In the detection condition, only the target appeared in one of the two peripheral locations; in the discrimination
28
W. Trammell Neill and Katherine M. Mathis
condition, the nontarget shape appeared in the opposite location. In contrast to the results of Terry et al., Pratt found equivalent IOR for the two tasks. In an attempt to clarify the discrepant results, Pratt and Abrams (in press) found similar results in discrimination tasks using both keypress responses and a “target-target” procedure. However, in two experiments using eyemovement responses in the target-target procedure, no IOR was found if the same target was repeated across trials, and a reversal was found in one of the experiments. Pratt and Abrams suggested that an initially foveated target might produce a location-specificrepetition priming effect that would counteract IOR. Because the stimuli used by Terry et al. (1994) were fairly close to fixation (centered 1.1”to the left or right), Pratt and Abrams suggested that location-specific repetition priming might also account for the earlier study’s results. It remains unclear, however, whether location-specificrepetition priming provides a sufficient explanation for the absence of IOR in some discrimination tasks. In a third experiment, Terry et al. required subjects to press one key if an A or B appeared to the left or right of fixation, and the other key if a C or D appeared. A facilitation occurred for a changed target occurring at the same location, as long as the two targets required the same response (e.g., A then B in the same location). Neill, Valdes, and Terry (1992) similarly found facilitation for a changed probe target appearing in the prime-target location (albeit reduced, relative to identical targets). If IOR can be offset by repetition priming specific to the initially oriented location, a changed target should still show IOR. In the absence of a definitive understanding of when IOR does or does not occur, the question of whether IOR and location-specific negative priming are due to a common mechanism remains open. If IOR does not occur under certain conditions under which negative priming does occur, then they must be caused by different mechanisms. However, if IOR and negative priming share similar constraints, then negative priming might be plausibly attributed to IOR.
D. REPETITION BLINDNESS In “rapid serial visual presentation” (RSVP), a subject is shown a rapid sequence of stimuli such as letters or words, typically all at the same spatial location. The subject is then required to report either all of the items, or certain items specified as targets. Kanwisher (1987) discovered that if an item is repeated in the sequence, subjects often miss the second occurrence, hence, repetition blindness. In sequences of words that together comprised sentences, repetition blindness occurred despite strong grammatical constraints. For example, if the sequence consisted of “When the ink spilled,
Transfer-InappropriateProcessing
29
there was ink all over,” subjects often reported seeing “When the ink spilled, there was all over.” Yet, if nonrepeated words were presented in the same positions, the second critical word was recognized easily (for example, “When the liquid spilled, there was ink all over.”). Kanwisher (1987) explained repetition blindness as a failure to discriminate between the two instances of the repeated word. More specifically, she distinguished between the activation of an abstract “type” required for identification, versus the formation of a specific “token” representing a specific occurrence of that type. Tokenization is required in order to report the stimulus as having occurred at a particular time and place. If a token has just been created for an activated type, a second token for that same type cannot be created for some duration afterwards. Kanwisher maintained that type activation occurs independently of token individuation, on the basis of an experiment in which subjects were required to identify only the last word in a sequence of words: If the same word appeared earlier in the sequence, identification accuracy of the target word was enhanced. According to Kanwisher, the task did not require token individuation, and so simply benefitted from type activation. Whittlesea, Dorken, and Podrouzek (1995) failed to replicate Kanwisher’s finding of facilitation in three similar experiments in which subjects were required to pronounce the last word as quickly as possible. However, as pointed out in a reply by Downing and Kanwisher (1995), the subjects in Whittlesea et al. were also required to judge whether or not the target word had appeared earlier in the list; hence, subjects would have been obligated to individuate the first and second occurrences. Neil1 and Cestaro (unpublished data, 1993) found results similar to Kanwisher’s in an experiment designed to explore “illusory conjunctions” in RSVP (cf. Botella & Eriksen, 1992; Botella, Garcia, & Barriopedro, 1992). Subjects were required to identify a single target word printed in capital letters among lowercase distractor words. Sequences of 10 words were presented at 67 msec per word. The target word appeared at the sixth or seventh position in the sequence. On a third of the trials, the third or fourth word was identical to the target, except in lower case. In another third of the trials, the word immediately after the target was primed by an identical word earlier in the list. In the remaining third, no word was repeated. Priming the post-target distractor significantly increased intrusions by that word, relative to the control condition (20% versus 4% of trials). More pertinent to the present issue, priming the target word also increased the likelihood of its report (42% versus 28%). Repetition blindness in RSVP therefore presents a paradox: If the first occurrence is attended (i.e., requires report), then report of the second occurrence is suppressed. However, if the first occurrence is unattended
30
W. Trammel1 Neil1 and Katherine M. Mathis
(i.e., does not require report), it facilitates the report of the second occurrence. Such results appear to confound expectations based on repetition priming and negative priming: When a target is accompanied by simultaneous distractors, the attended target facilitates report of a subsequent identical target, whereas an ignored distractor hampers the report of a subsequent identical target. In addition to Kanwisher’s (1987) token-individuation hypothesis, a variety of competing explanations for repetition blindness have been proposed, ranging from those that emphasize the initial perception of the repeated stimulus, to those that emphasize competition for memory retrieval at the time of report. In experiments by Luo and Caramazza (1995), subjects were presented with only two spatially separated letters on each trial, which should presumably minimize contributions of memory-retrieval difficulties. Repetition blindness occurred when the two letters were identical. On the other hand, repetition blindness also occurred when subjects were required to report just the second letter, which should not, according to Kanwisher, require token individuation. Luo and Caramazza argued that repetition blindness must then be due to refractoriness of type activation; that is, the second occurrence was not identified. At the other end of the spectrum, Fagot and Pashler (1995) argued that repetition blindness is purely a memory-retrieval phenomenon. In one experiment, they found repetition blindness for the second occurrence of a letter (imbedded in a stream of six letters) if subjects reported the letters in order of presentation, but not if they reported them in reverse order. In another experiment, one letter in a sequence was colored red; after a sequence, the subject was cued to either first report the red letter, or first report the whole sequence. Although whole report yielded repetition blindness (even if the repeated letter was the red letter), a repeated letter did not show any deficit if it had to be reported as the red item. Similarly, Armstrong and Mewhort (1995) found repetition blindness in whole report, but not in cued report, in which subjects were provided with one item in the list and required to report the next item. To further complicate matters, Neill, VerWys, and Neely (1997) obtained results that appear to contradict all of the extant theories. Researchers have tended to assume that repetition blindness occurs for the second of two repeated items. However, in many experiments, report may be biased in favor of the first item by the requirement (or expectation) that items should be repeated in order. In many other experiments, repetition blindness is measured by the correct report of both critical items, in which case it is not possible to distinguish between effects on report of the first item and report of the second item. A similar problem arises if studies only
Transfer-InappropriateProcessing
31
describe the performance on the second item, either unconditionally or conditional on correct report of the first item. On most trials, Neill et al. presented two letters (selected from A, B, C, and D) successively in two separate locations (left and right of fixation), followed by pattern masks, similar to the procedure used by Luo and Caramazza (1995). Luo and Caramazza required report of both stimuli in their first experiment (and scored only correct report of both letters) and required report of only the second letter in their second experiment. In neither case can one determine the effects of repetition on report of the first letter. Neill et al. instead randomly probed either the left letter or right letter for report. In two experiments, only one target was probed on each trial; in a third, both targets were probed in random order. All three experiments found a significant repetition blindness effect for the first presented target as well as the second, and in two of the experiments the magnitude of the repetition deficit was significantly greater for the first target than the second! On one-fifth of the trials in these experiments, no second target was presented. In the experiments in which only one location was probed on each trial, the “blank” location was probed on half of such trials. In the experiment in which both locations were probed, the blank location was randomly probed before or after the target location was probed. Subjects exhibited no bias either toward or against reporting the same item as the target in the opposite location. Furthermore, on two-target nonrepetition trials in which one target was reported incorrectly, subjects showed no bias either toward or against reporting the same item as had appeared in the other location. In other words, subjects were not biased against reporting the same item twice, contrary to the “censorship bias” hypothesized by Fagot and Pashler (1995). Furthermore, in the experiment in which both targets were probed on each trial, repetition blindness did not interact with order of report. And, contrary to both Fagot and Pashler (1995) and Armstrong and Mewhort (1999, repetition blindness was manifested here in a cued report procedure. On the other hand, the fact that repetition blindness occurred as strongly for the first item as for the second is inconsistent with either of the two more perceptual theories offered for the phenomenon. If activation of a type node is refractory after initial activation (Luo & Caramazza, 1995), or if tokenization of the first target interferes with subsequent tokenization of the second target (Kanwisher, 1987), then repetition blindness should occur primarily for the second target. A modification of Kanwisher’s theory might, however, account for the results. “Tokenization” may really consist of two separate processes: (a) creation of a tokenperse, as the instantiation of a type, and (b) association of that token to a particular time or place of
32
W. Trammell NeU and Katherine M. Mathis
occurrence. Hence, repetition blindness may reflect a failure to individuate the two occurrences of the same target identity. But, there is still a choice between two times or locations to which the token should be associated. In a task in which the choices are spatially separated, and report itself is spatially cued, subjects may often attribute the singly perceived object as having appeared at the most recent location. E. THEBEFORE-DISRUFTION EFFECT Reicher (1969) first noted that identification of a brief tachistoscopic letter was depressed if subjects saw the relevant alternatives immediately before the target presentation, rather than after the target presentation. Smith, Haviland, Reder, Brownell, and Adams (1976) confirmed this beforedisruption effect in experiments requiring either forced-choice recognition (two alternatives) or same-different matching (one alternative). The effect is quite counterintuitive, because one might expect a benefit of pre-exposure due to either (a) narrowing down the possible alternatives before the target, or (b) a repetition priming effect. Smith et al. attributed before-disruption to subjects erroneously identifying masking pattern features as part of the target that they were anticipating. Consistent with this explanation, subjects showed a strong bias to respond “same” if a single alternative was preexposed. In addition, no before-disruption occurred if target visibility was limited by reduced illumination rather than masking; in fact, pre-exposure of the alternatives facilitated identification under this condition. A series of experiments by Neill and Walling (1981; Neill, 1985),however, posed difficulties for the explanation that before-disruption is caused by the incorporation of mask features into the target perception. Several experiments indicated a dissociation between the before-disruption effect and the same-bias effect. In one experiment, one group of subjects always selected between similar alternatives (0-Q, F-E, P-R, C-G), whereas another group selected between dissimilar alternatives (0-E, F-Q, C-R, P-G). Subjects presented with similar alternatives exhibited beforedisruption, whereas the other group did not. However, both groups showed a similar bias to respond “same” when precued with the alternative. Furthermore, subjects in both groups were as likely to make errors on a target that possessed the distinguishing feature, as on a target that lacked the distinguishing feature, contrary to the hypothesis that precuing led subjects to incorporate mask features into the target. Neill (1985) hypothesized that precuing with similar alternatives led subjects to attend to a level of processing (e.g., feature detection) that was more susceptible to disruption by a pattern mask. In contrast, on trials that were not precued, or were precued with dissimilar alternatives, subjects
Transfer-InappropriateProcessing
33
were likely to attend to a more abstract level of processing (e.g., letter name). In support of this explanation, precuing with similar alternatives did not cause before-disruption if subjects were uncertain of the target letter-case, even though before-disruption did occur for the same alternatives and targets if subjects could anticipate the letter-case. Another experiment found facilitation from precuing when the stimuli were unfamiliar patterns; presumably, such patterns could not be easily represented at a more abstract level, and so processing was constrained to the same level for all conditions. In a third experiment, subjects were required to make both detection (yesho) and two-choice discrimination judgments on each trial. Although discrimination was worse on precued trials, detection was actually better, supporting the hypothesis that precuing led subjects to attend more to visual appearance than letter name. The explanation for before-disruption posed by Neil1 (1985) assumes that the effect reflects a processing strategy (albeit maladaptive) adopted by the subject. However, phenomena like repetition blindness and even negative priming from attended objects (Wood & Milliken, in press) raise the possibility that mere attention to the precue (or its features) might be sufficient to impair identification of a similar target. The experiment remains to be done in which subjects attend to certain letters beforehand, but do not expect them to be relevant to target identification. VI. Toward a TIP-TAP Theory of Priming
Negative priming, inhibition of return, repetition blindness, and beforedisruption are all cases in which the initial processing of a stimulus somehow disrupts the subsequent processing of a similar stimulus. In a quite literal sense, they demonstrate “transfer-inappropriate processing,” which may or may not be related to “proactive interference” as described in the classic verbal learning literature. As an explanatory construct, transfer-inappropriate processing may be marginally superior to transferappropriate processing: Whereas it seems trivially obvious that task performance should benefit from the prior acquisition of task-relevant information, it is perhaps a little less obvious that task performance should suffer from the acquisition of irrelevant information. Nonetheless, “transferinappropriate processing” by itself is simply a restatement of the fact that repetitions can hurt performance. Rather than invoking transfer-appropriate processing (TAP) and transfer-inappropriate processing (TIP)as explanatory constructs, a more appropriate goal is to specify conditions under which repetitions are likely to facilitate performance (TAP)or hamper performance (TIP). It is possible
34
W. Trammel1 Neill and Katherine M. Mathis
that exogenous identity-specific negative priming, endogenous negative priming, location-specific negative priming, inhibition of return, repetition blindness, and before-disruption (and any other superficially similar phenomena) are all caused by qualitatively different mechanisms, and as such require quite unrelated explanations. On the other hand, it is possible that such phenomena can be described within a common set of processing assumptions; in the best of all worlds, it may be possible to formulate a testable theory of TIP and TAP. Toward this goal (however naive it may be), we consider here the possibility that a modified episodic-retrieval theory of negative priming might be extended to describe other TIP phenomena. The modified theory makes the following assumptions: 1. A perceived stimulus automatically activates multiple abstract representations in memory (i.e., “type nodes”; cf. Kanwisher, 1987). 2. Further processing (e.g., response selection) is influenced by two processes that occur in parallel: (a) controlled, algorithmic processing based on whatever activated type is relevant to immediate goals; (b) cued retrieval of previous processing episodes containing tokens (instances) of the same activated type (cf. Logan, 1988). 3. Processing episodes (or “object files”; cf. Kahneman et al., 1992) are assumed to include not only information about stimulus identity (i.e., an instance or token corresponding to a particular type), but also information about how that instance was processed. 4. Retrieval of a past processing episode tends to reinstate similar processing of the present instance (cf. Kolers, 1976; Kolers & Ostry, 1974). 5. Reinstated processing is either transfer-appropriate or transferinappropriate (TAP or TIP) depending on whether it is compatible or incompatible with current goals. 6. Whether retrieval actually affects performance depends on the speed of retrieval relative to the controlled algorithmic processing (cf. Logan, 1988). 7. Repetition can facilitate performance by either episodic TAP retrieval, or persisting type activation (see No.1 above). 8. Repetition hampers performance by episodic TIP retrieval. Point 4 above, which owes much to Kolers (1976; Kolers & Ostry, 1974), constitutes the major substantive change to the episodic retrieval theory of negative priming as previously described (Neill, 1997; Neill & Valdes, 1996; Neill et al., 1995; Neill, Valdes, Terry, & Gorfein, 1992). Whereas it was assumed that negative priming was caused by retrieving information specifically that an instance of the current stimulus had been ignored, it is now suggested that retrieving any processing information that is incompati-
Transfer-InappropriateProcessing
35
ble with current processing requirements may hamper performance. This change seems especially warranted by the demonstration by Wood and Milliken (in press) that negative priming can be caused by certain study conditions in which a figure is unarguably attended. Fortuitously, this modification affords the possibility of extending the theory to other TIP phenomena in which the initial occurrence is attended. In inhibition of return, attention is attracted to a location and then withdrawn. If a processing episode/object file is created for the initially attended location, the included processing information would be that attention was withdrawn from that location. Insofar as withdrawal of attention is incompatible with allocating attention, reinstatement of the processing specified by the retrieved episode would obviously conflict with reorienting to that location. Although attention was initially allocated to that location, which should be compatible with reallocation, the initial allocation would have less effect than the withdrawal on subsequent reallocation. If the initial orienting and the withdrawal are encoded as separate processing episodes, the more recent episode is expected to dominate. Alternatively, if information is “updated” in an object file, as suggested by Kahneman et al. (1992), the withdrawal information might actually replace the initial orientation information. In repetition blindness, the two occurrences of a stimulus might on the surface seem to require the same processing, and so transfer-appropriate facilitation should occur. However, as argued by Kanwisher (1987), the two occurrences do not require identical processing; rather, the task requirements are that each instance be associated with a uniquely different position in the list of items. Hence, one possibility would be that a second token is formed, but retrieval of the first token results in the second token being similarly bound to the earlier position in the list. However, this mechanism would not account for the finding of repetition blindness for the first occurrence (Neil1 et al., 1997). An alternative description would be that there is a failure to individuate separate tokens, but sometimes an association to the second position can override the association to the first position (i.e., retroactive interference instead of proactive interference). If an item in an RSVP sequence is not attended, type activation is still likely to occur, but (unlike an interfering distractor in selective-attention paradigms) it may receive no further processing. A target that is a repetition of the unattended stimulus will then benefit from the persisting activation. Hence, the surface paradox between negative priming experiments and repetition blindness experiments is solved: In the former, attended stimuli result in TAP episodes but ignored stimuli result in TIP episodes. In contrast, in RSVP, attended stimuli result in TIP episodes but ignored stimuli typically just cause activation.
36
W. Trammel1 NeiU and Katherine M. Mathis
The before-disruption effect appears to depend on the subject attending to the featural difference between two similar alternatives. From the present TIP-TAP perspective, identification of a target may suffer because the featural processing is reinstated by a retrieved processing episode, rather than because the subject is actively looking for particular features. This featural processing may then be more susceptible to interference from a pattern mask, than would processing that was directed to more abstract qualities. The theoretical meat of the present TIP-TAP theory is that TIP is episodic, whereas facilitatory effects can occur from either TAP episodes or persisting activation. Some reviews of negative priming (Fox, 1995; May et al., 1995; Tipper & Milliken, 1996) have argued that negative priming is caused by both episodic retrieval and persisting inhibition of a cognitive representation (i.e., a type). In contrast, Neill (1997; Neill & Valdes, 1996) has argued that the episodic retrieval theory is sufficient to account for the extant data on negative priming. This is clearly an issue that remains to be resolved, and a pure episodic-retrieval account may ultimately fail. If so, the present TIP-TAP theory would require modification. A theory that postulates that either episodes or abstract representations can produce both TIP and TAP effects might have too many degress of freedom to be testable. On the other hand, it may still be possible to formulate a theory that would predict more precisely when the mixture of TIP and TAP effects would yield a net facilitation or net retardation of performance. By the same token, it is possible that different mixtures of TIP and TAP effects occur in the various TIP phenomena, that is, negative priming, inhibition of return, repetition blindness, and before-disruption. For example, negative priming may prove to depend entirely on episodic retrieval, but before-disruption may depend entirely on whether attention is directed to visual-level versus name-level types during target processing. A possible program of research might be to determine which phenomena are sensitive to context and other task-irrelevant detail, hallmarks of episodic retrieval.
MI. Summary Although repetitions of stimuli are often beneficial to performance, there are a variety of phenomena in which stimulus repetitions hurt performance, including several forms of negative priming, inhibition of return, repetition blindness, and the before-disruption effect. In general, such effects occur when the processing of the initial occurrence differs from the processing requirements of the second occurrence in some critical way. In other words, the processing of the first occurrence is sometimes “transfer-inappropriate”
Transfer-InappropriateProcessing
37
(TIP) to the second occurrence. Research on negative priming indicates that it is caused, at least in part, by the cued retrieval of past instances or episodes involving a similar stimulus. In contrast, facilitatory effects of repetitions appear to be caused by persisting activation of abstract representations, as well as by episodic retrieval of transfer-appropriate processing (TAP). The present chapter suggests that other cases of transferinappropriate processing may similarly be caused by episodic retrieval processes, and that a general TIP-TAP theory may be possible to describe both the facilitatory and inhibitory consequences of repetition. REFERENCES Anderson, J. R. (1976). Language, memory, and thought. Hillsdale, NJ: Erlbaum. Anderson, J. R., & Bower, G. H. (1973). Human associative memory. Washington, D C Winston. Anderson, M. C., & Neely, J. H. (1996). Interference and inhibition in memory retrieval. In E. L. Bjork & R. A. Bjork (Eds.), Memory (pp. 237-313). San Diego: Academic Press. Armstrong, I. T., & Mewhort, D. J. K. (1995). Repetition deficit in rapid-serial-visualpresentation displays: Encoding failure or retrieval failure? Journal of Experimental Psychology: Human Perception and Performance, 21, 1044-1052. Baddeley, A. D. (1976). The psychology of human memory. New York: Basic. Beller. H. K. (1971). Priming: The effects of advanced information on letter matching. Journal of Experimental Psychology, 87, 176-182. Bertelson, P. (1963). S-R relationships and reaction times to new versus repeated signals in a serial task. Journal of Experimental Psychology, 65, 478-484. Bjork, R. A,, & Whitten, W. B. (1974). Recency-sensitive retrieval processes in long-term free recall. Cognitive Psychology, 6, 173-189. Blaxton, T. A. (1989). Investigating dissociations among memory measures: Support for a transfer-appropriate processing framework. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 657-668. Botella, J., & Eriksen, C. W. (1992). Filtering versus parallel processing in RSVP tasks. Perception & Psychophysics, 51, 334-343. Botella, J., Garcia, M. L., & Bamopedro, M. (1992). Intrusion patterns in rapid serial visual presentation tasks with two response dimensions. Perception & Psychophysics, 52. 547-552. Burgess, C., & Simpson, G. B. (1988). Cerebral mechanisms in the retrieval of ambiguous word meanings. Brain and Language, 33, 86-103. Collins, A. M., & Loftus, E. L. (1975). A spreading-activationtheory of semantic processing. Psychological Review, 82, 407-428. Craik, F. I. M., & Lockhart, R. S. (1972). Levels of processing: A framework for memory research. Journal of Verbal Learning and Verbal Behavior, 11, 671-684. Dalrymple-Alford, E. C., & Budayr, D. (1966). Examination of some aspects of the Stroop color-word test. Perceptual & Motor Skills, 23, 1211-1214. DeSchepper, B., & Treisman, A. (1996). Visual memory for novel shapes: Implicit coding without attention. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 27-47.
38
W. Trammell Neil1 and Katherine M. Mathis
Downing, P., & Kanwisher, N. (1995). Types and tokens unscathed: A reply to Whittlesea, Dorken, and Podrouzek (1995) and Whittlesea and Podrouzek (1995). Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 1698-1702. Ebbinghaus, H. (1913). Memory (H. A. Ruger & C. E. Bussenius, Trans.). New York: Teachers College. (Original work published 1885). Eriksen, B. A., & Eriksen, C. W. (1974). Effects of noise letters upon the identification of a target letter in a nonsearch task. Perception & Psychophysics, 16, 143-149. Eriksen, C. W., & Hoffman, J. E. (1973). The extent of processing noise elements during selective encoding from visual displays. Perception & Psychophysics, 14, 155-160. Fagot, C., & Pashler, H. (1995). Repetition blindness. Journal of Experimental Psychology: Human Perception and Performance, 21, 275-292. Fox, E. (1995). Negative priming from ignored distractors in visual selection: A review. Psychonomic Bulletin & Review, 2, 145-173. Glenberg, A. M. (1987). Temporal context and recency. In D. S. Gorfein & R. R. Hoffman (Eds.), Memory and learning: The Ebbinghaus centennial conference (pp. 173-190). Hillsdale, NJ: Erlbaum. Glenberg, A. M., & Swanson, N. G. (1986). A temporal distinctiveness theory of recency and modality effects. Journal of Experimental Psychology: Learning, Memory, and Cognition, 12, 3-15. Graf, P., & Mandler, G. (1984). Activation makes words more accessible, but not necessarily more retrievable. Journal of Verbal Learning and Verbal Behavior, 23, 553-568. Graf, P., Mandler, G., & Haden, P. (1982). Simulating amnesic symptoms in normal subjects. Science, 218, 1243-1244. Greene, R. L. (1986). Word stems as cues in recall and completion tasks. Quarterly Journal of Experimental Psychology, 38A, 663-673. Hasher, L., Stoltzfus, E. R., Zacks, R. L., & Rypma, B. (1991). Age and inhibition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 163-169. Hasher, L., Zacks, R. T., Stoltzfus, E. R., Kane, M. J., & Connelly, S. L. (1996). On the time course of negative priming: Another look. Psychonomic Bulletin & Review, 3, 231-237. Hekkanen, S. T. (1981). The effects of context on the processing of ambiguous words (Doctoral dissertation, University of South Florida, 1981). Dissertation Abstracts International, 42, 1646B. Hintzman, D. L. (1976). Repetition and memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 10, pp. 47-91). New York: Academic Press. Hintzman, D. L. (1986). “Schema abstraction” in a multiple-tracee memory model. Psychological Review, 93, 411-428. Hintzman, D. L., & Block, R. A. (1971). Repetition and memory: Evidence for a multipletrace hypothesis. Journal of Experimental Psychology, 88, 297-306. Hyde, T. S., & Jenkins, J. J. (1969). The differential effects of incidental tasks on the organization of recall of a list of highly associated words. Journal of Experimental Psychology, 82, 472-481. Hyde, T. S., &Jenkins, J. J. (1973). Recall for words as a function of semantic, graphic, and syntactic orientic tasks. Journal of Verbal Learning and Verbal Behavior, 12, 471-480. Hyman, R. (1953). Stimulus information as a determinant of reaction time. Journal of Experimental Psychology, 45, 188-196. Jacoby, L. L. (1983a). Perceptual enhancement: Persistent effects of an experience. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9, 21-38. Jacoby, L. L. (1983b). Remembering the data: Analyzing interactive processes in reading. Journal of Verbal Learning and Verbal Behavior, 22, 485-508.
Transfer-InappropriateProcessing
39
Jacoby, L. L., & Dallas, M. (1981). On the relationship between autobiographical memory and perceptual learning. Journal of Experimental Psychology: General, 110, 306-340. Jacoby, L. L., & Hayman, C. A. (1987). Specific visual transfer in word identification. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 599-612. Kahneman, D., Treisman, A., & Gibbs, B. J. (1992). The reviewing of object files: Objectspecific integration of information. Cognitive Psychology, 24, 175-219. Kanwisher, N. G. (1987). Repetition blindness: Type recognition without token individuation. Cognition, 27, 117-143. Keele, S. W. (1969). Repetition effect: A memory-dependent process. Journal of Experimental Psychology, 80, 243-248. Kolers, P. A. (1976). Reading a year later. Journal of Experimental Psychology: Human Learning and Memory, 2, 554-565. Kolers, P. A., & Ostry, D. J. (1974). Time course of loss of information regarding pattern analyzing operations. Journal of Verbal Learning and Verbal Behavior, 13, 599-612. Koppenaal, R. J. (1963). Time changes in the strengths of A-B, A-C lists: Spontaneous recovery? Journal of Verbal Learning and Verbal Behavior, 2, 310-319. Leahey, T. H. (1987). A history of psychology: Main currents in psychological thought (2nd ed.). Englewood Cliffs, NJ: Prentice-Hall. Logan, G. D. (1988). Toward an instance theory of automatization. Psychological Review, 95, 492-527. Logan, G . D. (1990). Repetition priming and automaticity: Common underlying mechanisms? Cognitive Psychology, 22, 1-35. Lowe, D. G. (1979). Strategies, context, and the mechanism of response inhibition. Memory & Cognition, 7, 382-389. Luo, C. R., & Caramazza, A. (1995). Repetition blindness under minimum memory load: Effects of spatial and temporal proximity and the encoding effectiveness of the first item. Perception & Psychophysics, 57, 1053-1064. Malley, G. B., & Strayer, D. L. (1995). Effects of stimulus repetition on positive and negative identity priming. Perception & Psychophysics, 57, 657-667. Marcel, A. J. (1980). Conscious and preconscious recognition of polysemous words: Locating the selective effects of prior verbal context. In R. S. Nickerson (Ed.), Attention and performance VIll (pp. 435-437). Hillsdale, NJ: Erlbaum. May, C. P., Kane, M. J., & Hasher, L. (1995). Determinants of negative priming. Psychological Bulletin, 118, 35-54. Maylor, E. A. (1985). Facilitatory and inhibitory components of orienting in visual space. In M. I. Posner & 0. S. M. Marin (Eds.), Attention and performance XI (pp. 189-204). Hillsdale, NJ: Erlbaum. Maylor, E. A., & Hockey, R. (1985). Inhibitory components of externally controlled overt orienting in visual space. Journal of Experimental Psychology: Human Perception and Performance, 1I , 777-787. McGeoch, J. A. (1932). Forgetting and the law of disuse. Psychological Review, 39, 352-370. Melton, A. W., & Irwin, J. M. (1940). The influence of degree of interpolated learning on retroactive inhibition and the overt transfer of specific responses. American Journal of Psychology, 53, 173-203. Melton, A. W., & von Lackum, W. J. (1941). Retroactive and proactive inhibition in retention: Evidence for a two-factor theory of retroactive inhibition. American Journal of Psychology, 54, 157-173. Milliken, B., & Joordens, S. (1996). Negative priming without overt prime selection. Canadian Journal of Experimental Psychology, 50, 333-346.
40
W.Trammel1 NeiU and Katherine M. Mathis
Milliken, B., Joordens, S., Merikle, P., & Seiffert, A. (in press). Selective attention: A reevaluation of the implications of negative priming. Psychological Review. Milliken, B., Tipper, S . P., & Weaver, B. (1994). Negative priming in a spatial localization task: Feature mismatching and distractor inhibition. Journal of Experimental Psychology: Human Perception and Performance, 20, 624-646. Moore, C. M. (1994). Negative priming depends on probe-trial conflict: Where has all the inhibition gone? Perception & Psychophysics, 56, 133-147. Morris, C. D., Bransford, J. D., & Franks, J. 3. (1977). Levels of processing versus transfer appropriate processing. Journal of Verbal Learning and Verbal Behavior, 16, 519-533. Morton, J. (1969). Interaction of information in word recognition. Psychological Review, 76, 165-178. Morton, J. (1979). Facilitation in word recognition: Experiments causing change in the logogen model. In P. A. Kolers, M. E. Wrolstad, & H. Bouma (Eds.), Processing visible language (Vol. 1, pp. 259-268). New York: Plenum. Murdock, B. B. (1974). Human memory: Theory and data. Hillsdale, NJ: Erlbaum. Neath, I., & Crowder, R. G. (1990). Schedules of presentation and temporal distinctiveness in human memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 316-327. Neely, J . H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limited-capacity attention. Journal of Experimental Psychology: General, 106, 226-254. Neill, W. T. (1977a, April). The effect of semantic and nonsemantic orienting tasks on memory for nonsemantic information. Paper presented at the meeting of the Western Psychological Association, Seattle. Neill, W. T. (1977b). Inhibitory and facilitatory processes in selective attention. Journal of Experimental Psychology: Human Perception and Performance, 3, 444-450. Neill, W. T. (1979). Switching attention between categories: Evidence for intracategory inhibition. Memory & Cognition, 7, 283-290. Neill, W. T. (1985). Levels of processing in disruptive effects of prior information. Memory & Cognition, 13, 477-484. Neill, W. T. (1989). Ambiguity and context: An activation-suppression model. In D. S. Gorfein (Ed.), Resolving semantic ambiguity (pp. 63-83). New York: Springer-Verlag. Neill, W. T. (1997). Episodic retrieval in negative priming and repetition priming. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23, 1-15. Neill, W. T., Beck, J. L., Bottalico, K. S., & Molloy, R. D. (1990). Effects of intentional versus incidental learning on explicit and implicit tests of memory. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 457-463. Neill, W. T., Hilliard, D. V., & Cooper, E. A. (1988). The detection of lexical ambiguity: Evidence for context-sensitive parallel access. Journal of Memory and Language, 27, 279-287. Neill, W. T., & Kahan, T. A. (1997, November). Time and context in negative priming. Paper presented at the meeting of the Psychonomic Society, Philadelphia, PA. Neill, W. T., Kahan, T. A., & VerWys, C. A. (1996, November). Reversal of lexicalpriming by probe interference. Paper presented at the meeting of the Psychonomic Society, Chicago. Neill, W. T., Lissner, L. S., & Beck, J. L. (1990). Negative priming in same-different matching: Further evidence for a central locus of inhibition. Perception & Psychophysics, 48,398-400. Neill, W. T., & Terry, K. M. (1995). Negative priming without reaction time. Psychonomic Bulletin & Review, 2, 121-123. Neill, W. T., Terry, K. M., & Valdes, L. A. (1994). Negative priming without probe selection. Psychonomic Bulletin & Review, I, 119-121.
Transfer-Inappropriate Processing
41
Neill, W. T., & Valdes, L. A. (1992). The persistenceof negative priming: Steady-state or decay? Journal of Experimental Psychology: Learning, Memory, and Cognition, IS, 565-576. Neill, W. T., & Valdes, L. A. (1996). Facilitatory and inhibitory aspects of attention. In A. F. Krarner, M. G. H. Coles, & G. D. Logan (Eds.), Converging operations in the study of visualselective attention (pp. 77-106). Washington,D C American PsychologicalAssociation. Neill, W. T., Valdes, L. A., & Kennedy, C. E. (1994, November). Negative priming: Effects of S-R compatibility. Paper presented at the meeting of the Eastern Psychological Association, Providence, RI. Neill, W. T., Valdes, L. A., & Terry, K. M. (1992, November). Negative priming in target localization. Paper presented at the meeting of the Psychonomic Society, St. Louis, MO. Neill, W.T., Valdes, L. A., & Terry, K. M. (1995). Selective attention and the inhibitory control of cognition. In F. N. Dempster & C. J. Brainerd (Eds.), Interferenceand inhibition in cognition (pp. 207-261). San Diego, CA: Academic Press. Neill, W. T., Valdes, L. A., Terry, K. M., & Gorfein, D. S. (1992). Persistence of negative priming: 11. Evidence for episodic trace retrieval. Journal of Experimental Psychology: Learning, Memory, and Cognition, IS, 993-1000. Neill, W. T., VerWys, C. A., & Neely, J. H. (1997, November). Repetition blindness occurs for both presentations of a repeated letter. Paper presented at the meeting of the Psychonomic Society, Philadelphia, PA. Neill, W. T., & Walling, J. R. (1981). Disruptive effects of prior information on tachistoscopic recognition. Memory & Cognition, 9, 217-224. Neill, W. T., & Westberry, R. L. (1987). Selective attention and the suppression of cognitive noise. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 327-334. Neumann, E., & DeSchepper, B. G. (1992). An inhibition based fan effect: Evidence for an active suppression mechanism in selective attention. Canadian Journal of Experimental Psychology, 46, 1-40. Park, J., & Kanwisher, N. (1994). Negative priming for spatial locations: Identity mismatching, not distractor inhibition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 613-623. Posner, M. I. (1980). Orienting of attention: The Seventh Bartlett Memorial Lecture. Quarterly Journal of Experimental Psychology, 32, 3-25. Posner, M. I., & Cohen, Y. (1980, November). Consequences of visual orienting. Paper presented at the meeting of the Psychonomic Society, St. Louis. Posner, M. I., & Cohen, Y.(1984). Components of visual orienting. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X: Control of language processes (pp. 531556). Hillsdale, NJ: Erlbaum. Posner, M. I., Cohen, Y., Choate, L., Hockey, R., & Maylor, E. A. (1984). Sustained concentration: Passive filtering or active orienting? In S. Komblum & J. Requin (Eds.), Preparatory states and processes (pp. 49-65). Hillsdale, NJ: Erlbaum. Posner, M. I., Davidson, B. J., & Snyder, C. R. R. (1980). Attention and the detection of signals. Journal of Experimental Psychology: General, 109, 160-174. Posner, M. I., & Keele, S. W. (1968). On the genesis of abstract ideas. Journal of Experimental Psychology, 77, 353-363. Posner, M. I., & Keele, S. W. (1970). Retention of abstract ideas. Journal of Experimental Psychology, 83, 304-308. Posner, M. I., & Snyder, C. R. R. (1975). Facilitation and inhibition in the processing of signals. In P. M. A. Rabbitt & S. Dornic (Eds.), Attention and performance V (pp. 669-682). New York: Academic Press.
42
W. Trammel1 Neil1 and Katherine M. Mathis
Postman, L., Stark, K., & Fraser, J. (1968). Temporal changes in interference. Journal of Verbal Learning and Verbal Behavior, 7, 672-694. Pratt, J. (1994). Inhibition of return in a discrimination task. Psychonomic Bulletin & Review, 2, 117-120. Pratt, J., & Abrams, R. A. (in press). Inhibition of return in discrimination tasks. Journal of Experimental Psychology: Human Perception and Performance. Reicher, G. M. (1969). Perceptual recognition as a function of meaningfulness of stimulus material. Journal of Experimental Psychology, 81, 275-270. Richardson-Klavehn, A., & Bjork,R. (1988). Measures of memory. Annual Review of Psychology, 39, 475-543. Roediger, H. L., & Blaxton, T. A. (1987a). Effects of varying modality, surface features, and retention interval on priming in word fragment completion. Memory & Cognition, 15, 379-388. Roediger, H. L., & Blaxton, T. A. (1987b). Retrieval modes produce dissociations in memory for surface information. In D. S. Gorfein & R. R. Hoffman (Eds.), Memory and learning: The Ebbinghaus centennial conference (pp. 349-379). Hillsdale, NJ: Erlbaum. Roediger, H. L., & Neely, J. H. (1982). Retrieval blocks in episodic and semantic memory. Canadian Journal of Psychology, 36, 213-242. Roediger, H. L., Weldon, M. S., & Challis, B. H. (1989). Explaining dissociations between implicit and explicit measures of retention: A processing account. In H. L. Roediger & F. I. M. Craik (Eds.), Varieties of memory and consciousness: Essays in honour of Endel Tulving (pp. 3-41). Hillsdale, NJ: Erlbaum. Scarborough, D. L., Cortese, C., & Scarborough, H. (1977). Frequency and repetition effects in lexical memory. Journal of Experimental Psychology: Human Perception and Performance, 3, 1-17. Schacter, D. L. (1987). Implicit memory: History and current status. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 501-518. Schacter, D. L. (1990). Perceptual representation systems and implicit memory: Toward a resolution of the multiple systems debate. Annals of the New York Academy of Sciences, 608, 543-571. Schacter, D. L., Cooper, L. A., & Delaney, S. M. (1990). Implicit memory for unfamiliar objects depends on access to structural descriptions. Journal of Experimental Psychology: General, I 19, 5-24. Simpson, G. B. (1981). Meaning dominance and semantic context in the processing of lexical ambiguity. Journal of Verbal Learning and Verbal Behavior, 20, 120-136. Simpson, G.B. (1984). Lexical ambiguity and its role in models of word recognition. Psychological Bulletin, 96, 316-340. Simpson, G. B., & Burgess, C. (1985). Activation and selection processes in the recognition of ambiguous words. Journal of Experimental Psychology: Human Perception and Performance, I I , 28-39. Simpson, G. B., & Kellas, G. (1989). Dynamic contextual processes and lexical access. In D. S. Gorfein (Ed.), Resolving semantic ambiguity (pp. 40-56). New York: SpringerVerlag. Slamecka, N. J., & Graf, P. (1989). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4, 592-604. Sloman, S. A., Hayman, C. A., Ohta, N., Law, J., & Tulving, E. (1988). Forgetting in primed fragment completion. Journal of Experimental Psychology: Learning, Memory, and Cognition, 14, 223-239. Smith, E. E., Haviland, S . E., Reder, L. M., Brownell, H., & Adams, N. (1976). When preparation fails: Disruptive effects of prior information on perceptual recognition. Journal of Experimental Psychology: Human Perception and Performance, 2, 151-161.
Transfer-Inappropriate Processing
43
Stoltzfus, E. R., Hasher, L., Zacks, R. T., Ulivi, M. S., & Goldstein, D. (1993). Investigations of inhibition and interference in younger and older adults. Journal of Gerontology, 48, 179-188. Stroop. J. R. (1935). Studies of interference in serial verbal reactions. Journal of Experimental Psychology, IS, 643-662. Tanaka, Y., & Shimojo, H. (in press). Vision, attention, and action-inhibition and facilitation in sensory-motor links revealed by the reaction time and the line motion. Vision Research. Tenpenny, P. L. (1995). Abstractionist versus episodic theories of repetition priming and word identification. Psychonomic Bulletin & Review, 2, 339-363. Terry, K. M.. Valdes, L. A., & Neill, W. T. (1994). Does “inhibition of return” occur in discrimination tasks? Perception & Psychophysics, 55, 279-286. Tipper, S. P. (1985). The negative priming effect: Inhibitory priming by ignored objects. Quarterly Journal of Experimental Psychology, 37(A), 571-590. Tipper, S. P., Brehaut, J. C., & Driver, J. (1990). Selection of moving and static objects for the control of spatially directed action. Journal of Experimental Psychology: Human Perception and Performance, 16, 492-504. Tipper, S. P., & Cranston, M. (1985). Selective attention and priming: Inhibitory and facilitatory effects of ignored primes. Quarterly Journal of Experimental Psychology, 37(A), 581-61 1. Tipper, S. P., & Driver, J. (1988). Negative priming between response modalities: Evidence for semantic processing of ignored stimuli. Memory & Cognition, 16, 64-70. Tipper, S. P., Driver, J., & Weaver, B. (1991). Object-centered inhibition of return of visual attention. Quarterly Journal of Experimental Psychology, 43A, 289-298. Tipper, S. P., & Milliken, B. (1996). Distinguishing between inhibition-based and episodic retrieval-based accounts of negative priming. In A. F. Gamer, M. G . H. Coles, & G . D. Logan (Eds.), Converging operations in the study of visual selective attention (pp. 337-363). Washington, DC: American Psychological Association. Tipper, S. P., Weaver, B., Cameron, S., Brehaut, J. C., & Bastedo, J. (1991). Inhibitory mechanisms of attention in identification tasks: Time-course and disruption. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 681-692. Tipper, S. P., Weaver, B., Jerreat, L. M., & Burak, A. L. (1994). Object-based and environmental-based inhibition of return of visual attention. Journal of Experimental Psychology: Human Perception and Performance, 20, 478-499. Tipper, S. P., Weaver, B., & Milliken, B. (1995). Spatial negative priming without mismatching: Comment on Park and Kanwisher (1994). Journal of Experimental Psychology: Human Perception and Performance, 21, 1220-1229. Treisman, A. M., & DeSchepper, B. (1996). Object tokens, attention, and visual memory. In T. Inui & J. L. McClelland (Eds.), Attention andperformance X V I (pp. 15-46). Cambridge, MA: MIT Press. Tulving, E. (1972). Episodic and semantic memory. In E. Tulving & W. Donaldson (Eds.), Organization of memory (pp. 381-403). New York: Academic Press. Tulving, E. (1983). Elements of episodic memory. Oxford: Oxford University Press. Tulving, E., Schacter, D. L., & Stark, H. (1982). Priming effects in word-fragment completion are independent of recognition memory. Journal of Experimental Psychology: Human Learning and Memory, 8, 336-342. Underwood, B. J. (1945). The effect of successive interpolations on retroactive and proactive inhibition. Psychological Monographs, 59 (Whole No. 273). Underwood, B. J. (1948). Retroactive and proactive inhibition after five and forty-eight hours. Journal of Experimental Psychology, 38, 29-38. Wedin, M. V. (1988). Mind and imagination in Aristotle. New Haven, CT: Yale.
44
W.Trammel1 Neill and Katherine M. Mathis
Whittlesea, B. W. A., Dorken, M. D., & Podrouzek, K. W. (1995). Repeated events in rapid lists: Part 1. Encoding and representation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 1670-1688. Wickelgren, W. A. (1970). Multitrace strength theory. In D. A. Norman (Ed.), Models of human memory (pp. 65-102). New York: Academic Press. Wickelgren, W. A. (1974). Single-trace fragility theory of memory dynamics. Memory & Cognition, 2, 775-780. Wickelgren, W. A., & Norman, D. A. (1966). Strength models and serial position in shortterm recognition memory. Journal of Mathematical Psychology, 3, 316-347. Winnick, W. A., & Daniel, S. A. (1970). Two kinds of response priming in tachistoscopic recognition. Journal of Experimental Psychology, 84, 74-81. Wood, T. J., & Milliken, B. (in press). Negative priming without ignoring. Psychonomic Bulletin & Review. Yee, P. L. (1991). Semantic inhibition of ignored words during a figure classification task. Quarterly Journal of Experimental Psychology, 43A, 127-153.
CUE COMPETITION IN THE ABSENCE OF COMPOUND TRAINING Its Relation to Paradigms of Interference between Outcomes Helena Mature Oskar PineAo
I. Introduction Cue competition is an established phenomenon in many different learning situations such as Pavlovian conditioning, causal learning, categorization, and many others. Research with both humans and animals has shown that organisms tend to discount the potential predictive role of a cue, X, in predicting an outcome, 0, if they know that another cue, A, which occurs in compound with X, is a better predictor of the outcome. In general, whenever two cues occur together, they interact with one another, so that responding to each of them is not only a function of its own pairings with the outcome, but also a function of the degree to which the other cues that occur in compound with it are paired with the same outcome. Instances of these cue competition effects can also be frequently observed in real life. As an example, if we have developed an allergic reaction (outcome) after having taken two different medicines together, A and X, we would probably attribute the allergic reaction to A rather than to X, if we had also taken A alone and developed the reaction as well. One of the best well-known experimental designs on cue competition is Kamin’s (1968,1969) forward blocking, in which the A + 0 pairings occur before the AX + 0 pairings, and this prior knowledge that A predicts 0, THE PSYCHOLOGY OF LEARNING AND MOTIVATION, VOL. 38
45
Copyright 0 1998 by Academic Press. All rights of reproduction in any form reserved. 0079-7421/98 $25.00
46
Helena Matute and Oskar Pineiio
reduces the ability of X to predict that the outcome will occur, as compared to a control group which has not received the A + 0 pairings. Similar results can be observed in backward blocking experiments (e.g., Shanks, 1985), in which it is the subsequent (rather than previous) knowledge that A predicts 0 that causes impaired responding to X (i.e., AX + 0 pairings followed by A + 0 pairings). There are many other designs in which cue competition effects have been observed (e.g., the effect of relative stimulus validity, described by Wagner, Logan, Haberlandt, & Price, 1968), but, to the best of our knowledge, all of them have included compound training of the target cue, X, and the competing cue, A. In consequence, current theories of learning generally assume that compound training is necessary for cue competition to occur. In other words, they predict that competition would not occur if A and X were not trained in compound. In this chapter we will describe a series of recent experiments in which we have tested this assumption. These experiments are problematic for all theories of learning of which we are aware because they show that cue competition can occur, under some circumstances, even when A and X are trained separately. Our basic design consisted of a simplification of the forward and backward blocking designs so that X was never trained in compound with A (Matute & Pineiio, 1998). In our forward condition, subjects received A + 0 pairings followed by X + 0 pairings, instead of the typical forward blocking training consisting on A -+ 0 followed by AX + 0. In the backward condition, subjects received X + 0 followed by A + 0, instead of the typical backward blocking training consisting of AX + 0 followed by A + 0. At test, we observed weak responding to X in the backward condition, as compared to an appropriate control group, but not in the forward condition. That is, the acquisition of a second cue + outcome association (A + 0)disrupted responding to a previously acquired association between a different cue and the same outcome (X + 0),even though the two cues had never received compound training. Because this effect was new and quite surprising, we have conducted a number of experiments in trying to make sense of the observed data. This chapter will summarize what we so far have learned about this effect. We will first briefly describe the findings reported in Matute and Pineiio (1998) and will argue that none of the theories that explain cue competition effects can explain those results. We will then describe several unpublished experiments that we have conducted while trying to seek an explanation for this effect. Perhaps the most interesting finding is that, by borrowing a few ideas from the literature of interference between outcomes (e.g., extinction and counterconditioning, see Bouton, 1993), we were able to make sense of the initial results on competition between elementally trained cues, and to make accurate predictions for the newer experiments. Interestingly,
Cue Competition and Extinction
47
competition between cues seems to be subject to the same manipulations that have been used in the study of interference between outcomes. This finding is important because, although the study of competition between outcomes is currently being integrated within the larger field of interference paradigms, a rather fragmented picture emerges when our textbooks, still today, have to make use of different theories to explain, for example, extinction and cue competition effects (e.g., Matute, in press). The present results suggest that models that explain competition between outcomes can be extended to account for competition between cues. This, we will argue, has the advantage of allowing for integration of two areas of research that are being developed quite independently of each other. 11. Cue Competition in the Absence of Compound Training:
The Basic Effect The dominant explanation of cue competition effects during many years has been provided by traditional associative theories (e.g., Mackintosh, 1975; Pearce & Hall, 1980; Rescorla & Wagner, 1972). The RescorlaWagner model, for example, assumes that learning occurs only as long as the outcome is surprising. In this way, this theory explains traditional (i.e., compounded) forward blocking by assuming that because A is fully predicting the outcome by the time that the AX +. 0 pairings occur, the outcome will not be surprising when X is first presented, and for this reason, the X 0 association is not acquired. However, this theory has found problems in explaining several other well-known phenomena (for a review of successes and failures of the Rescorla-Wagner model see Miller, Barnet, & Grahame, 1995). Quite possibly, one of the greatest problems for traditional associative theories has been the observation of backward blocking, which was first observed with human subjects by Shanks (1985), and has also been recently observed in rats (Miller & Matute, 1996). To recap, in backward blocking, the AX +. 0 pairings are given before the A -+ 0 pairings. Traditional associative theories cannot explain why responding to X should be impaired by the exposure to subsequent A +. 0 trials. The problem is that those theories predict that the X -+ 0 association should be acquired during Stage 1 of backward blocking, and that the further A -+ 0 pairings given during Stage 2 cannot affect the already acquired X +. 0 association. If X is not present during Stage 2, no learning about X can take place. With regard to the experiments that we will describe here, in which the A -+ 0 and X + 0 pairings occur in different stages of the experiments, rather than in compound, the Rescorla-Wagner model predicts an absence of cue competition between A and X: If A and X are not trained in
-
48
Helena Matute and Oskar Pmeiio
compound, the outcome will be surprising when paired with X, regardless of whether or not it has previously or subsequently been paired with A. Nevertheless, according to these theories, the context in which X and A are trained could be regarded as a compounded cue that could produce, if it acquired enough associative strength during Stage 1, forward blocking of X by context in the forward condition (but not in the backward one, which is analogous to backward blocking). Thus, if we were to observe competition only in the forward condition, the experiments could be interpretable in terms of forward blocking by context. A blocking by context effect would also be predicted in the present experiments, although through a very different mechanism, by comparator theories (e.g., Miller & Matzel, 1988; Shanks & Dickinson, 1987). In this case, equal forward and backward blocking by context should be observed. For example, according to Miller and Matzel, traditional forward and backward blocking effects do not represent a failure to acquire the X + 0 association, as postulated by Rescorla and Wagner (1972), but a failure to express it. According to Miller and Matzel (1988), the two cues in a traditional (i.e., compounded) forward or backward blocking experiment acquire an association to the outcome. Later, when X is presented during testing, it activates a representation of the outcome as well as a representation of A (through the X-A within-compound association that is formed during compound training). The two cues then activate their own representation of the outcome. At this point, blocking will be observed if the representation of the outcome that is activated by A is stronger than the representation of the outcome that is activated by X, which is the cue that is actually present. Thus, although this theory predicts an absence of competition between cues that have been elementally trained (because there is no association between them), this theory predicts that, if the context acquires enough associative strength, both forward and backward blocking by context should occur. If this were correct, blocking by context should be observed both in our forward condition and in our backward condition, with no differences between them being predicted by these theories. Several nonassociative (rule-based) models would also predict impaired responding to X in the two groups, regardless of trial order, not because of associative interference, but because of the identical statistical contingency to which the two groups were exposed (e.g., Allan, 1980; Busemeyer, Myung, & McDaniel, 1993; Cheng & Novick, 1992; Rescorla, 1968). Moreover, some recent revisions of rule-based theories (e.g., Cheng, 1997) make the explicit prediction that competition between elementally trained cues should not be observed (see, e.g., Cheng, Park, Yarlas, & Holyoak, 1996, p. 345).
Cue Competition and Extinction
49
The first experiments that we will describe were aimed to test the prediction of all of these traditional theories that competition between elementally trained cues should not be observed.
A. METHODOLOGICAL ASPECTS COMMON TO ALLEXPERIMENTS In all of the experiments to be described in this chapter, our subjects were volunteer college students and the experiments were run in a large computer room that allowed for simultaneous running of all subjects in each experiment. In all of the experiments, except for those in Section V, the preparation that we used was a behavioral task for use with humans that was recently developed by Arcediano, Ortega, & Matute (1996). In this preparation, the experiments are superimposed on a video game in which the task of the subjects is to shoot a laser gun (the space bar) at Martians that are trying to invade the Earth. Martians appear, one at a time, every 0.2 seconds, which gives a behavioral baseline of about 5 responses per second. But subjects are told that, from time to time, the Martians may connect an antilaser shield that is evidenced through a white flashing of the screen and that, if they shoot while the shield is connected, many Martians will suddenly invade. Therefore, the subjects suppress their bar-pressing behavior when the shield is connected. On this baseline game, we superimpose the experiments: Changes in the background color of the screen are used as cues that sometimes signal that the shield (outcome) is about to be connected, and sometimes signal nothing. For example, a change from black (the default color of the screen during the intertrial intervals) to blue may signal that the shield is about to be connected, whereas a change to yellow may signal nothing. During the intertrial intervals and during the presentation of the cues, bar-presses are reinforced with Martians exploding. However, barpresses that occur while the shield is connected are punished with an invasion. The experiments are conducted with no breaks or interruptions between stages (i.e., the last trial of each stage and the first trial of the following stage are separated by a regular intertrial interval in which the subjects are shooting at Martians). The learning curves obtained with this preparation by Arcediano et al. (1996) show that the subjects learn to suppress bar-pressing when the screen color changes to that of the cue that signals the outcome and not otherwise. This suppression of ongoing bar-pressing is our dependent variable and is used as an index of whether the subject is expecting the outcome to occur immediately after the target cue. The suppression ratio is computed as a/ (a + b) where a is the number of bar-presses during the presentation of the target cue (e.g., color blue) and b is the number of bar-presses during
Helena Matute and Oskar Pinelo
50
a period of time identical to that of the presentation of the cue and that immediately precedes its presentation. This preparation has also shown to be reliable in the study of traditional (i.e., compounded) blocking effects with humans (Arcediano, Matute, & Miller, 1997). In the preparation described by Arcediano et al. (1996), Martians appear, one at a time, in horizontal rows. In the experiments reported here, we use this horizontal format when only one context is needed. However, some of the experiments to be described here use two different contexts. In those cases, a vertical presentation of Martians, with Martians appearing one at a time in vertical columns, is used as the second context (Matute & Pinefio, 1998).
B. EXPERIMENT 1: BACKWARD, BUT NOTFORWARD, COMPETITION BETWEEN ELEMENTALLY TRAINED CUES In this experiment we (Matute & Pinefio, 1998) modified the typical forward and backward blocking designs to test which of them, if any, produced the strongest cue competition effect in the absence of compound training. The design is summarized in Table 1. Our forward-competition group received A + 0 training during Stage 1,and X + 0 training, instead of the AX + 0 training typical of more traditional blocking studies, during Stage 2. The test phase consisted of one presentation of X. The forward-control group received identical training and testing on X, but was simply exposed to the experimental context (i.e., shooting at Martians in the video game) during Stage 1, with no cues or outcomes being presented during that stage. The backward-competition and -control group received the equivalent training but the order of the two training phases was reversed. That is, for the backward-competition group the A + 0 pairings occurred during Stage TABLE I DESIGN SUMMARY OF EXPERIMENT 1 Treatment
Group
Stage 2
Stage I Forward-Competition Forward-Control Backward-Competition Backward-Control
A
--$
-
OTS ---f no 0
X + OIC -+ no 0 X+O/C+noO
Test
X + OIC -+ no 0 X + O/C -+ no 0 A + O/B + no 0
-
X X X X
Note. A and X were the critical CSs and were yellow and blue, counterbalanced: B and C were two distractor stimuli that were included to prevent strong stimulus generalization (see text). These stimuli were red and brown, counterbalanced. Presentations of A and X were always followed by the outcome whereas presentations B and C were never followed by the outcome.
Cue Competition and Extinction
51
2, as in traditional backward blocking experiments, but Stage 1 consisted of X + 0 pairings rather than the typical AX + 0 pairings. Finally, the backward-control group was exposed to X + 0 pairings during Stage 1, and to the experimental context (shooting at Martians) during Stage 2, receiving no exposure to any cues or outcomes during this stage. The other cues that are shown in Table 1,cues B and C, are distractor stimuli, which were never paired with the outcome, and which are generally needed in this type of simple computerized human preparations in order to prevent subject from just responding to anything that appears in the screen. A and X were yellow and blue background colors of the screen, counterbalanced; B and C were red and brown, counterbalanced. The results of this experiment are shown in Fig. 1. There was no tendency towards competition in the forward condition. The forward-competition group showed good suppression of responding to X, which means that these subjects were expecting the outcome to occur after X when this stimulus was presented at test. If anything, a (nonsignificant) tendency in the opposite direction was observed, in that suppression to X in the forwardcompetition group tended to be better than that in the forward-control group. Most importantly, a reliable cue competition effect was observed in the backward condition. Backward-control subjects showed good suppression of responding when X was presented at test, but suppression to X was “blocked” in the backward-competition group. That is, subjects in the backward-competitior group responded at test as if they were not expecting the outcome to occur when X was presented. This result is similar to backward blocking except that the competing cue, A, was not presented in
8m
0.4
di
-
Competition Control
T
c
.-08
03
p! n
n
$
02
n ”.r,
Forward
Backward
Group
Fig. 1. Conditioned responding to the target cue, X, during the test stage of Experiment 1. The dependent variable was mean suppression ratio. Thus, a lower value indicates stronger conditioning. Error bars represent standard errors of means. After Matute and Pineiio (1998).
52
Helena Malute and Oskar Pmeiio
compound with X during Stage 1. Thus, this suggests the existence of competition between cues that were elementally trained in the backward condition. Moreover, these results suggest that acquisition of a second cue + outcome association (A + 0) can disrupt responding to a previously acquired association between the same outcome and another cue (X + 0). The present results are problematic for traditional theories of learning because they predict an absence of competition between elementally trained cues. As previously mentioned, traditional associative theories predicted blocking by context in the forward, rather than backward condition (e.g., Rescorla & Wagner, 1972). Comparator theories predicted equal forward and backward blocking by context (e.g., Miller & Matzel, 1988). And rulebased theories predicted an absence of cue competition and similar degrees of responding in both forward- and backward-competition groups (e.g., Cheng, 1997). Instead, the present results suggest that (a) the effect is not due to blocking by context or contingency effects because it only occurred in the backward condition, and (b) the effect seems to be due to competition between the two elementally trained cues, A and X. But before going any further, some less interesting but plausible alternative interpretations need to be discarded. This was the purpose of the next experiment. 2: THEEFFECT Is NOT DUETO C. EXPERIMENT MEMORY OVERLOAD The effect observed in Experiment 1 cannot be attributed to differences in retention intervals because the backward-control group was exposed to the same retention interval as the backward-competition group. Nor can it be attributed to blocking by context because the effect was only observed in the backward condition. Nevertheless, it might well be that instead of cue competition in the absence of compound training, some simpler explanations could still account for the results of Experiment 1. For example, the presentation of any stimulus during Stage 2 may produce a memory overload that might interfere with what was learned during Stage 1. Would A + no 0 trials given during Stage 2 produce the same effect? How about giving the A + 0 trials in a different context? If our effect were due to memory overload, giving A + no 0 trials, or giving A + 0 trials in a second context should produce a detrimental effect similar to that observed in Experiment 1. A related possibility is that our presenting two novel cues during Stage 2 (A plus the distractor cue C) might have also produced some form of memory overload. In this case, eliminating the presentation of novel cues during Stage 2 should attenuate the effect. Experiment 2 was designed to control for these possibilities. Thus, in this experiment we tested whether the effect of backward competition between individually trained cues observed in Experiment l could
Cue Competition and Extinction
53
be due to novelty of cues or to memory overload rather than associative interference. For this reason, we now used two new groups of subjects in addition to the critical backward-competition group that replicated (with the changes mentioned above) that of Experiment 1. The design is summarized in Table 2. One of the new groups (group A + no 0) received A + no 0 training during Stage 2; the other group received A -+ 0 trials, as our critical group, although these trials were given in a second context. This group was called group 1-2-1 because Stage 1 and testing took place in Context 1, and Stage 2 took place in Context 2. The critical group, which replicated the backward-competition group of Experiment 1, is here called group 1-1-1because all phases of the experiment took place in Context 1. The other substantial difference with respect to Experiment 1 was that now, all three groups were exposed to only two cues (A and X) instead of four. This was done in order to reduce the number of stimuli involved, and thus, the possibility of memory overload or of novelty of the cues influencing the results. Thus, in this experiment, all three groups received discrimination training with X -+ 0 and A + no 0 during Stage 1, but received no distractors during Stage 2 (only cue A was presented at that stage). The horizontal and vertical contexts described in Section ILA, counterbalanced, served as Contexts 1 and 2 in this experiment. A and X were blue and yellow background colors, counterbalanced. The results of this experiment are shown in Fig. 2. As can be seen in this figure, our critical group, group 1-1-1, replicated the findings of the backward-competition group in Experiment 1. This occurred despite our having reduced the number of stimuli involved, as well as the novelty of the stimuli presented during Stage 2. Moreover, the results also show that this effect did not occur in the group exposed to A +. no 0 trials during Stage 2 or in the group receiving the A + 0 pairings in a second context. Thus, this experiment replicates our findings that the acquisition of a new
TABLE I1 DESIGNSUMMARY OF EXPERIMENT 2 Group
Treatment Stage 1 ~~
1-1-1 A --* no 0 1-2-1
Test Stage 2
~
(X -+ OIA -+ no 0)1 (X + O/A -+ no O ) , (X -+ O/A -+ no O ) ,
(A O), (A -+ no 0)1 (A -+ 0)2 -+
(X)l (X), (XI1
Nore. A and X were the critical CSs and were yellow and blue, counterbalanced: 1 and 2 refer to Contexts 1 and 2, respectively.
Helena Matute and Oskar Pinefio
54
c 0
’P
g
0.3
0.2
-n r.
1-1-1
A-2
no0
1-2-1
Group Fig. 2. Conditioned responding to the target cue, X, during the test stage of Experiment 2. The dependent variable was mean suppression ratio. Thus, a lower value indicates stronger conditioning. Error bars represent standard errors of means. After Matute and Pineao (1998).
cue + outcome association can disrupt responding to a previously acquired association between another cue and the same outcome, and shows that this effect is not an artifact produced by memory overload, nor novelty of the cues presented during Stage 2. Moreover, these results also suggest that the effect is context specific. We are not aware of previous studies that have tested the possibility of associative competition between elementally trained cues. However, there are some results in the “miscuing effect” literature that are consistent with our observations (Lipp, Siddle, & Dall, 1993; Packer & Siddle, 1989;Siddle, 1985;Siddle, Broekhuizen, & Packer, 1990). For example, Lipp et al. showed that giving just one miscued trial in which the outcome was signaled by a cue that previously had signalled its absence (i.e., one A +0 trial given after an X + 0,A + no 0 discrimination) reduced electrodermal responding to the outcome in a subsequent test stage in which the outcome was preceded by the excitatory cue, X. In those studies, reduced responding at test was observed when the outcome was presented, but not when X was presented. Nevertheless, according to Lipp et al., their observing the effect only in response to the outcome could be due to their use of electrodermal responding as the dependent variable. The present experiment can be regarded as an extension of the experiment by Lipp et al. because it shows that the miscuing effect (a) can also be observed in response to X, (b) can be obtained with a very different preparation (although, unlike Lipp et al., we used more than one miscued trial), and (c) is context-specific. Other experiments in our laboratory have demonstrated a miscuing effect with just one trial in Stage 2 using the behavioral preparation used in the present
Cue Competition and Extinction
55
experiments, and have shown that the miscuing effect is not due to blocking by context (Ortega & Matute, 1997). That is, giving the outcome alone during Stage 2 does not produce the miscuing effect. This last observation is also consistent with the results of Experiment 1, as well as with experiments to be described below, which also show that blocking by context cannot account for the results of competition between elementally trained cues reported herein. Although the present experiment can be regarded as an extention of the miscuing effect, Experiment 1 (as well as the experiments to be reported next) did not use miscued trials but used a different design in which a novel cue, rather than an inhibitory cue, was paired with the outcome during Stage 2. Thus, the effect observed in the present series appears to be more general than that produced by miscued trials. However, the similarities between our findings and the miscuing effect suggest convergent evidence of stimulus competition effects in the absence of compound training. As previously stated, this is contrary to traditional theories of learning. In the next section we investigate the possibility that the X + 0 association was unlearned during the A + 0 trials of Stage 2.
In. The Unlearning Explanation There are some recent revisions of the Rescorla-Wagner model that could potentially explain the results of Experiments 1 and 2 (Dickinson & Burke, 1996; Markman, 1989; Tassoni, 1995; Van Hamme & Wasserman, 1994). These revised versions of the Rescorla-Wagner model were developed to account for backward blocking effects and other related effects that suggested retrospective revaluation of cues after their training had been complete (e.g., weak responding to X observed if AX + 0 training is followed by A + 0 training). According to these models, if a cue is omitted in a trial in which it was expected, the representation of that cue becomes negatively activated. If the outcome occurs, this negative activation produces a loss of associative strength. That is, according to this view, each of the A + 0 trials of Stage 2 in a traditional backward blocking design produces unlearning of X. Therefore, if we interpret X as an absent cue that was expected during the A + 0 pairings in our backward (but not forward) condition, these models would explain our effect through unlearning of X during the A + 0 pairings of Stage 2 in the backward condition. According to Van Hamme and Wasserman (1994), any cue that has acquired some degree of associative strength in a given context (or that has been mentioned in the instructions of the experiment) should be regarded as a “relevant” cue that will be reevaluated in trials that occur in the same
56
Helena Matute and Oskar PineAo
context in which the cue itself is absent (p. 132). This could explain our observation of impaired responding to X when the A + 0 pairings occurred in the same context as the X + 0 pairings, as well as the absence of this effect when the A + 0 pairings occurred in a second context. According to Dickinson and Burke, however, only those cues that have a withincompound association to cues that are present in a given trial are expected, and thus, reevaluated in trials in which they are absent. Therefore a literal interpretation of Dickinson and Burke’s version of the modified RescorlaWagner model would predict no unlearning of the X + 0 association in our experiments, whereas Van Hamme and Wasserman’s version does predict unlearning. Alternatively, we could probably integrate the predictions of these two versions by assuming that the two of them could rest on the association formed between X and the context during Stage 1, rather than between X and A. In this case, both versions would predict unlearning of the X -+ 0 association in our experiments, as long as the A + 0 pairings occur in the same context as the X + 0 pairings. In our next experiment we decided to provide a more direct test for this prediction. 3: COMPETITION OCCURS WHEN THE TARGET CUE A. EXPERIMENT IS TESTEDIN A CONTEXT IN WHICH THE COMPETING CUE IS MORESTRONGLY ACTIVATED THAN THE TARGET CUE
The revised Rescorla-Wagner model (e.g., Dickinson & Burke, 1996; Van Hamme & Wasserman, 1994) could perhaps explain the results of Experiments 1 and 2 through unlearning of the X + 0 association during the A + 0 trials of Stage 2.Another potential interpretation of our Experiments 1 and 2 is that each of the associations (i.e., X + 0 and A + 0)was acquired independently of each other and neither of them was unlearned. Instead, because the A -+ 0 association was presumably more strongly activated by the test context (because the test trial was presented immediately after the last Stage 2 trial, in which A + 0, but not X --f 0 was activated), A competed with X in the activation of the outcome representation during the test stage. The observation of no impairment in responding to X when the A + 0 pairings occurred prior to the X + 0 pairings (forward-competition group), as well as when the context was shifted between Stage 2 and testing (Group 1-2-1),or when A + no 0, rather than A --$ 0 pairings occurred during Stage 2 (Group A + no 0) gives some support to this hypothesis. In Group 1-2-1,the test context no longer resembled the context in which A + 0 had been trained, and the return to the experimental context of Stage 1 for testing could have produced a reactivation of the X + 0 association. Alternatively, the context shift that occurred between Stages 1 and 2 in that group, rather than the context
Cue Competition and Extinction
57
shift occurring between Stage 2 and testing, might have been the critical factor, as predicted by the negative activation models (e.g., Dickinson & Burke, 1996; Van Hamme & Wasserman, 1994). Thus, in this experiment, we tested whether the critical factor was training A -+ 0 in the context in which the X + 0 pairings had occurred, as predicted by the negative activation models, or whether it was testing X in the context in which the A + 0 pairings occurred. A design summary for this experiment is presented in Table 3. Four groups of subjects were exposed to the same contingencies (X -+ O/C -+ no 0 in Stage 1, followed by A -+ 0 in Stage 2, and test on X), and we manipulated the context in which each stage took place for each group. Stage 1 was conducted in Context 1 for all four groups. Stage 2 was conducted in Context 1 for two of the groups (groups 1-1-1and 1-1-2) and in Context 2 for the other two groups (groups 1-2-1 and 1-2-2). Orthogonally, the test stage took place in either Context 1 (groups 1-1-1-and 1-2-1) or Context 2 (groups 1-1-2 and 1-2-2). The horizontal and vertical contexts, counterbalanced, served as Contexts 1 and 2. A and X were blue and yellow, counterbalanced; C was red. If our effect were due to unlearning of X during Stage 2, it should only be observed if the A + 0 pairings occurred in the context in which X was trained (i.e., it should occur in groups 1-1-1and 1-1-2, but not in groups 1-2-1 and 1-2-2). Alternatively, if the effect were due to A being more strongly activated than X during testing, the effect should be observed if X were tested in the context in which the A + 0 pairings occurred, regardless of whether A and X were trained in the same or in different contexts (it should occur in groups 1-11 and 1-2-2 but not in groups 1-1-2 and 1-2-1). TABLE I11 DESIGN SUMMARY FOR EXPERIMENT 3 Group
Treatment ~
1-1-1 1-1-2 1-2-1 1-2-2
~
Test ~~
Stage 1
Stage 2
(X O/C + no O ) , ( X .+ O/C + no O)i (X + O/C + no O ) , (X -+ O/C 4 no O)i
(A .+ O)i (A -+ O ) , (A + O), (A + O)*
--j
(X), (X),
(X), (X),
Note. The four groups received the identical treatment and testing but we manipulated the contexts (1 and 2) in which the different stages occurred. Group names refer to the context in which each of the three stages took place for that group. A and X were yellow and blue, counterbalanced; C was red.
Helena Matute and Oskar Pineiio
58
.s
0.4
2c .-0
V)
!i
03
n
;0 2 .
n -. i
1-1-1
1-1-2
1-2-1
1-2-2
Group Fig. 3. Conditioned responding to the target cue, X, during the test stage of Experiment 3. The dependent variable was mean suppression ratio. Thus, a lower value indicates stronger conditioning. Error bars represent standard errors of means. After Matute and Pineiio (1998).
The results of this experiment can be seen in Fig. 3. Neither the context of Stage 2, nor the test context appeared critical in producing the effect. Instead, the interaction of these two contexts was significant: The effect was observed only in those groups that were tested in the context in which the competing A + 0 association had been trained (i.e., groups 1-1-1and 1-2-2)* Thus, contrary to the predictions of the negative activation models, the A + 0 pairings given during Stage 2 did not produce unlearning of the X + 0 association that had been acquired during Stage 1. Instead, good responding to X was observed after the A + 0 pairings as long as these occurred in a context different from that in which X was tested (groups 12-1 and 1-1-2). Other results that also speak against the unlearning explanation commonly used by associative theories to explain weak responding are spontaneous recovery from extinction (Pavlov, 1927), as well as the observation that good responding after extinction can be obtained through the use of several contextual (or other) manipulations (see e.g., Bouton, 1993).Indeed, it is well known today that extinction does not result, as predicted by traditional associative theories, in the destruction of the X + 0 association (see, e.g., Rescorla, 1996a; Rosas & Bouton, 1996). This “catastrophic interference” prediction, as it is sometimes called, is also known to be a problem of many neural network models (see McCloskey & Cohen, 1989). Thus, how should we interpret these data? It seems clear that competition occurred during retrieval and that the X + 0 association was not unlearned because impaired suppression to X was observed only when the test context
Cue Competition and Extinction
59
and that in which A had been trained, were the same one. A theory such as Miller and Matzel’s (1988), which assumes that cue competition effects occur after, rather than during acquisition, might, at least in principle, be able to explain these results. However, this theory, like most any other theories, has no means by which two elementally trained cues can interact with one another. This theory would in principle explain the results of our Experiments 1and 2 by assuming that backward blocking by context, rather than competition between A and X, had occurred. However, by the same reasoning, this theory would predict forward blocking by context, which is inconsistent with the trial order effects that we observed in Experiment 1. Moreover, Miller and Matzel’s hypothesis actually predicts backward blocking by context in Group 1-1-2 in the present experiment, rather than in our critical group throughout the three experiments (e.g., Group 1-1-l), because in Group 1-1-1both the training context and X are physically present during testing, and thus, no competition between them should be expected (if anything, some summation should be expected in Group 1-11). However, the role played by context in the present research appears to be a modulatory one rather than a competitive one. We know of no one theory of cue competition that can currently explain these effects. In the three experiments described so far, the acquisition of a new cue -+ outcome association impaired responding to a previously acquired association between that outcome and another cue. The only principle that appears clear is that competition occurred whenever the target cue was tested in the context in which the competing cue had been trained. Quite possibly, this occurred because the competing association was more strongly activated than the target one in that context. In the next section we look at the possibility that theories that have been developed to account for extinction and other interference effects that could not be explained by cue competition theories could perhaps be extended to account for the present results.
IV. Theories of Interference between Outcomes Our eliminating compound training from the backward blocking design has resulted in a design surprisingly similar to those generally used in the study of interference between outcomes, such as counterconditioning and extinction (e.g., Bouton, 1993; Bouton & Ricker, 1994; Rescorla, 1996a, 1996b). For example, counterconditioning is another form of retroactive interference in which the acquisition of a second cue + outcome association (X + 0 2 ) also interferes with (but does not destroy) the retrieval of a previously acquired association (X -+ 01). Similarly, extinction is yet an-
60
Helena Matute and Oskar Pineilo
other instance of retroactive interference in which what is learned during Stage 2 (i.e., X + no 0)interferes, but does not destroy, what was learned during Stage 1 (i.e., X + 0).Moreover, weak responding to X after extinction has taken place is often observed if X is tested in the context used for Stage 2, but responding to X can be renewed if X is tested in the context used for Stage 1or in a new context (i.e., renewal effect, see, e.g., Bouton & Bolles, 1979; Rosas & Bouton, 1997). Thus, quite amazingly, extinction (as well as counterconditioning) do not only show that, like in the present experiments, the acquisition of the second cue + outcome association can interfere with a previously acquired one, without destroying it, but also appear to be subject to contextual manipulations very similar to those used in Experiment 3. Both counterconditioning and extinction are generally regarded as something that has very little to do with cue competition. However, the similarities between those studies, that could be regarded as showing competition between outcomes, and the cue competition effect that we have been observing, suggest that similar processes may be involved. The only difference between those studies and our effect is that, in our experiments, the event that changes from one stage to another is the first event in the association (i.e,, the cue) whereas in the studies of competition between outcomes, it is the second event in the association (i.e., the outcome) the one that changes from one stage to the other one. How critical is this difference? It could be argued that this difference is critical because competition between cues has generally been reported in compounded conditions (e.g., blocking) whereas competition between outcomes has generally been reported in noncompounded conditions in which the two outcomes were trained separately (e.g., counterconditioning). However, the present experiments show that compound training is not necessary for the occurrence of competition between cues, just as it is not necessary for the occurrence of competition between outcomes. Moreover, several animal experiments have shown that outcomes may also compete with one another if they are trained in compound, in a way similar to that observed between compounded cues (e.g., blocking of compounded outcomes in animals, see Esmoris-Arranz, Miller, & Matute, 1997; Miller & Matute, in press; Rescorla, 1980). Similarly, several human causal judgment studies have shown competition between multiple compounded effects of one cause (e.g., Matute, Arcediano, & Miller, 1996; Price & Yates, 1993, 1995; Shanks, 1991; Shanks & Lbpez, 1996). Thus, the compound versus elemental training dimension does not seem to be a critical one in differentiating between competition between outcomes and competition between cues. Moreover, if both cues and outcomes can compete regardless of whether they are trained in an elemental or compounded manner, we find no a priori reason
Cue Competition and Extinction
61
to assume that they are governed by different mechanisms, as postulated by current theories. In contrast, we will argue that models of competition between outcomes can be extended to account for competition between cues. It could also be argued that in Pavlovian research with animals, cues (conditioned stimuli, or CSs, such as lights and tones) are known to be processed differently than outcomes (unconditioned stimuli, or USs, such as food or footshock). However, it has been shown that differences in the processing of CSs and USs are not dependent on whether they are cues or outcomes in the associative pair. Instead, as Pavlov (1927) had suggested long ago, differences in biological significance is what appears to differentiate a CS from a US (Gunther, Miller, & Matute, 1997). One more line of research that calls for integration of studies of competition between cues and between outcomes is the old paired associate literature. Indeed, the two paradigms that we have been discussing so far, competition between cues and between outcomes, were extensively studied in the paired associate literature (e.g., Underwood, 1966) and were known as the A-B C-B paradigm (analogous to the effect of competition between individually trained cues described in this chapter) and A-B A-C paradigm (analogous to counterconditioning experiments in which the training with a second outcome interferes with retrieval of a previously acquired association between the same cue and a different outcome). Both the A-B C-B and the A-B A-C paradigms are known to produce interference (see Underwood), and once again, there appear to be no a priori reasons to assume that separate explanations should be developed to account for those two effects. In his review of several interference paradigms in the animal Pavlovian conditioning literature, Bouton (1993; see also Bouton, 1994) noted that the effects that produce either proactive or retroactive interference through pairing of multiple outcomes to a common cue in separate stages (e.g., extinction, counterconditioning, latent inhibition), are generally explained by different mechanisms in the associative learning literature. For example, the unlearning explanation is often used to account for extinction, but other interference effects such as, for example, latent inhibition, are generally explained in a very different way (e.g., Mackintosh, 1975; Pearce & Hall, 1980). He suggested that, by accepting four basic assumptions, many of those interference paradigms in Pavlovian conditioning that involve training with multiple outcomes can be explained in an integrative manner. These assumptions are (a) contextual stimuli guide retrieval, (b) time is context, (c) different memories are differentially dependent on context (i.e., excitatory associations are more readily generalized to new contexts than inhibitory associations), and (d) interference occurs at performance output.
62
Helena Matute end Oskar Pineiio
In Bouton’s view, if a cue becomes associated to two different outcomes in different stages, none of those associations is unlearned, but the cue is not able to retrieve both outcomes simultaneously. In those cases, the cue will predict one or the other outcomes as a function of the context in which retrieval occurs (with the context for each stage being either a physically distinct context or the temporal context in which it occurred). If retrieval occurs in the temporal andlor physical context of Stage 1, the cue predicts Outcome 1; if it occurs in the context of Stage 2, the cue predicts Outcome 2. This would explain, for example, why renewal of responding after extinction is observed if the target association is tested in the context used for acquisition, or in a new one, rather than in that used for extinction. Similarly, spontaneous recovery can be explained using the same principle if we assume that the passage of time removes the subject from the temporal context of cue + no outcome pairings that are controlling extinction performance (Bouton, 1991). Other phenomena, such as reinstatement (Rescorla & Heth, 1975) showing that a mere reexposure to the outcome before testing also produces good responding to an extinguished cue, can also be interpreted in similar terms. In this case, according to Bouton, the cue is returned to a feature of the conditioning context when the outcome is presented before testing. Spontaneous recovery, reinstatement, and renewal effects after extinction has taken place have also been reported in human causal learning situations (Vila, Miranda, Renteria, & Romero, 1997) as well as in other paradigms showing interference between outcomes in animals, such as, for example, counterconditioning and latent inhibition (Bouton & Peck, 1992; Brooks, Hale, Nelson, & Bouton, 1995; Kraemer, Randall, & Carbary, 1991; Peck & Bouton, 1990). Bouton’s (1993) theory predicts competition between independent outcomes during retrieval, but does not predict competition between independent cues. However, we suggest that these assumptions can be extended so as to account, not only for interference effects involving competition between outcomes, but also for those involving competition between cues, such as those described in this chapter. This has the advantage of allowing for a more integrative account of all of the interference paradigms, including cue competition, which, in this way, would no longer need to be regarded as an exception to other interference paradigms. From the above argument, it follows that traditional cue competition studies, such as traditional blocking effects, should be subject to the same manipulations that have been used in the study of competition between outcomes. That is, cue competition should occur during retrieval, and should be sensitive to manipulations of physical and temporal contexts (e.g., spontaneous recovery from cue competition should be observed if the test stage occurs in a temporal test context that is different from that in which
Cue Competition and Extinction
63
competition took place). This is contrary to the dominant explanations of cue competition, which have generally assumed that cue competition effects are due to a failure to acquire the target association (e.g., Rescorla & Wagner, 1972). However, it is consistent with the results of Experiments 1-3 as well as with many other experiments that have shown that traditional (i.e., compounded) cue competition is subject to spontaneous recovery (Cole, Gunther, & Miller, 1997; Kraemer, Lariviere, & Spear, 1988), reinstatement by presentation of the outcome or the cue before testing (Balaz, Gutsin, Cacheiro, & Miller, 1982; Cole, Denniston, & Miller, 1996), and several other postacquisition phenomena that strongly suggest that cue competition takes place during retrieval rather than during acquisition (see Miller & Matzel, 1988, for a review). Moreover, studies on backward blocking also suggest that a failure to acquire the target X + 0 association is not the cause of the observed weak responding to X at test (e.g., Denniston, Miller, & Matute, 1996; Miller & Matute, 1996; Shanks, 1985). Although there are currently several theories of cue competition that can explain the backward blocking effect (e.g., Dickinson & Burke, 1996; Miller & Matzel, 1988), none of those theories can explain spontaneous recovery or reinstatement in cue competition, extinction, and counterconditioning studies. In the next experiments, we tested whether Bouton’s theory, which can explain those effects in situations of interference between outcomes, could be extended to account for competition between elementally trained cues. If this were the case, similar manipulations should yield similar effects in the two paradigms.
4:THEROLEOF TEMPORAL CONTEXTS A. EXPERIMENT The results of Experiments 2 and 3, taken together, suggest that competition in the absence of compound training occurred when X was tested in a physical context in which the A + 0 association was more strongly activated than the X + 0 association. Additionally, these experiments can also be interpreted as demonstrating a “renewal” effect when X was tested in its original (physical) context (group 1-2-1) or in a new context (group 1-1-2) after it had been subject to cue competition training. If this is true, our findings are not too dissimilar to those observed in the literature of interference between outcomes. This suggests that similar theories should be able to explain them. In Experiment 1, however, we did not manipulate the physical context. At first glance, this might suggest that Experiment 1 should be explained through a different mechanism. But using a different mechanism to explain each of the possible ways in which the effect can be obtained does not seem to be the most parsimonious option. Instead, Bouton’s (1993) notion of temporal contexts appears to provide a unified expla-
64
Helena Matute and Oskar Pinelo
nation of what occurred in the three experiments. In Experiment 1, the test of X consisted of one trial that was given immediately after the last A + 0 trial, Therefore, in Experiment 1the effect was observed when the test of X occurred in the temporal context in which A had been trained, just like in Experiments 2 and 3 the effect was observed when X was tested in the physical context of A. But why did subjects create different temporal contexts for each of the training stages in Experiment l? Quite possibly, the presentation of two stimuli, X and C, during Stage 1, and of two different stimuli, A and B, during Stage 2, allowed subjects to discriminate between the two training stages. In Stage 2, the stimuli that were presented had changed, and this may have allowed for the discrimination between the two temporal contexts even though the physical context did not change in that experiment. If this were the case, the same principle that explains Experiments 2 and 3 could explain Experiment 1as well. Namely, competition occurs when the competing cue is more strongly activated than the target cue in the (physical or temporal) context in which retrieval takes place. Now, if this is true, we should be able to manipulate the conditions under which temporal contexts can be created and can affect competition. In the present experiment, our competition group received X + O/C + no 0 discrimination training during Stage 1, and A + 0 training during Stage 2. This manipulation should create, like in Experiment 1,two different temporal contexts coincident with the two stages of the experiment. Group random, by contrast, received random presentations of X + 0, C + no 0, and A + 0 trials, both during Stage 1 and during Stage 2. The number of times that each cue- outcome (or cue -+ no outcome) pair was presented was identical in the two groups. The only manipulation was whether the different types of trials were presented in random or sequential order. In the random group, no differential temporal contexts could be created for each stage because the trial types of each stage did not differ from the trial types of the other stage. Thus, in this group, both A and X should be equally activated in the test temporal context. The competition group, however, should replicate the conditions of Experiment 1 in which two different temporal contexts were presumably created. Because A is presumably more strongly activated than X in the temporal context in which the test of X takes place in this group, responding to X in this group should be impaired, as compared to group random. Therefore, if our hypothesis is correct, competition should only be observed in the competition group. The results of this experiment were as expected. Weak suppression to X at test was observed in group competition, in which different temporal contexts for A and X could be created (M = 0.23; SE = 0.2) as compared to group random (M = 0.16; SE = 0.2), which could not create two different
Cue Competition and Extinction
65
temporal contexts for A and X.This replicates and extends the results of the previous experiments by showing that in this experiment, like in the previous ones, competition occurred only when X was tested in a (temporal or physical) context in which another cue, A, was more strongly activated. Moreover, although the results of Experiment 2 could be interpreted in terms of backward blocking by context, the present results, as well as those of Experiment 1 and 3, show that blocking by context cannot account for this effect. In the event that blocking by context had occurred, it should have also taken place in group random. Instead, the present results suggest that competition between A and X does occur only if they are trained in different (temporal or physical) contexts and X is tested in the context used for the training of A. B. EXPERIMENT 5: A RETRIEVAL CUEFOR THE ACQUISITION THE EFFECT STAGE ATTENUATES One manipulation that Bouton and his colleagues have sometimes used to show the influence of temporal contexts on extinction is the use of a retrieval cue to retrieve one or the other stage of the study just before testing (Brooks & Bouton, 1993). Presumably, if subjects retrieve the information acquired during the Stage 1, they should show behavior appropriate for that stage, whereas if they retrieve the information acquired during the second stage, their behavior should be quite different. For example, Brooks and Bouton (1993) have shown that a retrieval cue for extinction can attenuate spontaneous recovery. Similar manipulations have also been shown to attenuate cue competition in traditional (i.e., compounded) situations by Miller and his colleagues (e.g., Miller & Matzel, 1988). The purpose of this experiment was to test whether a similar manipulation to that used by Brooks and Bouton (1993) could be used to retrieve responding to X in our paradigm. If this were the case, good responding to X should be expected even after exposure to cue competition training with A. Thus, in this experiment our competition group was used as the control group against which to assess retrieval of the X += 0 association in the retrieval group. Our purpose was to test the prediction that a retrieval cue for the temporal context in which X had been trained could renew responding to X. Thus, the two groups were exposed to X + 0, C += no 0 in Stage 1, followed by A += 0, B + no 0 in Stage 2, and tested on X. According to what we have seen so far, impaired responding should be observed. However, just before testing took place in group retrieval, we presented C, which was the nonreinforced cue that had been trained in the same temporal context as X.Thus, presumably, the retrieval, just before testing of the temporal context in which X was trained, should produce a
66
Helena Matute and O s h r Pineilo
renewed responding to X. The control (competition) group, however, received a control cue instead (i.e., this group simply received one more trial with cue B at the end of Stage 2, just before testing). The results showed renewal of suppression to X in the retrieval group (M = 0.20; S E = 0.3) as compared to the competition group (A4 = 0.33; SE = 0.5). Thus, this result adds to the growing evidence reported above that suggests the effect of competition between elementally trained cues only occurs when the target association is tested in a (physical or temporal) context in which the competing association is more strongly activated. Moreover, the similarities between this experiment and that of Brooks and Bouton (1993) suggest that similar processes may be operating here.
V. Competition between Elementally Trained Cues in Judgmental Tasks All of the experiments that we have reported so far had been conducted using the behavioral (Martians) preparation described in Section II.A, in which learning is assessed through a behavioral index (suppression ratio). However, most of the studies that are currently being conducted on human associative learning make use of judgmental tasks that assess learning through the subjective judgment of the degree to which the cue predicts the outcome. Therefore, in order to assess the generality of our effect, we next explored whether the effect could be obtained using a judgmental preparation. The most commonly used judgmental task to study associative learning in humans is the “allergy task” (Wasserman, 1990). In this task, the subjects play the role of allergists who are studying the medical files of fictitious patients. The fictitious patients have taken some foods or medicines (cues) and some of them have developed an allergic reaction (outcome). Each of the fictitious patients represents one trial, and after having observed the records of a number of patients, the subjects are asked to rate the degree to which they think that each of the potential cues is the cause of the outcome. Several variations of this task have been successfully used in many demonstrations of traditional (i.e., compounded) cue competition effects in humans, both in our laboratory and in many others (e.g., Dickinson & Burke, 1996; Matute et al., 1996; Shanks & Lopez, 1996; Van Hamme & Wasserman, 1994). Thus, we adapted our design of competition between elementally trained cues to this preparation. Surprisingly,however, we run into failure after failure to replicate the effect. Below we describe some of the hypothesis that we have entertained while trying to explain these failures, as well as some of the experiments that we have conducted,
Cue Competition and Extinction
67
some of them showing the effect and some of them not showing the effect. We found it important to report those hypothesis as well as the two types of studies because, taken together, these studies show, on the one hand, that the effect is not specific to the Martians preparation, but on the other hand, that specific conditions need to be met in order to obtain the effect in a judgmental task. One difference between the Martians task and the judgment preparation is the reaction time. In a judgment preparation, subjects can take all the time they need before they give their judgments at test. However, in the Martians task, X is presented for a very short time (generally 1 second during training and 3 seconds during testing). Thus, the judgmental response might represent a more elaborated reasoning process in which subjects may take into account all of what they learned during the two training stages, whereas, perhaps in the Martians task, there is no time for elaborated processing. This is a difference that might be important. However, it cannot fully account for all of the data. For example, Escobar and Matute (1998) happened to observe this effect of competition in the absence of compound training in a very different series of experiments, which was looking at differences between differential (i.e., elemental) and Pavlovian (i.e., compounded) inhibition, using a judgmental task quite different from the allergy task but in which the subjects could respond at their own pace (Escobar & Matute, 1998). We will not describe the details of those experiments in this chapter because they were primarily dealing with inhibition rather than cue competition but the important point, for our present purposes, was that they suggested that the effect could be obtained in a judgmental preparation different from the allergy task. The question was why. There were at least two differences between the judgmental task used by Escobar and Matute (1998) and the allergy task that we had been using in our attempts to obtain this effect. First, in the allergy task we were assessing subjects’ judgments at the end of the experiment whereas in Escobar and Matute’s task, the judgments had been assessed in a trial-bytrial basis. This could be important because judgmental preparations have been shown to be sensitive to frequency of judgment manipulations (Catena, Maldonado, & Chdido, in press). Thus, we conducted yet one more experiment with the allergy task using a trial-by-trial assessment of causal judgments. Again, the results showed an absence of competition between elementally trained cues. Thus, at least with our parameters, this variable did not seem to be critical in obtaining competition between elementally trained cues. The other apparent difference between the allergy task and the one used by Escobar and Matute was that the new task involved predictive judgments rather than causal judgments. That is, in each trial, instead of observing
68
Helena Matute and Oskar Pinelo
medical records, subjects had to predict whether an outcome was to appear on the screen based on the information provided by several lights that could be turned on and off. It could well be that predictive, but not causal judgments, were subject to the effect of competition between elementally trained cues. Indeed, the behavioral responses assessed during the Martians task quite probably reflected predictive rather than causal knowledge. To this end we decided to develop yet one more judgmental task designed to focus on predictive rather than causal judgments. Experiments 6 and 7 describe our results with that new judgmental task. 6 AND 7: ROLEOF CONTEXTS A. EXPERIMENTS IN JUDGMENTAL TASKS For these two experiments we used a new judgmental preparation, in which the computer screen presented in each trial one of two colored folders (cues). In each trial, the subjects had to open the folder that was present and see whether it contained a piece of paper (outcome). The design of Experiment 6 was X + 0, A + no 0 in Stage 1, followed by A + 0 in Stage 2. A and X were blue and yellow folders, counterbalanced. At test, subjects were shown folder X and were asked to respond, in a 0 to 9 scale, to which degree they expected the paper to be in that folder. This should have been a direct replication of Experiment 2 with a predictive judgment task. However, the results showed no tendency toward any impairment in the ratings of X at test. That is, subjects were perfectly aware that folder X should contain the piece of paper. They had seen many trials in which that was the case during Stage 1 and they seemed to be unaffected by the A + 0 trials given during Stage 2. Thus, contrary to what we had observed in the experiments reported above, which used the behavioral (Martians) task, and in the experiments reported by Escobar and Matute (1998), which used the lights-outcome judgmental task, the subjects in this experiment were apparently taking into account both the temporal context of Stage 1 and the temporal context of Stage 2 when they gave their ratings of X at test. If our hypothesis that the effect of competition in the absence of compound training could only occur when X was tested in the context of A were correct, perhaps we were failing to obtain the effect in this experiment because we were actually testing X in a whole new context. That is, in the behavioral Martians task (as well as in the judgmental light task used by Escobar & Matute, 1998) the test of X was actually one more trial given at the end of Stage 2, and thus, the test of X occurred in the temporal context of Stage 2. However, in the folders judgmental preparation used in this experiment (as well as in the allergy task that we had used in previous attempts) the test trial had taken place in a whole new context in which the relative attractive graphics
Cue Competition and Extinction
69
mode used during training was substituted for a bland text screen showing instructions and questions. If this were true, the use of differential physical contexts should produce the effect in this task. Thus, in the next experiment, we decided to explicitly manipulate the contexts in which each phase of the folders study took place. In Experiment 7, we again used the judgmental folders task used in Experiment 6, but we introduced two different physical contexts, which were defined by whether the folder belonged to a man or to a woman. If the folder in a particular trial belonged to a man, a picture of a businessman appeared on the screen, next to the folder. If the folder belonged to a woman, a picture of a businesswoman appeared next to the folder. These contexts were counterbalanced. Subjects were warned by the instructions that the folders could contain the paper or not, as a function, not only of their color, but also of whether the owner was the man or the woman. The design of this experiment replicated that of Experiment 6, by using the same cues and trial types, except that we now manipulated the contexts in which each stage took place: Both groups performed Stage 1 in Context 1, and Stage 2 in Context 2. However, group 1-2-1 performed the test stage in Context 1, while group 1-2-2 performed the test stage in Context 2. We expected to obtain a lower judgment to X in group 1-2-2 because the A + 0 association should presumably be more strongly activated than the X + 0 association in the test context for this group, and a greater judgment for X in group 1-2-1, because in this group the X + 0 association should be more strongly activated than the A + 0 association in the test context. As expected, the results showed that the judgment in group 1-2-2 (M = 2.57; SE = 1.44) was significantly lower than that in group 1-2-1 (M = 9; SE = 0). Thus, this showed that competition between elementally trained cues could also be obtained in judgmental situations as long as physical contexts are explicitly manipulated so that it is made explicit that the test of X is taking place in the context in which A, rather than X, was trained. It could be argued that the weak responding observed in group 1-2-2 was due to a generalization decrement because X was tested in a context in which it had never been presented before. This explanation is discarded in the next experiment. Moreover, in order to further explore the generality of the effect, the next experiment makes use of the more traditional allergy task, and of an explicit manipulation of temporal (rather than physical) contexts. 8: MAKING TEMPORAL CONTEXTS EXPLICIT IN THE B. EXPERIMENT ALLERGY TASK The experiments that we have reported so far suggest that competition can be observed whenever the target cue is tested in a (temporal or physical)
70
Helena Metnte and Oskar Pineiio
context in which the competing cue is more strongly activated. Moreover, Experiments 6 and 7 strongly suggest that subjects might not as readily create temporal contexts in judgmental tasks as compared to behavioral tasks, and that, in judgmental tasks, an explicit contextual manipulation might be necessary to obtain this effect. In this experiment we explicitly manipulated the temporal contexts in the allergy task. Given that everything is fictitious in this task, and that it apparently works well with other cue competition designs (e.g., Dickinson & Burke, 1996; Matute et al., 1996; Shanks & Lbpez, 1996; Van Hamme & Wasserman, 1994), we anticipated that fictitious manipulations of time should also be possible in such a setting. To this end, we developed a “time machine” to be used in conjunction with the allergy task. The preparation worked as in our previous attempts, with the subjects seeing the records of fictitious patients, one at a time, on the computer screen, and their judgments of the degree to which each of the medicines were causing the allergic reaction being assessed after all of the training trials had been presented. The main difference with respect to our previous attempts to obtain the effect using this preparation was the explicit manipulation of temporal contexts. Subjects were told that, from time to time, they would be traveling in a time machine, and that, when this occurred, they could either return to the same century or appear in a different one. Thus, they should pay attention, not only to the medicines and allergies that were presented in each trial, but also to the messages in the time machine, which indicated the century they were at. Between Stages 1 and 2, and between Stage 2 and testing, the screen was cleared and the message “Time machine activated . . . Traveling through time” was displayed. Moreover, a message stating “You are in the [Nth] century’’ was always displayed at the top of the screen. In this way, we manipulated three different explicit temporal contexts: 15th, 16th, and 17th centuries, counterbalanced. The design for this experiment is shown in Table 4.Three groups of subjects were exposed to the identical contingencies: Intermixed X + 0, C .--) no 0 trials during Stage 1, followed by A + 0 during Stage 2, and test on X.Also, the three groups were identical in the explicit temporal contexts used for Stage 1 and Stage 2 (i.e., all groups were exposed to Context 1 in Stage 1 and to Context 2 in Stage 2). The critical difference between them was the temporal context in which the test took place: It was Context 1 for Group 1-2-1, Context 2 for Group 1-2-2, and Context 3 for Group 1-2-3. Like in the other allergy task experiments we had conducted, the physical contexts used for training and testing were very different from each other. Training was conducted with a relatively attractive graphic mode using large style fonts showing, in colored screens, the medical files of each patient, their patient number,
Cue Competition and Extinction
71
TABLE IV DESIGN SUMMARY OF EXPERIMENT 8 Group
Treatment Stage 1
1-2-1
1-2-2 1-2-3
( X -+ OIC + no O ) , (X + O/C -+ no O ) , (X + O/C -+ no O ) ,
Test Stage 2 (A + O)Z (X), (A + 0 ) ~ (X)Z (A -+ O)z (X),
Nore. Experiment 8 used a causal judgment preparation. A, X. and C were potential causes of an effect (0).Presentations of X and A were always followed by the effect, whereas presentations of C were never followed by the effect. Numbers 1.2. and 3 symbolize the contexts that we manipulated.
as well as the time machine messages, and several other decorative stimuli (colored borders and so on). By contrast, testing was conducted in blackand-white regular font text screens showing just the instructions on how to respond and the actual test questions. The instructions explained that they should respond by typing a number between 0 (definitely not) and 8 (definitely) to indicate the degree to which they believed that each of the medicines was the cause of the allergic reaction. The test questions used in this study took the form of “You are in the [Nth] century. To which degree is [medicine XI the cause of the allergic reaction? (Please, type a number between 0 and 8 according to the instructions above.)” We expected to obtain a weaker judgmental response to X in Group 12-2 when compared to Group 1-2-1, because of the strong activation of the A -+ 0 association in the temporal context in which the test of X was performed in Group 1-2-2. Group 1-2-3 served to assess the judgment to X in a novel temporal context, in order to know whether the impaired judgment in Group 1-2-2 was due to generalization decrement produced by presenting X in a temporal context in which it had not received training (in which case judgment in Groups 1-2-2 and 1-2-3 should be similar) or whether it was due to genuine cue competition (in which case judgment in Group 1-2-2 should be lower than in Group 1-2-3). Thus, we expected to find higher judgments in Groups 1-2-1 and 1-2-3 when compared to Group 1-2-2. As can be seen in Fig. 4, we obtained exactly those results. This means that competition between individually trained stimuli can occur using a judgmental preparation, not only by manipulating physical contexts but also by manipulating the conditions under which temporal contexts may be formed. This discards the possibility that the effect was due to an artifact of our behavioral task. However, the results of the judgmental studies
72
Helena Matute and Oskar Pineilo
1-2-1
1-2-2
1-2-3
Group
Fig. 4. Mean causal judgment to the target cause, X, during the test stage of Experiment 8. Error bars represent standard errors of means. The lack of an error bar in group 1-2-1 is due to S E = 0 in that group.
described herein also suggest that, in judgmental situations, subjects do not appear to create different temporal contexts for each stage and, thus, the explicit manipulation of temporal or physical contexts seems to be a factor modulating this effect. A more traditional explanation of this experiment can also be constructed for those probably many readers, who might be skeptical about our use of a time machine to manipulate temporal contexts. The sentence “You are in the [Nth] century” inserted in the test question can be regarded as a retrieval cue for a given stage of the experiment. In this case, Groups 1-22 and 1-2-1 are similar to groups competition and retrieval, respectively, of Experiment 5 , although in this case, the retrieval cue used during testing was phsyically different from that used during training. Thus, like in Experiment 5 , the presentation of a retrieval cue for Stage 1 during testing in Group 1-2-1effectively retrieved the context in which the X + 0 association had been trained, and in this way, good responding to X was observed. Note that in this interpretation it is also necessary to assume that temporal contexts (in this case implicit ones) have been created. As previously mentioned, the physical contexts used during training and testing were very different from each other. Thus, it seems that the retrieval of temporal contexts, either through the time machine, or perhaps more conservatively, through the use of a retrieval cue, favored the retrieval of the X --f 0 association after it had been subject to cue competition training. This result is similar to those observed in the literature of interference between outcomes (e.g., Brooks & Bouton, 1993). The results of Group 1-2-3 are also interesting. On the one hand, the comparison between Group 1-2-3and Group 1-2-1shows that the judgment
Cue Competition and Extinction
73
was weaker in Group 1-2-3.This can probably be attributed to some generalization decrement that occurs when X is tested in a new context. The comparison between Group 1-2-3 and Group 1-2-2, shows, on the other hand, that impaired responding to X was greater in Group 1-2-2. This suggests that the effect observed in Group 1-2-2 cannot be attributed to a simple generalization decrement effect. In other words, competition was not observed when the test context was new (and thus, neither A nor X were presumably activated in that context). This confirms the results of previous experiments that suggested that competition only took place when X was tested in the context in which A had been trained. Additionally, this is, once again, quite similar to results in the literature of interference between outcomes, which also suggest that extinction and counterconditioning are observed if the test stage occurs in the test context used for Stage 2, but not if it occurs in the context of Stage 1 or in a new context (e.g., Bouton, 1993). According to Bouton, different memories are differentially dependent on context, with excitatory associations generally predominating over inhibitory ones unless the test context is that in which the inhibitory association was trained. That is, in a new context, the excitatory association should predominate over the inhibitory one. This opens the possibility that an inhibitory association might be formed during Stage 2 in the present experiments. This potential explanation, as well as the relative activation hypothesis that we have been assuming through this chapter, are discussed in the next section.
VI. Extending Models of Competition between Outcomes to Account for Competition between Cues In the previous sections we have summarized what we so far have learned about the effect of cue competition in the absence of compound training. Because these results were new and hard to explain, our primary efforts have focused on obtaining data that could suggest possible avenues to explain this effect. At this point, we know that cue competition is much more similar than we had thought to several effects observed in the literature of interference between outcomes. This calls for integration of the two areas. Given the results reported in this chapter, as well as many other results in the literature (see above), it appears easier to extend theories of interference between outcomes to account for competition between cues than vice versa. However, we are still uncertain as to the best way in which theories that could account for both phenomena should be constructed. Thus, in this section we will only speculate about some possible ways in which models
74
Helena Matute and Oskar Pineilo
of interference between outcomes could be extended to account for cue competition effects. One possibility is that an inhibitory association was formed during Stage 2 of the present experiments, and this prevented responding to X when the test trial was performed in the context used for Stage 2. This inhibitory association might have formed, for example, between A and X (because when A is present, X is absent, see Espinet, Iraola, Bennett, & Mackintosh, 1995). This inhibitory association, if it really exists, appears to be specific to the context in which it is acquired. This is consistent with the contextualspecificity of inhibitory associations observed in the literature of interference between outcomes (e.g., Bouton, 1993, 1994). A related possibility is that the potential inhibitory association might alternatively have formed between X and 0 (or between X and the response, R). However, this possibility requires some additional assumptions. In extinction experiments, an inhibitory association between, say, X and 0, can be formed when we present X and it is not followed by 0 (e.g., Rescorla, 1993). However, in order to extend this idea to the present experiments, we would need to assume that the inhibitory association between X and 0 is formed when we present 0 and X does not precede it. Dickinson and Burke (1996) had proposed a revision of Wagner’s (1981) SOP model that makes this prediction. According to Dickinson and Burke, if the representation of an absent cue is activated (in this case by the withincompound association between X and the context) and the outcome occurs, an inhibitory association is formed between the absent cue, X, and the outcome. Although this view alone would not explain several of the results presented in this chapter (e.g., the observation of an interaction between the test context and the context used for Stage 2; see e.g., Experiment 3), if we add to this Bouton’s assumption that inhibitory associations are specific to the context in which they are learned, a plausible explanation emerges: An inhibitory association is formed between X and 0 during Stage 2 but it does not generalize to other contexts. We currently have no data to favor this veiw over the alternative ones that we are speculating about, but this is one more possibility that should certainly be considered. An alternative interpretation of how such inhibitory association between X and 0 might have formed assumes the existence of bidirectional associations, an idea that is quite foreign to most associative learning theories. However, although we have not yet tested whether this idea can account for the present results, there are some studies that strongly suggest that both animal (Miller & Barnet, 1993) and human subjects (Gerolin & Matute, 1998) can acquire bidirectional associations between events. For example, in the studies reported by Gerolin and Matute (1998), subjects were shown several cards that contained a figure in one side and a color in the
Cue Competition and Extinction
75
other one. After seeing several trials in which the color side was always presented before the figure side, subjects were presented one figure and asked which color was at the other side. Their responses were impressively accurate, as compared to a control group of subjects who had experienced a training phase in which the color and figures had been presented in an inconsistent manner. This does not necessarily imply that a backward inhibitory association between 0 and X was formed in the present experiments, but it does suggest that such possibility should also be considered. Another possibility is the one that we have been implying throughout this chapter. Because the effect appears to be so strongly dependent on the interaction between the (physical or temporal) contexts used for Stage 2 and testing, we have assumed that the activation of the competing association during testing (i.e., A + 0) prevents the activation of the target association (i.e., X + 0).At this point, two new possibilities emerge. First, it might be that the test context activates one of two different associations such as, for example, Context + ( A --+ 0)versus Context + (X + 0).If this were the case, this process might be similar to that described by Bouton (1993, 1994), according to which, if a cue is activating one outcome (e.g., X + Ol), it cannot simultaneously activate another one (e.g., X + 02). The only difference would be that a higher order association, rather than a simple one, would be required to explain our effect. Second, it might also well be that the test context is strongly activating the representation of A, rather than that of X, and thus, A, rather than X, is activating the outcome representation (i.e., context + A + 0 rather than context + X + 0).If we assume that there is a limit in the amount of activation that a given event representation can support (in this case, the representation of 0),and if A is fully activating it, this would make X ineffective in activating the representation of 0.This could explain why X fails to produce a response. However, this last explanation alone does not explain why the representation of A, which is presumably strongly activating the outcome representation when X is presented, fails to produce a response. In order to explain this, we should probably assume that the representation of cues that are present in a given trial (in this case X) differs from the representation of cues that are absent (e.g., Wagner, 1981), and that subjects do not respond to cues that are absent (in this case, cue A) unless they expect them to immediately occur, such as, for example, in sensory preconditioning experiments. At this point, one more possibility should be considered. Namely, that the potential inhibitory association between X and A (see above) is what prevents responding to the representation of the outcome which is activated by A when X is presented during testing. A related explanation has been suggested by Mackintosh and Bennett (1997) to ac-
16
Helena Matute and Oskar Pinefio
count for a different phenomenon (i.e., Espinet et al., 1995), but it also appears plausible in the present case. Finally, if, as the data apparently suggest, cue competition takes place during retrieval, rather than during acquisition, it might also be a good idea to revive the older theories that explained the formation of associations in terms of pure contiguity (Lee,without recourse to cue competition during acquisition). Although this view seems contrary to today’s dominant views, there is an increasing body of data suggesting that contiguity is sufficient for the formation of associations (e.g., Dickinson & Charnock, 1985;Matute, 1994; 1996; Matute et al., 1996; Miller & Bamet, 1993; Miller & Matzel, 1988; Rescorla, 1992).This assumption allows for a more symmetrical interpretation of competition between cues and between outcomes (whether compounded or not), because cues and outcomes would be treated similarly as contiguous (but not competing) associates. In this framework, competition could occur between intact associations for activation of a given event during retrieval, rather than between cues that compete for associative strength. VII. Conclusions Through this chapter, we have shown that cues can compete with each other even when they are not trained in compound. The acquisition of a second cue-outcome association interferes with (but does not destroy) a previously acquired association between the same outcome and a different cue. The experiments also suggest that this occurs whenever the target cue, X, is tested in a (temporal or physical) context in which the competing association (A + 0)is more strongly activated. These results are problematic for all current theories of cue Competition, and strongly suggest that theories of cue competition need to be revised to account for competition in the absence of compound training. An interesting finding has been the observation that competition between cues is subject to the same manipulations that have been used in the study of many other, apparently unrelated, interference paradigms. As an instance, studies of counterconditioning and extinction have also shown that the acquisition of a second cue-outcome association interferes with (though does not destroy) a previously acquired association between the same cue and another outcome, and that, likewise, this too is context-specific, This observation calls for integration of cue competition research into the larger body of literature that studies interference between outcomes (e.g., counterconditioning and extinction) as well as the other interference paradigms. Unfortunately, however, current theories of cue competition do not account
Cue Competition and Extinction
77
for interference between outcomes, and current theories of interference between outcomes do not account for competition between cues. We have suggested several avenues through which theories of interference between outcomes could be extended to explain these results. Although it is clear that a great deal of experimental work still needs to be done before a complete explanation of competition between (compounded and noncompounded) cues and outcomes can be constructed, we should probably assume that associations are acquired as a function of mere contiguity (i.e., events do not compete during acquisition) and are not subject to unlearning. If this is true, both elementally- and compound-trained associations to the same outcome or to the same cue could compete during retrieval, as a function of (a) their relative associative strength, and (b) their relative activation in the (temporal or physical) context in which retrieval takes place. Or, alternatively, it is also possible that some form of inhibition is created during cue competition treatment, and that these inhibitory associations are affected by the same manipulations that have been shown to affect interference between outcomes (e.g., extinction). In any case, the most important point in these studies should be that they allow for a more symmetrical interpretation of studies showing interference between cues and between outcomes, as well as of studies showing interference in compounded and in noncompounded situations. ACKNOWLEDGMENTS The preparation of this chapter was supported by grants PB95-0440 and PI96-006 to HM from Direction General de EnseAanza Superior (Spanish Ministry of Education) and Departamento de Investigacion, Universidades, e Investigation (Basque Government).
REFERENCES Allan, L. G. (1980). A note on measurement of contingency between two binary variables in judgment tasks. Bulletin of rhe Psychonomic Society, 15, 147-149. Arcediano, F., Matute, H., & Miller, R. R. (1997). Blocking of Pavlovian conditioning in humans. Learning & Motivation, 28, 188-199. Arcediano, F., Ortega, N., & Matute, H. (1996). A behavioural preparation for the study of human Pavlovian conditioning. Quarterly Journal of Experimental Psychology, 498, 270-283. Balaz, M. A., Gutsin, P., Cacheiro, H., & Miller, R. R. (1982). Blocking as a retrieval failure: Reactivation of associations to a blocked stimulus. Quarterly Journal of Experimental PSychology, 34B, 99-113. Bouton, M. E. (1991). Context and retrieval in extinction and in other examples of interference in simple associative learning. In L. W. Dachowski & C. F. Flaherty (Eds.), Current topics in animal learning: Brain, emotion, and cognition (pp. 25-53). Hillsdale, NJ: Erlbaum.
78
Helena Matute and Oskar Pineiio
Bouton, M. E. (1993). Context, time, and memory retrieval in the interference paradigms of Pavlovian learning. Psychological Bulletin, 114, 80-99. Bouton, M. E. (1994). Context, ambiguity, and classical conditioning. Current Directions in Psychological Science, 3, 49-53. Bouton, M. E., & Bolles, R. C. (1979). Contextual control of the extinction of conditioned fear. Learning and Motivation. 10,44-466. Bouton, M. E., & Peck, C. A. (1992). Spontaneous recovery in cross-motivational transfer (counterconditioning). Animal Learning & Behavior, 20, 313-321. Bouton, M. E., & Ricker, S. T. (1994). Renewal of extinguished responding in a second context. Animal Learning & Behavior, 22, 317-324. Brooks, D. C., & Bouton, M. E. (1993). A retrieval cue for extinction attenuates spontaneous recovery. Journal of Experimental Psychology: Animal Behavior Processes, 19, 77-89. Brooks, D. C., Hale, B., Nelson, J. B., & Bouton, M. E. (1995). Reinstatement after counterconditioning. Animal Learning & Behavior, 23, 383-390. Busemeyer, J. R.. Myung, I. J., & McDaniel, M. A. (1993). Cue competition effects: Theoretical implications for adaptive network learning models. Psychological Science, 4, 196-202. Catena, A., Maldonado, A,, & CBndido, A. (in press). The effect of the frequency of judgment and the type of trials on covariation learning. Journalof Experimental Psychology: Human Perception and Performance. Cheng, P. W. (1997). From covariation to causation: A causal power theory. Psychological Review, 104, 367-405. Cheng, P. W., & Novick, L. R. (1992). Covariation in natural causal induction. Psychological Review, 99, 365-382. Cheng, P. W., Park, J., Yarlas, A. S.. & Holyoak, K. J. (1996). A causal-power theory of focal sets. In D. R. Shanks, K. J. Holyoak, and D. L. Medin (Ed.), The psychology of learning and motivation, Vol. 34: Causal learning (pp. 133-166). San Diego, CA: Academic Press. Cole, R. P., Denniston, J. C., & Miller, R. R. (1996). Reminder-induced attenuation of the effect of relative stimulus validity. Animal Learning and Behavior, 24, 256-265. Cole, R. P., Gunther, L. M., & Miller, R. R. (1997). Spontaneous recovery from the effect of relative stimulus validity. Learning and Motivation, 28, 1-19. Denniston, J. C., Miller, R. R., & Matute, H. (1996). Biological significance as a determinant of cue competition. Psychological Science, 7, 325-331. Dickinson, A., & Burke, J. (1996). Within-compound associations mediate the retrospective revaluation of causality judgments. Quarterly Journal of Experimental Psychology, 49B, 60-80. Dickinson, A., & Charnock, D. J . (1985). Contingency effects with maintained instrumental reinforcement. Quarterly Journal of Experimental Psychology, 37B, 397-416. Escobar, M., & Matute, H. (1998). Elemental training can sometimes produce stronger cue interaction than compound training. Manuscript submitted for publication. Esmoris-Arranz, F. J., Miller, R. R., & Matute, H. (1997). Blocking of antecedent and suhsequent events. Journal of Experimental Psychology: Animal Behavior Processes, 23, 145-156. Espinet A., Iraola, J. A., Bennett, C. H., & Mackintosh, N. J. (1995). Inhibitory associations between neutral stimuli in flavor-aversion conditioning. Animal Learning & Behavior, 23, 361-368. Gerolin, M., & Matute. H. (1998). Evidence of bidirectional associations. Manuscript submitted for publication. Gunther, L. M., Miller, R. R., & Matute, H. (1997). CSs and USs: What’s the difference? Journal of Experimental Psychology: Animal Behavior Processes, 2.3, 15-30.
Cue Competition and Extinction
79
Kamin, L. J. (1968). “Attention-like” processes in classical conditioning. In M. R. Jones (Ed.), Miami symposium on the prediction of behavior: Aversive stimulation (pp. 9-31). Miami, FL: University of Miami Press. Kamin, L. J. (1969). Predictability, surprise, attention, and conditioning. In B. A. Campbell & R. M. Church (Eds.), Punishment and aversive behavior (pp. 279-296). New York: Appleton. Kraemer, P. J., Lariviere, N. A., & Spear, N. E. (1988). Expression of a taste aversion conditioned with an odor-taste compound: Overshadowing is relatively weak in weanlings and decreases over a retention interval in adults. Animal Learning & Behavior, 16, 164-168. Kraemer, P. J., Randall, C. K., & Carbary, T. J. (1991). Release from latent inhibition with delayed testing. Animal Learning & Behavior, 19, 139-145. Lipp, 0.V., Siddle, D. A. T., & Dall, P. J. (1993). Effects of miscuing on Pavlovian conditioned responding and on probe reaction time. Australian Journal of Psychology, 45, 161-167. Mackintosh, N. J. (1975). A theory of attention: Variations in the associability of stimuli with reinforcement. Psychological Review, 82, 276-298. Mackintosh, N. J., & Bennett, C. H. (1997). Inhibitory associations in taste aversion conditioning. IX Congreso de la Sociedad EspaAola de Psicologia Comparada, Salamanca (Spain). Markman, A. B. (1989). LMS rules and the inverse base-rate effect: Comment on Cluck and Bower (1988). Journal of Experimental Psychology: General, 118, 417-421. Matute, H. (1994). Learned helplessness and superstitious behavior as opposite effects of uncontrollable reinforcement in humans. Learning and Motivation, 25, 216-232. Matute, H. (1996). Illusion of control. Detecting response-outcome independence in analytic but not in naturalistic conditions. Psychological Science, 7, 289-293. Matute, H. (in press). Learning and conditioning. In M. Eysenck (Ed.): Psychology: A European Text. London: Addison Wesley Longman. Matute, H., Arcediano, F., & Miller, R. R. (1996). Test question modulates cue competition between causes and between effects. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 182-196. Matute, H., & Pinefio, 0. (1998). Stimulus competition in the absence of compound conditioning. Animal Learning and Behavior, 26, 3-14. McCloskey, M., & Cohen, N. J. (1989). Catastrophic interference in connectionist networks: The sequential learning problem. In G. H. Bower (Ed.), The psychology of learning and motivation, Vol. 24 (pp. 109-165). San Diego, CA: Academic Press. Miller, R. R., & Barnet, R. C. (1993). The role of time in elementary associations. Current Directions in Psychological Science, 2, 106-111. Miller, R. R., Barnet, R. C., & Grahame, N. J. (1995). Assessment of the Rescorla-Wagner model. Psychological Bulletin, 118, 363-386. Miller, R. R., & Matute, H. (1996). Biological significance in forward and backward blocking: Resolution of a discrepancy between animal conditioning and human causal judgment. Journal of Experimental Psychology: General, 125, 370-386. Miller, R. R., & Matute, H. (in press). Competition between outcomes. Psychological Science. Miller, R. R., & Matzel, L. D. (1988). The comparator hypothesis: A response rule for the expression of associations. In G. H. Bower (Ed.), The psychology of learning and motivation, Vol. 22 (pp. 51-92). San Diego, CA: Academic Press. Ortega, N., & Matute, H. (1997). Competition between individually-trained cues can occur in just one trial. IX Congreso de la Sociedad Espariola de Psicologia Comparada. Salamanca (Spain). Packer, J. S., & Siddle, D. A. T. (1989). Stimulus miscuing, electrodermal activity and the allocation of processing resources. Psychophysiology, 26, 192-200.
80
Helena Matnte and Oskar Pinefio
Pavlov, I. P. (1927). Condiiioned reflexes. London: Clarendon Press. Pearce, J. M., & Hall, G. (1980). A model for Pavlovian learning: Variations in the effectiveness of conditioned but not of unconditioned stimuli. Psychological Review, 87, 532-552. Peck, C. A., & Bouton, M. E.(1990). Context and performance in aversive-to-appetitive and appetitive-to-aversive transfer. Learning and Moiivaiion, 21, 1-31. Price, P. C., & Yates, J. F. (1993). Judgmental overshadowing: Further evidence of cue interaction in contingency judgment. Memory & Cogniiion, 21, 561-572. Price, P. C., & Yates, J. F. (1995). Associative and rule-based accounts of cue interaction in contingency judgment. Journal of Experimenial Pscyhology: Learning, Memory, and Cogniiion, 21, 1639-1655. Rescorla, R. A. (1968). Probability of shock in the presence and absence of CS in fear conditioning, Journal of Comparative and Physiological Psychology, 66, 1-5. Rescorla, R. A. (1980). Pavlovian second-order conditioning. Hillsdale, NJ: Erlbaum. Rescorla, R. A. (1992). Response-independent outcome presentation can leave instrumental R - 0 associations intact. Animal Learning & Behavior, 20, 104-111. Rescorla, R. A. (1993). Preservation of response-outcome associations through extinction. Animal Learning & Behavior, 21, 238-245. Rescorla, R. A. (1996a). Preservation of Pavlovian associations through extinction. Quarterly Journal of Experimental Psychology, 498, 245-258. Rescorla, R. A. (1996b). Spontaneous recovery after training with multiple outcomes. Animal Learning & Behavior, 24, 11-18. Rescorla, R. A., & Heth, C. D. (1975). Reinstatement of fear to an extinguished conditioned stimulus. Journal of Experimenial Psychology: Animal Behavior Processes, I , 88-96. Rescorla, R. A,, & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectivenessof reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical conditioning 11: Current research and theory (pp. 64-99). New York: Appleton. Rosas, J. M., & Bouton, M E. (1996). Spontaneous recovery after extinction of a conditioned taste aversion. Animal Learning & Behavior, 24, 341-348. Rosas, J. M., & Bouton, M. E. (1997). Renewal of a conditioned taste aversion upon return to the conditioning context after extinction in another one. Learning and Moiivaiion, 28, 216-229. Shanks, D. R. (1985). Forward and backward blocking in human contingency judgment. Quarierly Journal of Experimenial Psychology, 378, 1-21. Shanks, D. R. (1991). Categorization by a connectionist network. Journal of Experimental Psychology: Learning, Memory and Cogniiion, 17, 433-443. Shanks, D. R., & Dickinson, A. (1987). Associative accounts of causality judgment. In G. H. Bower (Ed.), The psychology of learning and motivation, Vol. 21 (pp. 229-261). San Diego, CA: Academic Press. Shanks, D. R., & Lbpez, F. J. (1996). Causal order does not affect cue selection in human associative learning. Memory & Cogniiion, 24, 511-522. Siddle, D. A. T. (1985). Effects of stimulus omission and stimulus change on dishabituation of the skin conductance response. Journal of Experimental Psychology: Learning, Memory and Cogniiion, 11, 206-216. Siddle, D. A. T., Broekhuizen, D., & Packer, J. S. (1990).Stimulus miscuing and dishabituation: Electrodermal activity and resources allocation. Biological Psychology, 31, 229-243. Tassoni, C. J. (1995). The least mean squares network with information coding: A model of cue learning. Journal of Experimental Psychology: Learning, Memory, and Cogniiion, 21, 193-204.
Cue Competition and Extinction
81
Underwood, B. J. (1966). Experimental psychology (2nd ed.). New York: AppletonCentury-Crofts. Van Hamme, L. J., & Wasserman, E. A. (1994). Cue competition in causality judgments: The role of nonpresentation of compound stimulus elements. Learning and Motivation, 25, 127-151. Vila, J., Miranda, F., Renteria, A., & Romero, M. (1997). Extinction and causal judgment: Spontaneous recovery and renewal. IX Congreso de la Sociedad Espaiiola de Psicologia Comparada. Salamanca (Spain). Wagner, A. R. (1981). SOP A model of automatic memory processing in animal behavior. In N. E. Spear & R. R. Miller (Eds.), Information processing in animals: Memory mechanisms (pp. 5-47). Hillsdale, N J Erlbaum. Wagner, A. R., Logan, F. A., Haberlandt, K., & Price, T. (1968). Stimulus selection and a “modified continuity theory.” Journal of Experimental Psychology, 76, 171-180. Wasserman, E. A. (1990). Attribution of causality to common and distinctive elements of compound stimuli. Psychological Science, 1, 298-302.
This Page Intentionally Left Blank
SOONER OR LATER The Psychology of Intertemporal Choice Gretchen B. Chapman
I. Introduction Exercising, studying for an exam, and investing for retirement are everyday activities that involve a trade-off between short-term and long-term consequences. Exercise, for example, requires a time commitment and physical discomfort right now (and on each day that one exercises), whereas many of the benefits, such as being in shape or improving health, are delayed. Similarly, studying involves an up-front investment of time for a delayed reward of a good grade and associated events such as graduation or job offers. Thus, intertemporal choice occurs in many situations. Discounted utility theory provides the normative standard for how decision makers should trade off immediate and delayed outcomes. According to this theory, delayed outcomes should be discounted. Thus, the objective value of a delayed outcome must exceed that of an immediate outcome, in order for the subjective value of the two to be the same. The discount rate is the percent increase in magnitude needed to offset a given delay. For example, if money were discounted at an annual rate of lo%, that would mean that $1000 received immediately would have the same discounted utility as $1100 to be received one year from now. That is, $1000 is the present value of the delayed $1100. Present value (vo) is expressed as THE PSYCHOLOGY OF LEARNING AND MOTIVATION. VOL. 3H
83
Copyright Q 1998 by Academic Press. All rights of reproduction in any form reSeNed. 0079-7421198 1625.00
84
Gretchen B. Chapman
vo
=
vd (1 + r)d
where v d is the amount of the future money, d is the delay in years, and r is the annual discount rate (10% in this example). A positive discount rate, or a positive time preference, indicates that an immediate monetary gain is preferable to the same amount of money later. A negative time preference or discount rate indicates that a delayed positive outcome is preferable to an immediate one. According to normative discounted utility theory (Fishurn & Rubenstein, 1982; Loewenstein & Prelec, 1992; Weinstein & Stason, 1977), the discount rate should almost always be positive, and the same discount rate should be applied to all decisions. One rationale for why future outcomes should be discounted at a positive rate is that many goods grow or increase with time. Chapman and Elstein (1995) offer the example that, if offered a bushel of corn now or a bushel of corn a year from now, one should choose the bushel now because the corn could be planted, yielding more than a bushel of corn a year from now. To be just as attractive as a bushel now, the corn offered later would have to total more than a bushel. A second rationale for discounting is that delay involves risks, which compound over time. If one is offered $1,000 one year from now, there is a risk that one could not enjoy the money when it was delivered, for example because of death. Thus a smaller amount of money available immediately would be just as attractive as the delayed $1OOO. The reason that discounted utility theory prescribes that all decisions should employ the same discount rate is that using different discount rates leads to inconsistent preferences (e.g., Keeler & Cretin, 1983). Chapman and Elstein (1995) give the following example: suppose someone discounted money at 10% and corn at 20% and that corn was valued at $10 per bushel. If given a choice between $1,000 and 120 bushels of corn, both delayed 10 years, one would prefer the money (present value = $385) over the corn (present value = 19.38 bushels, worth $193.80). Ten years later when the goods are actually delivered, one would prefer the corn (now worth $1200) over the $1000, constituting a preference reversal. The use of the same discount rate for both money and corn prevents this preference reversal. Thus, the same discount rate should be applied to all domains, including, for example, health. Unlike money, health cannot be invested or saved for future consumption. Consequently, one might argue that discounting of health should be small or even zero. However, when health can be traded for money or other goods, normative theory prescribes that it should be discounted at the same rate as money. (The questions of whether health
Intertemporal Choice
85
is generally perceived as tradeable for money is discussed later in Section 1II.E.) Keeler and Cretin (1983) point out that if health benefits were discounted at a lower rate than monetary costs, then any program that spends money now to improve present health would be delayed indefinitely. Delaying the program would reduce the present value of the monetary costs more so than the present value of the health benefits. Thus, the health units gained per dollar spent would increase as the program is delayed. In order to prevent this paradox, health and money must be discounted at the same rate. In summary, normative theory prescribes that discount rates should be almost always positive. Although it does not specify exactly how large the rate should be, generally small positive rates (e.g., 5%) are used in decision analyses. More importantly, the theory specifies that a decision maker should use the same discount rate for all decisions.
II. Violations of Normative Theory The field of behavioral decision making is rife with examples of how actual behavior violates normative theory, and intertemporal choice is no exception (Loewenstein & Thaler, 1989). Rather than using a single discount rate for all decisions, decision makers often adjust their rates according to a number of nonnormative factors. RATES A. HIGHAND VARIABLE One result from discounting research is that people demonstrate discount rates that are extremely high, often on the order of 50% annually or more (e.g., Benzion, Rapoport, & Yagil, 1989; Chapman, 1996b; Chapman & Elstein, 1995; Thaler, 1981). In a study showing similar results, I asked 82 community members from the Chicago area YMCA and Park Service classes to fill out a questionnaire. The respondents were 63% female, 68% African American, and had a mean age of 36 (range 19 to 76). One survey question asked them to imagine that they had won some money and could receive $10 a day for 1 year starting today, or $10 per day for a longer period of time starting one year from now. They were asked to specify how long the $10 per day would need to continue in the delayed option to make it just as attractive as the option that commenced immediately. The geometric mean response was 4.86 years, corresponding to a discount rate of 386%. (See Table I.) A similarly high discount rate was found when respondents were asked to imagine that they were in poor health and could have a treatment that restored full health for a year beginning immediately
Gretchen B. Chapman
86
TABLE I GEOMETRIC MEANDISCOUNT RATESFROM A STUDY OF CHICAGO-AREA COMMUNITY MEMBERS (n = 82)' Money questions ($10 per day for M years)
Health questions (Full health for M years)
Long delay (7 years)
Short delay
(1 year)
(1 year)
Long delay (7 years)
Low magnitude
3.86 (1.25)
0.66 (0.16)
4.21 (1.36)
0.71 (0.18)
(M = 1) High magnitude (M = 5)
3.07 (0.77)
0.50 (0.13)
3.29 (0.89)
0.52 (0.15)
Short delay ~~~
Nore: The short delay, low magnitude money question asked participants to fill in the blank to make the two prizes equally attractive. Prize A receive $10 per day for 1 year (M) starting today Prize B: receive $10 per day for -years starting one year from now (D) In the long delay question the delay (D) to prize B was 7 years. In the high magnitude questions, prize A involved receiving $10 per day for 5 years (M).The health questions offered full health for 1 (or 5) years starting today or for a longer duration starting 1 (or 7) years from now. * 70 participants provided complete data for the ANOVA upon which this table is based. The ANOVA showed a main effect of magnitude. F(1,63) = 165, MSE = 0.19, p < .OOO1, and of delay, F(1.63) = 898, MSE = 1.22, p < .O001.
or a treatment that gave full health for more than a year but only after a one year delay. The geometric mean discount rate on that question was 421%. Extremely high discount rates are not ubiquitous. Other research (e.g., Redelmeier & Heller, 1993) has found lower average discount rates, but a high degree of variability, with some subjects showing zero or negative discounting. High discount rates are not technically a violation of discounted utility theory, but they do go against the discount rates usually employed in decision analyses. B. DELAY EFFECT Discount rates tend to be higher for short delays than for long delays (e.g., Benzion et al., 1989; Chapman, 1996b; Chapman & Elstein, 1995; Thaler, 1981). For example, in the study described above, the community members made monetary choices that involved a one year delay and a seven-year delay. Whereas the one year delay yielded a discount rate of 386% (as reported above), the seven-year delay question yielded a geometric mean annual discount rate of only 66%. (This corresponds to an indifference between $10 per day for one year starting immediately and $10 per day
Intertemporal Choice
87
for 34.6 years starting seven years from now.) The health choices yielded a similar delay effect (discount rates of 421% versus 71% for the one-year and seven-year delays, respectively.) The delay effect can result in preference reversals (Kirby & Hermstein, 1995). For example, someone may prefer $200 eight years from now to $100 six years from now, but six years later the same person would prefer $100 right away to $200 two years from now. Note that both questions involve whether one wishes to delay payment for two years in order to receive double the money, but the second decision takes place six years after the first decision. Because both delays are longer in the first question, the subjective discount rate is lower and the person is willing to wait for the larger reward. As another example, consider a person who resolves in the evening to get up early the next morning to reap the long-term reward of a good day’s work. The next morning, however, one’s preference switches to the short-term rewards of sleeping in. Christensen-Szalanski (1984) demonstrated such a preference reversal in women deciding whether to have anesthesia for childbirth. One month before labor and one month postpartum the women preferred to avoid using anesthesia during labor. During active labor, their preferences suddenly shifted toward using anesthesia to avoid pain. The anesthesia offers immediate pain relief but also the risk of long-term side effects. At short delays (during labor), the women showed a stronger preference for the immediate relief. This demonstration indicates that time preferences measured at any one time may not be representative of long-term preferences. These preference reversals can be modeled with a particular discount functional form, specifically a hyperbolic discount function rather than the exponential function specified by normative theory (Kirby & Marakovic, 1995). Unlike exponential discount functions, hyperbolic functions for a large and small reward can intersect, indicating a preference reversal. Ainslie (1975; 1982; 1984; 1986; 1991, see also Strotz, 1956) suggests that in order to solve the self-control problem introduced by these preference reversals or dynamic inconsistencies,decision makers must bind themselves to their earlier preferences to prevent a preference reversal from occurring during the “window of temptation.” One method is to develop a personal rule so that each decision sets a precedent for all succeeding decisions. For example, someone might follow a personal rule never to snack between meals and view any violation of this rule as an indication that one will always choose the short-term rewards of snacking over the long-term rewards of thinness and health. This method is equivalent to integrating the hyperbolic discount functions of all the small, more immediate outcomes and integrating the discount functions of all the large, more delayed outcomes. This
88
Gretchen B. Chapman
summation averages discount rates over many different delays and brings the discount rate toward a constant value, reducing preference reversals. Prelec and Loewenstein (1991) explain the delay effect in a manner similar to hyperbolic discounting. They propose the property of decreasing absolute sensitivity for time, which means that the difference between zero and two years seems larger than the difference between six and eight years. Thus, in the choice between $100 in six years versus $200 in eight years, the additional two years required for the larger prize does not seem very large, and consequently the discounting of the $200 is not as extreme as in the choice between $100 now and $200 in two years. C. MAGNITUDE EFFECT Subjective discount rates vary not only with the delay to the outcome, but also with the magnitude of the outcome. Smaller outcomes tend to elicit larger discount rates (Benzion et al., 1989; Chapman, 1996b; Chapman & Elstein, 1995; Chapman & Winquist, in press; Green, Myerson, & McFadden, 1997; Kirby & Marakovic, 1996; Thaler, 1981). Thus, someone may prefer $10 now to $15 in one year, but also prefer $1500 in one year to $1000 now, even though both choices offer a 50% return for waiting one year. The community sample previously described also showed the magnitude effect. When the magnitude of the outcome was small ($10 per day for one year), the mean subjective discount rate was 386% (for a one year delay). In contrast, when the magnitude was large ($10 per day for five years), the discount rate was 307%. The analogous rates for the health questions were 421% versus 329%. (See Table I.) Loewenstein and Prelec (1992; Prelec & Loewenstein, 1991) explained the magnitude effect by positing a value function in which the ratio between the values of $500 and $1000 is larger than the ratio of the values of $5 and $10. Thus, the added 50% seems like a larger ratio when choosing between $lo00 now versus $1500 in one year than in the choice between $10 now and $15 in one year. Chapman and Winquist (in press) found an analogous magnitude effect in decisions about tipping at a restaurant. Subjects indicated they would give a larger percent tip on a smaller restaurant bill. The fact that a magnitude effect occurs in settings other than intertemporal choice suggests that, consistent with Loewenstein and Prelec’s account, it is a result of the shape of the value function for money (or other outcomes).
D. SIGNEFFECT Loewenstein (1988) and others (Benzion et al., 1989; Chapman 1996b; MacKeigan et al., 1993) have demonstrated that discount rates for losses
Intertemporal Choice
89
are lower than discount rates for gains. A decision can often be framed as either a choice between two gains or as a choice between two losses. The sign effect indicates that a lower discount rate will be applied to the decision if it is framed as a choice between losses. Thus, for example, someone may view a gain of $10 right away to be equal in value to a $20 gain in one year (a 100% discount rate) but also see a loss of $10 right now to be equal in value to a $15 loss in one year (a 50% discount rate). Like the magnitude effect, the sign effect can be explained in terms of the value function for money (or whatever outcome is being delayed). Loewenstein and Prelec (1992; Prelec & Loewenstein, 1991) proposed the property of loss amplification; that is, the ratio of the subjective values of gains is smaller than the ratio of values of equivalent losses. Thus, the ratio between $20 and $10 seems larger if the two outcomes are losses than if they are gains. Shelley (1993; 1994) found that the sign effect is dependent on the frame of the question. For example, in a delay frame, the decision maker views the immediate outcome as the reference point. Delaying a gain then results in an immediate loss but a future gain. Similarly, delaying a loss results in an immediate gain but a future loss. When questions were posed in this delay frame, Shelley found that discount rates were higher for gains than for losses. That is, decision makers were reluctant to delay a gain, but not very eager to delay a loss. Conversely, in a speedup frame, decision makers view the delayed outcome as a reference point, such that expediting a gain leads to an immediate gain and a delayed loss, whereas speeding up a loss leads to an immediate loss but a delayed gain. Questions posed in this frame led to a higher discount rates for losses than for gains. That is, decision makers were reluctant to expedite a loss but not very eager to speed up a gain. Loewenstein (1988) also found a difference between delay and speed up questions. E. SEQUENCE EFFECT Preferences for sequences of outcomes often differ from choices between individual outcomes. For example, Loewenstein and Prelec (1991) showed that 80% of their subjects preferred dinner at a French restaurant in one month to dinner at a French restaurant in two months; that is, they made a present-oriented choice. When the outcomes were imbedded in a series, however, subjects were not so impulsive. Although all preferred French food to Greek food, 57% preferred dinner at a Greek restaurant in one month and dinner at a French restaurant in two months over the reverse order. Subjects no longer preferred to have the attractive French meal as soon as possible. Instead, they showed a slight preference for an improving sequence with the better outcome delayed until the end of the sequence.
90
Gretchen B. Chapman
Choices between individual outcomes almost always display a positive time preference, that is, a preference for good events sooner rather than later (e.g., Benzion et al., 1989;Chapman, 1996b;Chapman & Elstein, 1995; Thaler, 1981). In contrast, when presented with a sequence, subjects often demonstrate a negative time preference, preferring to postpone the more attractive events (Kahneman, Fredrickson, Schreibner, & Redelmeier, 1993; Loewenstein, 1987;Loewenstein & Prelec, 1993;Loewenstein & Sicherman, 1991; Varey & Kahneman, 1992). One explanation for this preference for improving sequences (Loewenstein & Sicherman, 1991) is that decision makers anticipate adapting to their current position in the sequence and, because of loss aversion, are averse to decreases in their position. An improving sequence offers a series of gains, which is seen as more attractive than a declining sequence, which offers a series of losses, even if both sequences include the same total amount of money (or restaurant meals). Chapman (1996a) found that preferences for sequences were influenced by expectations about how sequences are usually experienced. Subjects rated their preferences for sequences that described how their health or monetary income could change over their entire lifetime. Whereas they preferred increasing monetary sequences, they preferred declining health sequences, and these preferences were in line with their stated expectations about how health and money would change over their lifespan. When evaluating short sequences (one year or 12 days), expectations about changes in health and money were similar to one another, and improving sequences were preferred in both domains. Unlike the delay, magnitude and sign effects that have all been demonstrated in both health and monetary decisions, the sequence effect is influenced by the decision domain. A preference for improvement has been demonstrated in the vast majority of studies using monetary sequences, whereas, in contrast, decision makers sometimes prefer declining health sequences if such sequences are in line with their expectations. These findings suggest that preferences for sequences may be governed by mechanisms, such as expectations, that do not play such an important role in other intertemporal choice phenomena.
F. SPREADING EFFECT Loewenstein and Prelec (1993) found that, in addition to a preference for improving sequences, decision makers also prefer outcomes that are spread evenly across the sequence interval. They found that 84% of subjects preferred to have a fancy French restaurant dinner on the second rather than the first of three weekends when it was specified that they would eat at home on the remaining weekends. However, when it was specified that
Intertemporal Choice
91
they would eat an exquisite lobster dinner on the third weekend, 54% of subjects preferred to have the fancy French dinner on the first rather than the second weekend. That is, the preference for having the French dinner sooner versus later was influenced by the event occurring on the third weekend. When a lobster dinner was scheduled for the third weekend, subjects were more likely to schedule the French dinner for the first weekend, so that the positive events would be spread across the three-week interval. This pattern violates the independence or separability axiom. Because the third-weekend event is held constant for the two choice options, ones preference for events in the first two weekends should be independent of what event is scheduled for the third weekend. I replicated this spreading effect in some additional scenarios that involved both gain and loss outcomes. Six scenarios were presented to 148 college students. Two scenarios were about money (either winning a prize or paying a fine), two were about health-related events (a painful trip to the dentist or a pain-relieving trip to the chiropractor), and two were about restaurant meals (a pleasant meal with one’s spouse or guest, or an unpleasant meal with an irritating coworker.) Subjects responded to three versions of each scenario (for a total of 18 questions.) In the first version, the event was to occur on either the second or third weekend of a fourweek interval. In this case, subjects generally preferred to have the event (either gain or loss) on the second weekend. In the second version of each scenario, two events were scheduled (e.g., two dentist visits, or two restaurant meals), one on the first weekend, and the other on either the second or third. Subjects had to choose the option (first and second or first and third) that they preferred. Finally, in the third scenario version, subjects choose between having two events scheduled on the first and fourth or second and fourth weekends. Notice that in all three versions, the event (or nonevent) scheduled for the first and fourth weekends were the same for the two choice options. The two options differed in whether there was an event scheduled for the second or third weekend. The three versions differed in terms of what was scheduled for the first and fourth weekends (for both options). The results (shown in Table 11) are very similar to Loewenstein and Prelec’s (1993) results. In the pleasant dinner scenario, for example, 42% (that is, about half) preferred the dinner on the second rather than third weekend (center column) when no other dinners were scheduled. When another dinner was scheduled for the first weekend for both options, only 8% of subjects (left column) preferred a dinner on the second weekend; most preferred it on the third weekend, so that the two dinners would be spread more evenly across the interval. Likewise, when another dinner was scheduled for the fourth weekend, 91% of subjects (right column) preferred
Gretchen B. Chapman
92
TABLE I1 PERCENTAGE OF SUBJECTS (N = 148) CHOOSING THE MORE IMMEDIATE EVENT IN EACHOF THREE VERSIONS OF SIX SCENARIOS Scenario
% choosing 1100 over 1010
% choosing 0100 over 0010
% choosing 0101 over 0011
Pleasant dinner Unpleasant dinner Chiropractor Dentist Monetary prize Monetary fine
8* 30* 12* 19* 28* 30*
42 76 68 85 72 64
91* 64 84* 650 72 83*
Nore: 1100 refers to an event (e.g., a pleasant dinner) on the first and second weekends but not the third and fourth weekends. Other sequences are similarly defined. * Percentage differs significantly from center column, p < .001 8 Percentage differs significantly from center column but in the direction counter to prediction, p < .Gal
the first dinner to be scheduled for the second weekend, so that the two dinners were evenly spread across the interval. Of twelve comparisons made (both the left and right columns compared to the center column), nine showed differences in the predicted direction, and one in the counterpredicted direction. A preference for spreading occurrred for both gains and losses and for restaurant dinners, health events, and monetary outcomes. As in the original Loewenstein and Prelec (1993) demonstration, the preference for spreading resulted in a violation of the independence axiom. Although for any particular choice pair the event scheduled for the first and fourth weekends were the same, subjects’preferences were nonetheless influenced by which event was specified for those weekends. G. ANTICIPATION EFFECT The sequence and spreading effects illustrate that decision makers do not always show positive time preferences. Another case in which delay actually increases the value of an event is the anticipation effect (Elster & Loewenstein, 1992). Loewenstein (1987) demonstrated that people sometimes prefer to put off good outcomes (e.g., a kiss from a movie star) to savor the event, or to speed up bad outcomes (e.g., an electric shock) to avoid dread. Anticipating the event before it occurs makes a positive event (such as a kiss) more enjoyable and a negative event (e.g., a shock) more unpleasant. Dread may explain why, in the spreading effect study described above, a majority of subjects preferred to have the unpleasant dinner, the dentist appointment, and the monetary fine on the second rather than the third
Intertemporal Choice
93
weekend (0100 preferred to 0010). This response pattern is consistent with a desire to “get it over with.” Little is known, however, about when savoring and dread have a large influence on decisions. H. GENERALITY OF INTERTEMPORAL BIASES
Most of the biases described above have been demonstrated in a number of different domains, for example both monetary and health deicisions (e.g., Chapman, 1996b). Few studies assess, however, whether these biases occur outside of laboratory questionnaire experiments. As one step toward answering this question, my survey of Chicago community members included some realistic decision scenarios that involved intertemporal choice. For example, in one scenario the decision maker had to decide whether to give up a habit of eating salted pretzels in order to lower blood pressure, thus decreasing risk of heart disease and increasing life expectancy. This decision involves an up-front cost (forgoing the salted pretzels) for a longterm gain (increased life expectancy). Respondents were asked to divide 100 points between the two choice options to express their preference. Each decision scenario was framed in one of two ways. For example, the pretzel scenario was described either as a choice between two gains or between two losses. In the gain format, decision makers chose between enjoying salted pretzels and gaining a year of life expectancy. In the loss format they chose between giving up the salted pretzels or losing one year of life expectancy. Because the research reviewed above indicates that discount rates are generally lower for losses than for gains, we would expect the long-term option (giving up the salted pretzels and gaining life expectancy) to be rated higher in the loss frame than in the gain frame. Each participant judged eight scenarios about health-related decisions. Two were framed as either a gain or a loss (corresponding to the sign effect), two as short or long delays (corresponding to the delay effect), two as large or small outcomes (corresponding to the magnitude effect), and two as individual outcomes or sequences of outcomes (corresponding to the sequence effect). Each participant responded to one scenario in the gain frame, one in the loss frame, one in the short delay frame, one in the long delay frame, and so on. Two versions of the questionnaire counterbalanced which scenario appeared in which frame. It was predicted that the “prudent” frames (those shown in past research to decrease discount rates) would lead to higher ratings of the long-term choice option than would the “impulsive” frames (those shown to increase discount rates), Thus, the long-term option should be rated most highly when framed as a loss, long delay, large magnitude, or sequence. An ANOVA was performed that included framing (prudent or impulsive)
94
Gretchen B. Chapman
and framing type (sign, delay, magnitude, or sequence) as within-subject variables and counterbalance condition (2 levels) as a between-subjects factor. Rating of the long-term choice option was the dependent measure. There was a main effect of frame, F(1,80) = 177, MSE = 0.002, p < .0001, indicating that overall the prudent frame led to higher ratings of the longterm choice option (means 75 5 35 versus 79 ? 32). There was also a frame by type of framing interaction, F(3,240) = 68, MSE = 0.001, p < .0001, indicating that some types of framing were more effective than others. Individual t-tests on each of the eight scenarios showed that only three were individually significant at the 0.05 level. The salted pretzel sign effect scenario described above showed an effect (62 versus 78) as well as one delay effect scenario (71 versus 81) and one magnitude scenario (78 versus 90). These results indicate some evidence that factors that influence discount rates also affect semi-realistic decisions. This effect, however, was moderate in size (only 4 points on a 100-point scale) and was individually significant (even with a liberal criterion) for less than half of the scenarios. Because different scenarios were used to demonstrate different types of framing effects, it is difficult to separate the effectiveness of particular scenarios from the influence of type of frames (sign, delay, etc.). What can be concluded is that factors demonstrated to affect discount rates appear to have a small influence on more realistic decisions. The Christensen-Szalanski (1984) study described earlier is a similar demonstration of a framing effect (specifically, the delay effect) influencing realistic choice. Such framing effects could prove to be useful in encouraging decision makers to make more future-oriented decisions about, say, health care. 111. Domain Effects
Normative discounted utility theory prescribes that the same discount rate be used in all decisions. One implication of this prescription is that the domain or content of the decision should not influence the discount rate. The majority of research on time preferences has focused on discounting of monetary outcomes, thus raising the question of whether other types of outcomes, such as health, are discounted differently from money (Olsen, 1993). One answer to this question comes from the findings reviewed in Section 11. By and large, most of the intertemporal biases reviewed have been demonstrated in health decision in addition to financial choices. They delay effect (Chapman, 1996b; Chapman & Elstein, 1995; Redelmeier & Heller, 1993), magnitude effect (Chapman, 1996b; Chapman & Elstein, 1995), and sign effect (Chapman, 1996b; MacKeigan et al., 1993) have all been documented in health decisions. As discussed above, the spreading
Intertemporal Choice
95
effect also appears in both health and money decisions. Preference for sequences (Chapman, 1996a) is the most notable area for which large differences between health and money decisions have been demonstrated. Thus, one method for comparing time preferences in different domains is to examine whether the same biases occur in each domain. (They generally do.) A second method is to examine whether the same mean discount rate is used in each domain. These comparisons have shown mixed results with health discount rates being sometimes higher than money rates (Chapman & Elstein, 1995), sometimes lower (Cairns, 1992), and sometimes the same (Chapman, 1996b).
A. DOMAIN INDEPENDENCE A final way to compare time preferences in health and money domains is to examine the correlation between rates in the two domains to determine whether subjects who use the highest (lowest) discount rates for money also use the highest (lowest) discount rates for health. Chapman and Elstein (1995) gave subjects a number of monetary discounting questions (the individual questions differed in delay and magnitude) and an equal number of health discounting questions. The intra-domain correlation between pairs of money questions or pairs of health questions was much larger than the inter-domain correlation between health-money pairs. In Experiment 2, for example, the mean correlation between pairs of money questions was 0.81 (N = 70, p < .001) and the analogous correlation for health was 0.82. Thus, subjects were quite consistent in their time preferences within a domain. In contrast, the mean between-domain correlation was much lower, 0.26. This comparison of correlations indicates domain independence, or a low relationship between domains despite a high within-domain reliability. Domain independence can also be demonstrated with confirmatory factor analysis (Chapman, 1996b), showing that all the monetary discount rates load on one factor and the health discount rates on a second factor. This phenomena has been demonstrated in several studies (Cairns, 1992; Chapman, 1996b; Chapman & Elstein, 1995). Domain independence represents a violation of normative discounted utility theory because, normatively, decision domain should not influence discount rates. Like other decision biases, domain independence may be a clue as to the psychological processes underlying intertemporal choice. Several potential causes of domain independence are discussed next. FOR HEALTH AND MONEY B. UTILITY FUNCTIONS
Chapman (1996b) assessed the possibility that domain independence might be the result, not of different time preferences for health and money, but
96
Gretchen B. Chapman
of different utility functions for health and money. Discount rates are typically computed as the ratio between the amounts of future and immediate money (or time in health) that are equivalent in value (e.g., Benzion et al., 1989; Chapman & Elstein, 1995; Thaler, 1981). According to normative discounted utility theory, the discount rate should actually be computed as the ratio between the (undiscounted) utilities of future and immediate money (or time in health) that are equivalent in value (e.g., Redelmeier & Heller, 1993). If, for example, a decision maker was indifferent between $10 immediately and $20 in one year, the ratio in amount of money would indicate a 100% annual discount rate (20/10 - 1). Suppose, however, that the utility function for money was not linear, but was instead curvilinear, described, perhaps as u(x) = x O . ~ . The ratio in utilities of monetary amounts would yield an annual discount rate of 62% [(20°.7/100.7)- 11. If the shape of the utility function for money differs (e.g., is more or less curved) from that for health, then the two domains could differ in terms of discount rates computed as ratios of amounts even if they had the same discount rate computed as ratios of utilities. If there are individual differences in utility functions for health and money, then domain independence might be observed in conventional discount rates even though the underlying utility discount rates showed complete consistency. To evaluate this possibility, I assessed subjects’ utility functions for health and money and converted health and money amounts into utility before computing discount rates (Chapman, 1996b). These utility discount rates showed the same domain independence pattern as the conventional rates did. Other methods for accounting for different utility functions for health and money also did not reduce the domain independence effect. Specifically, matching the health and money outcomes used in the time preference choices for utility did not change the results, nor did statistically controlling for each subject’s willingness to trade money for health. Thus, domain independence does not appear to be an artifact of utility functions but is instead a property of intertemporal choice. C. FAMILIARITY WITH DECISION DOMAINS A large majority of intertemporal choice studies have employed healthy, college-aged subjects. This factor could influence findings in that young people generally have higher discount rates than older people (Green, Fry, & Myerson, 1994). It could also influence domain independence, specifically. Whereas almost all decision makers (college students included) are familiar with monetary decision, young healthy college students are likely to be particularly unfamiliar with health decisions, having been faced with few weighty decisions about their own health. The low correlation
Intertemporal Choice
97
between health and money discount rates might indicate an effect of familiarity rather than of decision domain. Decision makers may have consistent time preferences for familiar domains (such as money) but use idiosyncratic strategies for unfamiliar domains (such as health), leading to a low correlation between the two. Chapman, Nelson, Winquist, Fu, Novak, and Hier (1996) tested this possibility by comparing discount rates for familiar and unfamiliar health domains. Two groups of patients participated-patients with migraine headaches visiting a neurology clinic, and patients with inflammatory bowel disease (IBD) who had received care from a surgery clinic. Both groups responded to intertemporal choice scenarios about both disorders. The scenarios involved choices between a somewhat effective, immediate medication and a more effective delayed medication. All patients made choices about a familiar domain (choices relating to their own disorder) and an unfamiliar domain (choices relating to the other group’s disorder), and each scenario was familiar for half the patients. In addition, all patients made intertemporal choices about money, a domain that is presumably familiar to everyone. If domain independence is due to familiarity, one would expect a high correlation between monetary discount rates and discount rates from the familiar health domain, but a low correlation between the familiar and unfamiliar health domain, In contrast to this prediction, the results showed a high correlation between the two health domains and a low correlation between the money domain and both of the health domains. Thus, it appears that domain independence is not caused by differential familiarity with the health and money domains. D. EMOTIONS Although normatively, discount rates should be based on growth and risk associated with temporal delays, descriptively, discount rates may be based on a number of other psychological factors including emotions. Some emotions are tied to knowing that a future event will occur. Such knowledge makes possible anticipation effects (see Section 1I.G) such as savoring, dread, and the unpleasant experience of waiting (e.g., abstinence, see Loewenstein, 1992). Other emotions are tied to knowing that a choice has been made. Options not selected may generate feelings of regret (Loomes & Sugden, 1982). It may be particularly difficult to accept a delayed option when one knows that an immediate option is available. A potential explanation for the lack of correlation between health and monetary discount rates is that the two domains may be differentially affected by anticipation and regret factors. If the influence of these
98
Gretchen B. Chapman
factors could be eliminated, the inter-domain correlation would then be increased. Although it may be impossible to eliminate the role of anticipation and regret in decisions made on one’s own behalf, these psychological factors could be irrelevant to decisions made on behalf of another person, provided the beneficiary merely experiences the outcome of the decision, but does not know of the other options available and does not expect the outcome before it occurs. Thus, decisions made on behalf of another person provide a means of manipulating the relevance of anticipation and regret. I conducted an experiment (Chapman, 1995) in which 347 college students made monetary and health discounting decisions on behalf of another person or for themselves. In the “other” condition, subjects were in charge of executing a late aunt’s will and had to make two decisions. In the money decision, they must distribute some money to cousin Bertha, either $1000 now or a different amount three years from now. Subjects had to specify the amount of the delayed money so that the two options were equally attractive in terms of which one they would like cousin Bertha to have. In the health decision, subjects had to distribute a new medication to cousin Louie who is in a state of poor health. Louie could have a one-year supply right now or a different amount in three years. Again, subjects had to specify the amount of the delayed medication that would make the two options equally attractive in terms of which one they would like cousin Louie to have. In the “self” condition, subjects made the analogous decisions for themselves. The order of the health and money questions were counterbalanced. There were four versions of the “other” condition that resulted from crossing two factors: regret and anticipation. Regret involves comparing an experienced outcome to possible alternatives. In the no-regret conditions, subjects were told that cousin Bertha (and Louie) would not know that two options were available or that the subject (her cousin) selected which one she would receive; she would only know about the alternative she did receive. In the regret conditions, cousin Bertha (and Louie) would know that there were two options and that the subjects selected one of them. In the no-anticipation conditions, cousin Bertha (and Louie) would not know that she had inherited anything until she received the check (or for Louie, until his physician delivered the medication), thus eliminating any possibility of anticipation (savoring, dread, discomfort of waiting, etc.). In the anticipation conditions, the cousins are told what they have inherited and when they will receive it. The “self” condition is analogous to the regret/ anticipation “other” condition because, when deciding for oneself, one always knows what options there are to choose from and what option has been selected, even before the outcome is received.
Intertemporal Choice
99
Subjects in each of the five conditions answered health and money intertemporal choice questions. The responses to these questions were converted to discount rates, and the correlation between health and money discount rates was computed for each condition. Table I11 shows the resulting correlations. The lowest correlation ( r = 0.12) occurred when the recipient of the decision experienced no anticipation and no regret. The correlations in each of the other conditions (includingthe "self" condition) were significant and of similar magnitude (about 0.30). However, none of the pairwise comparisons between correlations was significant ( p s > 0.12). Thus, eliminating the influence of regret and anticipation did not make health and money discount rates more similar. If anything the absence of these emotions decreased the correlations between health and money time preferences (although not significantly). In addition, the health-money correlation for decisions made on one's own behalf was similar to that for decisions made on behalf of someone else, especially if the recipient could experience regret and anticipation.
E. FUNGIBILITY OF HEALTH AND MONEY The reason that normative theory prescribes the same discount rate for both health and money is that violation of this prescription leads to anomalies in trading health for money. For example, if health is discounted at a lower rate than money, then any public health program that spends money now to gain health benefits now would be more cost effective if both the costs and benefits were delayed (indefinitely) into the future (Keeler & Cretin, 1983). Similarly, if health and money are discounted at different rates, then
TABLE I11 SPEARMAN CORRELATIONS BETWEEN HEALTH AND MONEY DISCOUNT RATESFOR DECISIONS INVOLVING OR REGRET ANTICIPATION Anticipation
Reeret
No Yes
No
Yes
0.12 (N = 62) 0.27* (N = 72)
0.35** (N = 70) 0.32** (N = 71)
Note: The analogous correlation for the "self" condition was 0.37 (N < 0.01).
= 72, p
* p < 0.05,** p < 0.01
100
Gretchen B. Chapman
a money pump could be constructed in which the decision makers trade future money for future health now, but in the future they trade the health back for the money (see Section I for a similar example). Both of these examples involve trading health for money and thus make clear that the normative prescription to use the same discount rate in the two domains holds only under the assumption of fungibility, that is, that health can, in fact, be traded for money (and vice versa) (Ganiats, 1994). If money cannot be traded for health, as is arguably the case in some personal decisions, then no paradoxes or inconsistencies will result if different discount rates are used for health and money. Thus, one explanation for domain inconsistency is that decision makers don’t view health and money as fungible and consequently don’t use consistent discount rates. To test this possibility, I distributed two versions of a questionnaire to 206 participants; half were college students and the remainder were community members at a commuter train station (Chapman, 1997). One version of the questionnaire described a situation in which money and health were tradable (abridged description): Imagine that you have migraine headaches 50 days per year. A medication is available to you that will eliminate all of your headaches at a cost of $1000/year, which your health insurance does not cover. You can choose how much of this medication you wish to purchase. You could spend no money and remain in your current state of health (with headaches). You could spend $lo00 per year for the rest of your life and be in good health for the rest of your life with no headaches. Or you could purchase more limited amounts of the medication and be in full health (no headaches) sometimes but in poor health (with headaches) other times. Your income is such that you are able to put $3000 per year in a savings account to be spent on whatever you like.
In the second version of the questionnaire, money and health were not tradable: Imagine that you have migraine headaches 50 days per year. A medication is available on a one-time basis that will eliminate all of your headaches for a limited period of time. There is no cost for the medication; all costs are covered by your insurance company. The medication is not available for sale to the public; the only way to get it is through your insurance company. This is the only opportunity to receive any of this medication.
Both versions of the questionnaire then asked participants to express their time preferences. In the health question they are told that their insurance company was offering them a free supply of the headache medicationeither a 12-month supply right now or a larger supply three years from now. Subjects were asked to specify the supply of delayed medication that would make it just as attractive as the immediate option. In the money
Intertemporal Choice
101
question subjects were told that their insurance company was offering a cash rebate-either $1000 right now or a larger amount in three years. Again, subjects specified the delayed amount that was just as attractive as the immediate amount. The order of the health and money questions was counterbalanced. The Spearman correlation between the health and money discount rates in the not-tradeable condition was 0.02 (N = 101, p > 0.8), a correlation similar in low magnitude to that seen in previous studies of domain independence. In contrast, the correlation in the tradeable condition was 0.44 (N = 105,p < .0001),significantly higher than in the not-tradeable condition (p < .001). 'Thus, in a scenario in which health and money are explicitly tradeable, the correlation between discount rates in the two domains, although not exceedingly high, is larger than seen in previous studies or in the not-tradeable condition in this study. Decision makers are, at least in part, following the prescription of normative discounted utility theory in that their discount rates are more consistent across decisions when the relevant domains are fungible. Thus, it appears that domain independence is not the result of psychological factors such a familiarity, emotions, or utility functions. Instead, it appears to be, at least in part, a result of an economic factor-the fungibility (or lack thereof) of health for money. By this account, domain independence is consistent with normative theory. A question that requires further research, however, is what exactly decision makers are doing differently in health as compared to money intertemporal choices.
IV. Agreement among Measures of Time Preferences Are time preferences stable characteristics of decision makers? Are some decision makers consistently short-sighted whereas others are invaribly future-minded? If this were the case, there should be agreement among different measures of time preferences. In addition, one would expect that decision makers with low discount rates would behave differently than their high discount rate counterparts. They might, for example, invest more money or engage in more preventive health behaviors such as exercise. The intertemporal choice biases reviewed above illustrate that time preferences are not always stable; instead, they are influenced by a number of features of the decision task such as sign, delay, or decision domain. These results suggest that different indicators of time preferences (for example, questionnaire responses and real behavior) may not agree unless many of the decision task features are similar on both measures.
102
Gretchen B. Chapman
A. CONSIDERATION OF FUTURE CONSEQUENCES Some researchers have hypothesized that time preference is a stable characteristic of the decision maker. For example, Strathman, Gleicher, Boninger, and Edwards (1994) posited consideration of future consequences to be a stable individual difference. They developed the Consideration of Future Consequences scale (CFC), which consists of 12 items such as “I am willing to sacrifice my immediate happiness or well-being in order to achieve future outcomes” and “I only act to satisfy immediate concerns, figuring the future will take care of itself.” Respondents rated each statement on a 5-point scale. In one study, Strathman et al. presented subjects with arguments for and against increases on an oil drilling project. Subjects who showed higher scores on Consideration of Future Consequences placed more weight on the long-term environmental effects of drilling and thus were more opposed to it. In a second study, CFC scores were correlated with cigarette consumption, environmental behavior (e.g., recycling), and general concern for health. Strathman et al. did not, however, compare CFC scores to discount rates. I conducted a study that compared the CFC scale to discount rates. Fifty college students completed a questionnaire with three sections. The first section contained 18 money-discounting questions. Subjects specified the amount of immediate money that would be just as attractive as a delayed amount. The magnitude, sign, and delay varied over the questions. The second section contained 18 analogous health-discounting questions. Subjects specified the amount of time in good (or poor) health that was just as attractive as a delayed period of good (or poor) health. The magnitude and delay of the delayed option as well as its sign (good or poor health) varied over questions. An average health discount rate and an average money discount rate was computed for each subject. The third section of the questionnaire contained the CFC scale (Strathman, et al. 1994). The CFC scale was scored so that higher scores indicated more futureoriented time preferences. It was marginally correlated with health discount rates (Spearman r = -0.27, N = 50, p < 0.06) so that, as expected, higher CFC scores were associated with lower discount rates. The CFC score was not, however, related to money discount rates ( r = 0.05, N = 50, p > 0.7), and health and money discount rates were not related to one another ( r = 0.06, N = 50, p > 0.6). Thus, as in previous demonstrations of domain independence (see Section 111) health and money discount rates were not correlated with one another. Consideration of future consequences was modestly related to health discount rates, suggesting that they are both measures of time preference. Consistent with the finding by Strathman et al. that CFC is related to health behavior, it was health discount rates (rather than money) that were related to CFC.
Intertemporal Choice
103
B. RELATION BETWEEN DISCOUNT RATESAND OTHER MEASURES OF TIMEPREFERENCES The study of Chicago community members described in Section I1 also included a comparison of a number of questions that could be considered measures of time preferences. Sixty of the 82 participants provided complete data on all four of the measures described below.
1. Discount Rates As described in Section II.A, participants answered health and money discount questions. Four money questions asked them to compare receiving $10 a day for one (or five) year(s) starting today versus $10 per day for a longer period of time starting one (or seven) year(s) from now. Four health questions asked them to compare a treatment that restored full health for one (or five) year(s) beginning immediately or a treatment that gave full health for more than a year but only after a one- (or seven)-year delay. The mean Spearman correlation between pairs of money discount rates was 0.71 (N = 60, p < O.OOOl), and the analogous correlation for health was 0.82 (N = 60, p < 0.0001). Thus, the discount rates were reliable across questions within a decision domain.
2. Decision Scenario Ratings As described in Section II.H, participants also responded to eight health decision scenarios. Each scenario presented a long-term and a short-term choice option, and participants divided 100 points between the two options. The scenarios were framed in different ways, which affected the ratings (see Section 1I.H). Apart from the framing effects, however, the points assigned to the long-term option could be seen as a measure of time preference, with higher ratings indicating a lower discount rate. As a measure of reliability of this measure, I computed the Spearman correlation between the ratings for the odd-numbered scenarios and those for the evennumbered scenarios. This split-half reliability was 0.38 (N = 60,p < 0.005). Thus, ratings were related across scenarios, but the correlation was only moderate. The framing manipulation may have reduced the correlation. 3. Self-Report of Health Behaviors Participants also provided self-reports of preventive health behaviors. They read a list of 24 behaviors used by Ajzen and Timko (1986) (e.g., “I brush my teeth at least twice a day,” “I avoid eating high cholesterol foods”) and used a 7-point scale (from always to never) to rate how each behavior matched their own behavior. Higher scores indicated more preventive
Gretchen B. Chapman
104
health behaviors, which should indicate a low discount rate. The split-half reliability of the scale was 0.77 (N = 60, p < 0.0001). 4.
Exercise
Participants were also asked specifically about their exercise behavior. (About half of the participants were recruited at one of several YMCA exercise classes). They indicated the number of times per week that they exercised, and the amount of time they kept exercising during a single session. The amount of time per week spent exercising was computed from these responses. The median amount was 3 hours per week (range 0 to 16 hours per week-only two subjects reported exercising more than 9 hours per week). Table IV shows the correlations among these measures. Of the ten correlations shown, only three are significant. Health and money discount rates are moderately correlated. The size of this correlation is noticeably larger than that seen in most of the domain independence results reported in Section 111. Although the correlation between health and money rates was significantly lower than the within-health correlation (Z = 2.87, p < O.Ol), it was not lower than the within-money correlation (Z = 1.43, p > 0.15). One explanation for the comparatively large correlation between health and money rates is that in this study, both types of outcomes were described as streams of events (e.g., $10 a day for a year). In previous studies (e.g., Chapman, 1996b), health was described as a stream of events, whereas money was depicted as a punctate event. Describing health and money outcomes more similarly may increase the correlation. A second significant correlation shown in Table IV is that between the scenario ratings and the scale of self-reported health behaviors. As expected, higher health behavior scores (indicating more preventive health behaviors) were associated with higher health decision scenario ratings of the long-term option. The final significant correlation was between health discount rates and time spent exercising. Contrary to prediction, lower discount rates were associated with less time exercising. More time spent exercising did not appear to be the result of a greater value placed on future health. Thus, of ten correlations, only three were significant, and one of these was contrary to prediction. Why were these various measures of time preferences not more strongly related? And why was the relation between decision scenario ratings and self-report of health behaviors the strongest relation (apart from that between health and money discount rates)? The overall poor relation among time preference measures could be due to noisy measures of time preferences; however, each of the measures had a fairly high
Intertemporal Choice
105
TABLE IV SPEARMAN CORRELATIONS AMONG MEASURES OF TIME PREFERENCES FROM A STUDY OF CHICAGO-AREA COMMUNITY MEMBERS (n = 60)" Monetary discount rate Health discount rate Decision scenario rating Time spent exercising Scale of health behaviors
0.52* 0.18 0.13
0.04
Health discount rate
Decision scenario rating
-0.06 0.348 -0.13
-0.07 0.39*
Time spent exercising
-0.03
* p < 0.01 in predicted direction 8 p < 0.01 in counter-predicted direction Of the 82 participants, only those (N = 60)providing complete data on all five measures are included in this table. a
reliability (with the exception of exercise behavior, for which there was no reliability check). An alternative explanation is that numerous factors influence time preferences. Although discount rate questions, exercise, decision scenario ratings, and health behaviors all involve trade-offs between short-term and long-term outcomes, the magnitudes of these outcomes, the perceived delays until their occurrence, the decision domain, and possibly other factors may vary both within and between these different measures. Because a decision maker does not have one constant time preference, but instead has many time preferences depending on the situation, it is difficult to detect stable individual differences across different measures of time preferences. The correlation between decision scenario rating and self-report of health behaviors may have occurred because both of these measures involve ratings of whether one does or would engage in specific health behaviors such as brushing teeth or cutting down on salt intake. Both the decision domain and presentation were very similar for these measures, which could have lead to similar time preferences. The fact that the correlation between health discount rates and time spent exercising was counter to prediction suggests that the main motivation for exercise is not to improve future health but rather for short-term consequences such as relieving stress or being in shape.
C. TIMEPREFERENCES AND HEALTH BEHAVIOR One reason for interest in the study of time preferences is its potential application to understanding preventive behavior. Preventive health behav-
106
Gretchen B. Chapman
ior in particular appears to be an obvious example of making an up-front investment in return for a long-term gain. If time preferences are a stable personality attribute, then one would expect people with low discount rates to be more likely to engage in preventive behavior. Fuchs (1982) hypothesized such a relation. He interviewed about 500 community members and asked about their monetary discount rates and several health behaviors: smoking, exercise, seat belt use, dental exams, and being overweight. Discount rates were not significantly correlated with any of these health behaviors, with the exception of a small relation to smoking. In addition, exercise was related to discount rates for men, but in the opposite direction to that expected, similar to the results reported in Table IV. Fuchs found that discount rates were related to other respondent characteristics, such as age and education, indicating that it is possible to detect correlates of time preferences. One possible explanation for the lack of relation between discount rates and health behaviors in Fuchs’ study is that he compared monetary discount rates to health behaviors. The results summarized in Section I11 indicates that monetary discount rates are largely unrelated to health discount rates. Thus, it is perhaps not surprising that money discount rates were not related to health behavior. It may be that health discount rates are more closely linked to health behaviors. To test this possibility, I assessed the time preferences of patients with high blood pressure and compared these to their adherence to medical therapy. Medication adherence was selected as a health behavior because it involves an intertemporal choice (Christensen-Szalanski & Northcraft, 1985). Hypertension generally does not produce symptoms, and thus medication to control blood pressure does not bring any symptom relief; in fact, it may produce unpleasant side effects. Although the medication does not produce any immediate benefits that are noticeable to the patient, it does provide long-term benefits in the form of reduced risk of stroke, heart attack, and kidney disease. Thus, adherence to antihypertension medication requires an immediate cost (in terms of time, expense, and side effects) but provides a long-term benefit. The study included 51 male outpatients seen at the hypertension clinic of a Veterans Administration hospital. They had a mean age of 59 (range 31 to 83) and most (78%) were African American. Patients were interviewed in the clinic waiting area before or after a routine appointment. The interview contained four time preference questions that were the result of crossing two domains (health and money) with two delays (one or three years). Subjects were asked to compare an immediate outcome ($1000 or one year of full health) with another outcome delayed by one or three years. Each question consisted of a series of choices that established a discount rate.
Intertemporal Choice
107
For example, in the three-year health question, patients were first asked whether they preferred one year of full health now or eight years of full health starting in three years. The amount of time in full health for the delayed option was adjusted up or down (to four or sixteen years) depending on the patient’s response. A cascade of six choices was used to establish the amount of time in full health for the delayed option that made it just as attractive as the immediate option. A discount rate was computed from this indifference point. The median discount rate was 188%for the health question with the one-year delay and 57%for the three-year health question. The analogous money rates were 206% and 65%. Two measures of adherence to medication were used: pharmacy records and medical appointments kept. Computerized hospital pharmacy records were accessed for each patient to determine what antihypertensive medications he took and his pattern of refilling the prescriptions. The actual number of days between refills was compared to the prescribed number of days (usually 30). Refill intervals that exceeded the target interval were coded as the number of days late; refills that occurred early or on time were coded as zero. The average across refills and multiple medications constituted the refill delay score for each patient. The average refill delay score across patients was 15.44, coresponding to refilling prescriptions on average about 15 days late (range 0 to 136.5, median 7.3). The second measure of adherence was computed as the percentage of hypertension clinic appointments that the patient kept. The average patient kept 88% of appointments. Routine checkups with a clinician is an important part of management of hypertension. Thus, although this measure does not assess medication-taking behavior directly, it is nevertheless a measure of compliance with the medical regimen. A positive correlation was expected between refill delay scores and discount rates (long refill delays corresponding to high discount rates), and a a negative correlation was expected between percent appointments kept and discount rates. Contrary to prediction, refill delay scores were not correlated with health (Spearman r = -0.10, N = 43, p > 0.5) or money discount rates (r = -0.20, n = 44, p > 0.20). Appointment keeping was also not correlated with health ( r = 0.17, n = 43, p > 0.25) or money discount rates ( r = 0.03, n = 44, p > 0.8). In addition, although one would expect that keeping a higher percentage of appointments would be associated with a lower refill delay score, the two adherence measures were not well correlated with one another ( r = -0.14, N = 42, p > .35). Several limitations of this study could have contributed to these low correlations. First, the measures of health behavior may not have been good indicators of adherence behavior. Appointment keeping was so high (88%) that this measure may have suffered from a ceiling effect. A similar problem is the
108
Gretchen B. Chapman
fact that the patients were attending a hypertension clinic, which may indicate that they are fairly compliant, in contrast to those who have not sought care. Thus, the sample in this study may not have provided enough variance in medication adherence to allow for detection of a correlation. Another measurement issue was that the discount rate measure also suffered from a ceiling effect. On each of the four discounting questions, an average of 24% of patients chose the immediate option on every choice in the cascade, indicating a discount rate too high to be measured by the current procedure. (For purposes of computing correlations, they were assigned the maximum measurable discount rate.)
D. SUMMARY The studies described in this section showed very weak, if any, relation between discount rates, other measures of time preferences, and actual behavior. Although in a few cases the hypothesized relations were observed, in many others there was no relation at all. Each of the studies had certain limitations that might explain why few or small correlations were detected. The preponderance of the evidence, however, is that discount rates are not strongly related to other indicators of time preference.
V. Conclusions The research reviewed in this chapter indicates that time preferences are quite labile. They are influenced by a number of decision parameters, such as magnitude, delay, and sign (see Section 11) and vary with decision domain (Section 111). Finally, there is not high agreement among possible measures of time preferences (Section IV). What implications do these findings have for the psychological processes underlying intertemporal choice? Two classes of explanatory accounts are discussed below. A. PSYCHOPHYSICS Some of these intertemporal biases can be explained in psychophysical terms. Kirby and Herrnstein (1995), for example, explain the delay effect (and resulting preference reversals) by a hyperbolic discount function. Thus, the particular algorithm by which the magnitude and delay of an outcome are combined results in the delay effect. Prelec and Loewenstein (1991) follow a similar approach in their accounts of the delay, magnitude, and sign effects. According to their account, the delay effect is due to a particular psychophysical function of time so that the same time interval has more psychological weight if it occurs after a short rather than long delay. The
Intertemporal Choice
109
magnitude and sign effects are due to the shape of the utility function for money (or other outcomes) so that a given ratio between two amounts carries more psychological weight if the amounts are (a) large rather than small and (b) losses rather than gains. This class of explanation of intertemporal biases point to their similarity to other psychophysicalbiases reviewed by Arkes (1991). Other examples of psychophysical biases include those resulting from Prospect Theory’s S-shaped value function (Kahneman & Tversky, 1979), for example the reflection effect (Tversky & Kahneman, 1981) or mental accounting (Thaler, 1985). According to psychophysical accounts, biases occur because the psychological dimension does not match the objective dimension. This type of explanation can account for a number of the biases reviewed in Section 11. It might also explain the weak relations among different time preference measures discussed in Section IV, if one posits that parameters of the psychophysical function (e.g., magnitude delay) differ across the various time preference measures. This account would make the prediction that agreement across measures would be increased by further specifying the parameters of each measure (e.g., magnitude and delay explicitly defined for discount rates questions, decision scenarios, CFC scale questions, etc.). The psychophysical approach cannot explain the domain independence results reviewed in Section 111, however. Even when both money and health outcomes are transformed from the physical scales of dollars and years in good health to the psychological scale of utility, discount rates from the two domains are still unrelated (Chapman, 1996b). A different class of explanation is needed to account for this result. B. CONSTRUCTIVE PREFERENCES Another interpretation of intertemporal biases is that time preferences, like other components of decision making, are not well defined but instead are constructed on-line in the context of the particular decision (Payne, Bettman, & Johnson, 1992). The constructive preferences approach posits that judgments or choices are made by constructing a preference based on information available in the environment or retrieved from memory. The judgment could be biased if irrelevant information is incorporated or if the judgment is based on a biased subset of relevant information. The information used and the strategies adopted for integrating the information can vary with context such as decision domain. The constructive preferences approach incorporates additional determinants of biased judgment in addition to psychophysical factors. For example, Arkes (1991) presents, in addition to psychophysical biases, other categories of biases such as association-based biases. These biases occur when normatively irrelevant
110
Gretchen B. Chapman
associations influence judgments. For instance, hindsight bias (Fischhoff, 1975) occurs because cues that would have predicted the outcome are more salient than cues that would have predicted an alternative outcome. Consequently, subjects report that they would have been able to predict the actual outcome. This type of account fits well with the constructive processes approach. How might constructive preferences explain intertemporal biases, particularly domain independence? The construction of time preferences may be influenced by the decision domain because, for example, different types of information are retrieved as the basis for each judgment. For example, in considering intertemporal monetary decisions, subjects might retrieve instances of previous decisions involving investments, loans, or credit cards. In contrast, in considering intertemporal health decisions, subjects may consider previous choices about vaccinations, diet, or medications. Thus, different schemas, exemplars, and strategies may be applied to decisions in the two domains, resulting in domain independence. For instance, suppose most decision makers formed monetary time preferences by referring to a credit card scenario. In contrast, in forming health-time preferences, they might show large individual differences in the scenarios considered. Some might think of buying health insurance, thus using a discount rate similar to that used in monetary transactions. Others might think of exercising or smoking and thus use a discount rate very different from that used in monetary decisions. The variation in health schemas but not in money schemas would result in a low correlation between health and money discount rates (Chapman, 1996b). Other research supports the idea that different decision processes are used depending on the decision context. Goldstein and Weber (1995) found that subjects tended to use attribute-based processes (weighing advantages and disadvantages) when deciding about consumer goods, such as a CD player. In contrast, when making social decisions, such as choosing a spouse, they tended to use schema-based processes (constructing a story about the way things would go). This type of contingent valuation is captured by the constructive preferences framework. Intertemporal biases, including domain independence, can be explained by positing that different decision processes are used depending on the decision domain or type of time preference question. Further research is needed to specify what these different decision processes might be. C. SUMMARY Time preferences vary according to a number of context factors. Some of these effects can be explained well by psychophysical accounts whereas
Intertemporal Choice
111
others, especially domain independence, are better explained by constructive preferences. Further research is needed on the different decision processes used in intertemporal choice and the manner in which they result in intertemporal biases.
ACKNOWLEDGMENTS This research was supported by NSF Grant 9796042 to the author. Thanks to Jennifer Winquist for research assistance, to Jean Anderson Eloy, Mei-ling Fu, Macy Ng, Lorelle Nevils, Brian Novak, and Ingrid Thomas, for data collection, and to Jon Baron for helpful comments. Correspondence concerning this paper should be addressed to Gretchen Chapman, Rutgers University, Department of Psychology,Busch Campus, 152 Frelinghuysen Road, Piscataway, NJ 08854-8020, E-mail:
[email protected].
REFERENCES Ainslie, G. (1975). Specious reward: A behavior theory of impulsiveness and impulse control. Psychological Bulletin, 82, 463-509. Ainslie, G. (1982). A behavioral economic approach to the defense mechanisms: Freud’s energy theory revisited. Social Science Information, 21, 735-779. Ainslie, G. (1984). Behavioral economics 11: Motivated involuntary behavior. Social Science Information, 23, 47-74. Ainslie, G. (1986). Beyond microeconomics: Conflict among interests in a multiple self as a determinant of value. In J. Elster (Ed.), The Multiple Se/$ Cambridge University Press. Ainslie, G. (1991). Derivation of “rational” economic behavior form hyperbolic discount curves. American Economic Reivew, 81, 334-340. Ajzen, I., & Timko, C. (1986). Correspondence between health attitudes and behavior. Basic and Applied Social Psychology, 7, 259-276. Arkes, H. R. (1991). Costs and benefits of judgment errors: Implications for debiasing. Psychological Bulletin, 110, 486-498. Benzion, U., Rapoport, A., & Yagil, J. (1989). Discount rates inferred from decisions: An experimental study. Management Science, 35, 270-284. Cairns, J. A. (1992). Health, wealth and time preference. Project Appraisal, 7, 31-40. Chapman, G. B. (1995). The role of emotions in decisions about future health and money. Paper presented at the 1995 meeting of the Society for Judgment and Decision Making, Los Angeles, CA. Chapman, G. B. (1996a). Expectations and preferences for sequences of health and money. Organizational Behavior and Human Decision Processes, 67, 59-75. Chapman, G. B. (1996b). Temporal discounting and utility for health and money. Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 771-779. Chapman, G. B. (1997). Your money or your health: How the fungibility of health and money affects temporal discount rates. Paper presented at the 1997 biannual Subjective Probability, Utility, and Decision Making (SPUDM) conference in Leeds, England. Chapman, G. B., & Elstein, A. S. (1995). Valuing the future: Temporal discounting of health and money. Medical Decision Making, 15, 373-386. Chapman, G. B., Nelson, R. L., Winquist, J. R., Fu, M. L., Novak, B. S., & Hier, D. B. (1996). Does familiarity with a decision affect patient preference judgments? Time preferences
112
Gretchen B. Chapman
in familiar and unfamiliar disease scenarios. Paper presented at the 1996 meeting of the Society for Medical Decision Making, Toronto, Ontario. Abstract published in Medical Decision Making, 16, 452. Chapman, G. B., & Winquist, J. R. (in press). The magnitude effect: Temporal discount rates and restaurant tips. Psychonomic Bulletin and Review. Christensen-Szalanski,J. J. J. (1984). Discount functions and the measurements of patients’ values: Women’s decisions during childbirth. Medical Decision Making, 4, 47-58. Christensen-Szalanski,J. J. J., & Northcraft, G. B. (1985). Patient compliance behavior: The effects of time on patients’ values of treatment regimens. Social Science and Medicine, 21, 263-273. Elster, J., & Loewenstein,G. (1992). Utility from memory and anticipation. In G. Loewenstein and J. Elster (Eds.), Choice Over Time. New York Russell Sage Foundation, pp. 213-234. Fischhoff,B. (1975). Hindsight # foresight: The effect of outcome knowlege on judgment under uncertainty.Journal of Experimental Psychology: Human Perception and Performance, 1, 288-299. Fishburn, P. C., & Rubenstein, A. (1982). Time preference. International Economic Review, 23, 677-694. Fuchs, V. R. (1982). Time preference and health: An exploratory study. In V. R. Fuchs (Ed.), Economic Aspects of Health. Chicago: University of Chicago Press, pp. 93-120. Ganiats, T. G. (1994). Discounting in cost-effectiveness research. Medical Decision Making, 14, 298-300. Goldstein, W. M., & Weber, E. U. (1995). Content and discontent: Indicationsand implications of domain specificity in preferential decision making. In J. R. Busemeyer, R. Hastie, and D. L. Medin (Eds.), The Psychology of Learning and Motivation, 32, 83-136. Green, L., Fry, A. F., & Myerson, J. (1994). Discounting of delayed rewards: A life-span comparison. Psychological Science, 5, 33-36. Green, L., Myerson, J., & McFadden, E. (1997). Rate of temporal discounting decreases with amount of reward. Memory and Cognition, 25, 715-723. Kahneman, D., Fredrickson, B. L., Schreibner, C. A., & Redelmeier, D. A. (1993). When more pain is preferred to less: Adding a better end. Psychological Science, 4, 401-405. Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decisions under risk. Econometrica, 47, 263-291. Keeler, E. G., & Cretin, S. (1983). Discounting of life-saving and other non-monetary effects. Management Science, 29, 300-306. Kirby, K. N., & Hermstein, R. J. (1995). Preference reversals due to myopic discounting of delayed reward. Psycjzological Science, 6, 83-89. Kirby, K.N., & Marakovic, N. N. (1995). Modeling myopic decisions: Evidence for hyperbolic delay-discounting within subjects and amounts. Organizational Behavior and Human Decision Processes, 64,,22-30. Kirby, K. N., & Marakovic, N. N. (1996). Delay-discounting probabilistic rewards: Rates decrease as amounts increase. Psychonomic Bulletin & Review, 3, 100-104. Loewenstein, G. (1987). Anticipation and the valuation of delayed consumption. Economic Journal, 97, 666-684. Loewenstein, G. (1988). Frames of mind in intertemporal choice. Management Science, 34, 200-214. Loewenstein, G . (1992). The fall and rise of psychological explanations in the economics of intertemporal choice. In G. Loewenstein and J. Elster (Eds.), Choice Over Time. New York Russell Sage Foundation, pp. 3-34. Loewenstein, G., & Prelec, D. (1991). Negative time preference. American Economic Review, 81, 347-352.
Intertemporal Choice
113
Loewenstein, G., & Prelec, D. (1992). Anomalies in intertemporal choice: Evidence and interpretation. Quarterly Journal of Economics, 107, 573-597. Loewenstein, G., & Prelec, D. (1993). Preferences for sequences of outcomes. Psychological Review, 100, 91-108. Loewenstein,G., & Sicherman,N. (1991). Do workers prefer increasing wage profiles? Journal of Labor Economics, 9, 67-84. Loewenstein, G., & Thaler, R. H. (1989). Anomalies: Intertemporal choice. Journal of Economic Perspectives, 3, 181-193. Loomes, G., & Sugden, R. (1982). Regret theory: An alternative theory of rational choice under uncertainty. Economic Journal, 92, 805-824. MacKeigan, L. D., Larson, L. N., Draugalis, J. R., Bootman, J. L., & Bruns, L. R. (1993). Time preference for health gains versus health losses. PharmacoEconomics, 3, 374-386. Olsen, J. A. (1993). On what basis should health be discounted? Journal of Health Economics, 12, 39-53. Payne, J. W., Bettman, J. R., & Johnson, E. J. (1992). Behavioral decision research: A constructive processing perspective. Annual Review of Psychology, 43, 87-131. Prelec, D., & Loewenstein, G. (1991). Decision making over time and under uncertainty: A common approach. Management Science, 37, 770-786. Redelmeier, D. A., & Heller, D. N. (1993). Time preferences in medical decision making and cost-effectiveness analysis. Medical Decision Making, 13, 212-217. Shelley, M. K. (1993). Outcome signs, question frames, and discount rates. Management Science, 39, 806-815. Shelley, M. K. (1994). Gain/loss asymmetry in risky intertemporal choice. Organizarional Behavior and Human Decision Processes, 59, 124-159. Strathman, A., Gleicher, F., Boninger, D. S., & Edwards, C. S. (1994). The consideration of future consequences: Weighing immediate and distant outcomes of behavior. Journal of Personality and Social Psychology, 66, 742-752. Strotz, R. H. (1956). Myopia and inconsistency in dynamic utility maximization. Review of Economic Studies, 23, 165-180. Thaler, R. H. (1981). Some empirical evidence on dynamic inconsistency. Economic Letters, 8, 201-207. Thaler, R. (1985). Mental accounting and consumer choice. Marketing Science, 4, 199-214. Tversky, A., & Kahneman, D. (1981). The framing of decisions and the rationality of choice. Science, 211, 453-458. Varey, C., & Kahneman, D. (1992). Experiences extended across time: Evaluation of moments and episodes. Journal of Behavioral Decision Making, 5, 169-185. Weinstein, M. C., & Stason, W. B. (1977). Foundations of cost-effectivenessanalysis for health and medical practices. New England Journal of Medicine, 296, 716-721.
This Page Intentionally Left Blank
STRATEGY ADAPTMTY AND INDIVIDUAL DIFFERENCES Christian D.Schunn Lynne M. Reder
I. Conceptions of Individual Differences Two approaches have traditionally dominated the study of individual differences in cognitive performance, including studies of aging, brain damage, child development, expertise, giftedness, intelligence, schizophrenia, and adult individual differences. The first is the parameters approach. This approach assumes that individuals vary in performance because of differences in some fundamental aptitude or parameter of the brain or cognitive architecture. This parameter approach includes both psychometric approaches to individual differences (e.g., Ackerman, 1989;Snow, Kyllonen, & Marshalek, 1984; Spearman, 1904), and information-processing approaches (e.g., Hunt, Joslyn, & Sanquist, 1996; Just & Carpenter, 1992; Lovett, Reder, & Lebiere, in press; Sternberg, 1977). Many different parameters have been proposed (e.g., g, gain, inductive reasoning, processing speed, working memory capacity, etc.). There are two especially popular parameters: processing speed and working memory capacity. For example, in terms of processing speed differences, researchers have argued that older children think faster than younger children (Fry & Hale, 1996; Kail, 1988), the elderly think slower than younger adults (Salthouse, 1994), the gifted think faster than average children (Saccuzzo, Johnson, & Guertin, 1994), and schizophrenics think slower than normals THE PSYCHOLOGY OF LEARNING AND MOTIVATION, VOL. 38
115
Copyright Q 1998 by Academic Press. All rights of reproduction in any form reserved. 0079-7421/98 S2S.00
116
Christian D. Schunn and Lynne M. Reder
(Schooler, Neumann, Caplan, & Roberts, 1997). In terms of working memory capacity, researchers have argued that aphasics have reduced working memory capacity (Miyake, Carpenter, & Just, 1995), children have higher working memory capacity (Case, 1985; Fry & Hale, 1996), and general intelligence differences depend heavily on working memory capacity differences (Just & Carpenter, 1992). The other traditional approach is the strategies approach that assumes there are many different ways in which a task can be attacked and that individuals vary in the strategies they use. For example, proposals have been made that older children use different strategies than younger children in a wide variety of domains (Siegler, 1983), the aged use different strategies than younger adults (Dunlosky & Connor, 1997; Reder, Wible, & Martin, 1986; Shapira & Kushnir, 1985), good students self-explain and poor students do not (Chi, Bassok, Lewis, Reimann, & Glaser, 1989), experts use different strategies than novices (Chi, Feltovich, & Glaser, 1981;Ericsson & Polson, 1988; Larkin, McDermott, Simon, & Simon, 1980), optimists use different strategies than pessimists (Carver & Scheier, 1992),and individuals from different cultures use different strategies (Greenfield & Lave, 1982; Wagner, 1978). A variant of the strategies approach is the styles approach, in which individuals are thought to vary in terms of their general cognitive styles, or typical modes of processing information (see Sternberg & Grigorenko, 1997 for a review). Typically researchers tend to emphasize one approach over the other. Yet, the strategies approach and the parameters approach need not be mutually exclusive. One could argue that individuals select different strategies in order to compensate for parameter differences. For example, the aged have been argued to rely more heavily on plausible reasoning strategies because their exact memory retrievals are more effortful than for younger adults (e.g., Reder, Wible, & Martin, 1986). More recently researchers have argued that the simple form of the strategies approach was incorrect. Specifically, they have argued that almost everyone uses multiple strategies, and the different groups of people shared many, if not most, strategies. This pattern was first found by Reder (1982, 1987) in a question answering domain, and subsequently has been found in every domain in which it has been studied (see Siegler, 1996 for a review). This observed ubiquity of multiple strategy use then led the strategies proponents to propose a general perspective on individual differences: groups vary in their distribution of use of strategies (i.e,, when each strategy is used). For example, although older and younger children do simple addition problems using both counting and retrieval, the way they vary is that older children use retrieval more often, especially on more difficult problems.
Individual Differences in Adaptivity
117
A new approach to individual differences is the strategy adaptivity approach (Reder & Schunn, in press). This approach builds on the multiplestrategies approach, but assumes that people vary in how adaptive they are in their strategy selections. That is, although two individuals may have the same set of strategies, they may differ in their abilities to select the best strategy for a given situation. For example, some people may be less capable of flexibly changing their strategy use as the situation changes. This approach does not assume that some people suffer from lack of metacognitive knowledge of the particular strategies; rather, the approach assumes that people vary in their general ability to detect situational change and/or select strategies appropriate to the new situation. This chapter will explore this approach to individual differences. The importance of this new approach is twofold. First, it provides a new view of individual differences that can be explored in many domains. Perhaps differences across individuals or groups in various domains that have been ascribed to parameter or strategy differences are in fact better characterized as differences in strategy adaptivity. Second, this approach can provide new information regarding the processes of strategy selection and strategy adaptivity. For example, correlations of individual differences with various ability tests can provide information regarding what types of underlying processes are important in strategy selection and strategy adaptivity. This chapter has two goals, and the structure of the chapter reflects these goals. First, the chapter seeks to explicate the concept of strategy adaptivity and its components. Second, the chapter seeks to demonstrate the existence of individual differences in strategy adaptivity, as well as explore the stability and possible causes of such individual differences. These issues will be explored in several different task domains, with a section devoted to each task domain.
11. Strategy Adaptivity in Arithmetic
A. RETRIEVE VERSUS COMPUTE A domain in which strategy selection has been studied quite frequently is arithmetic, which includes addition (Geary & Brown, 1991; Lebiere & Anderson, in press; LeFevre, Sadesky, & Bisanz, 1996; Siegler & Jenkins, 1988), subtraction (Siegler, 1987), multiplication (Lemaire & Siegler, 1995; Reder & Ritter, 1992; Schunn, Reder, Nhouyvanisvong, Richards, & Stroffolino, 1997; Siegler, 1988; Siegler & Lemaire, 1997), and artificial alphaarithmetic (Logan, 1988; Zbrodoff, 1995). In this domain, the primary strat-
118
Christian D. Schunn and Lynne M. Reder
egy decision is whether to retrieve the answer or compute the answer (e.g., count out the addition on one’s fingers for addition or multiplying out the numbers in multiplication). Although other decisions do occur (e.g., which strategy to use if computation is selected), this particular strategy choice accounts for the largest percentage of the variance in accuracy and reaction times of responses (Siegler & Shipley, 1995). Computation is slow but accurate, and retrieval is fast but error-prone. Thus, in this domain, the primary issue of strategy adaptivity is defined in terms of how people select to retrieve when they know the answer well and use computation when they do not know the answer well. Why is this a strategy choice? Under one commonly held view, people are assumed to always retrieve first and switch to computation once retrieval has failed (e.g., Lebiere & Anderson, in press; LeFevre, Greenham, & Waheed, 1993; Siegler & Shrager, 1984). Under this view, there is no strategic selection between retrieval and computation before a strategy is tried. Although this view is quite intuitive, several facts make it untenable. For example, people typically switch to computation faster than they would wait for retrieval to fail. Consider tip-of-the-tongue phenomena in which people will persevere in trying to retrieve for many seconds. Yet, for other problems, people either switch to computation or say “don’t know” right away (e.g., what is 24 X 38?). Another view is that people do retrieval and computation in parallel, and whichever finishes first is selected (Logan, 1988). Such a model can produce a very good account of the shape of learning curves. It also makes the selection of retrieve versus compute a non-issue. However, it also has fundamental problems. First, people don’t always try computation right away (consider tip-of-the-tongue phenomena). Second, making the answer more available by priming it does not lead to a higher rate of selecting to retrieve (Reder, 1979), whereas priming parts of the problem statement (which does not make the answer more available) does affect the rate of selecting to retrieve (Reder, 1987). The alternative view, which currently has the most empirical support, is that people make a decision to retrieve or compute before trying either (Reder, 1982,1987,1988;Reder & Ritter, 1992; Schunn et al., 1997;Siegler & Shipley, 1995). The key issue, then, is how people are able to make this strategy selection decision so rapidly. B. METHODS FOR STUDYING STRATEGY SELECTION
In studying the selection of retrieve versus compute strategies, the most common methodologies are to either look for overt strategy use (e.g., using pencil and paper to calculate, or fingers to count) with the assumption that
Individual Differences in Adaptivity
119
no overt strategy use implies retrieval, or to ask subjects after the fact which strategy they used. The problem with these approaches is that they leave open the possibility that for trials in which the subject apparently used computation they might have tried to retrieve first, and on trials in which the subject apparently used retrieval they might have initially started computation. One alternative approach is the game show or feeling-of-knowing paradigm developed by Reder (1987; Reder & Ritter, 1992). In this paradigm, subjects are shown a question or problem, and have to quickly decide whether the answer is known. For example in Reder and Ritter (1992) and Schunn et al. (1997), subjects had to decide in less than 850 ms whether they will be able to retrieve the answer to an arithmetic problem, or whether they will have to calculate the answer. This decision was reported with a key press. If they choose to retrieve, then the answer had to be produced in 2.4 s. If they choose to calculate, then they are given 30 s to calculate the answer. A payoff structure was provided to reward subjects for accurately predicting strategy use: 50 points if retrieve is chosen and the correct answer is produced within the 2.4 s deadline; 5 points if calculate is chosen and a correct answer is produced within 30 s; 1 point if a correct answer is produced too late; and 0 points for incorrect answers. The problems are selected such that the answers cannot be calculated within the 2.4 s (e.g., 26 X 37). Although none of the problems were initially familiar to the subjects (to control individual differences in preexisting knowledge), the problems were repeated many times such that the subjects learned the answer to many of the problems. In this paradigm, subjects quickly become quite adept at making their strategy selections within the 850 ms decision deadline, making retrieve-calculate strategy selections well before they are able to see whether retrieval will succeed. C. AN EXAMPLE STUDY
Schunn, Reder, Nhouyvanisvong, Richards, and Stroffolino (1997) contains a simple study (Experiment 2) that employs the game show paradigm and has many useful properties for the purposes of this chapter. This study consisted of two experimental sessions on consecutive days. On the first day, subjects were given 16 problems that appeared over and over again with varying frequencies (either 27, 18, or 12 times over the course of the session). The problems were either multiplication or sharp (a novel operator similar in difficulty to multiplication), and the operands were all in the 12-38 range). On the second day, subjects were given 16 completely new training problems, also involving multiplication and sharp and operands in the 12-38 range, and also varying in presentation frequency (20, 10, or 5
120
Christian D. Schunn and Lynne M. Reder
times over the course of the session). On every trial during both sessions, subjects used the game show paradigm: first rapidly deciding whether to retrieve or calculate by selecting between two keys, and then providing the answer within the appropriate deadline. On only the second day, there were a few operator swap problems intermingled with the training problems. These operator swap problems were copies of the training problems except that the operator was swapped (e.g., if 34 X 17 was a training problem, then 34 # 17 would be an operator swap problem). Some of the swap problems were swaps of training problems from Day 1, and some were swaps of training problems from Day 2. The purpose of these swap problems was to investigate whether subjects were using the familiarity of the problem (rather than the availability of the answer) to make their retrieve-calculate strategy selections. If the subjects choose to retrieve just as often for the operator-swapped problems (to which they did not know the answer) as for the original training problems, then they could not be using the retrievability of the answer to make their strategy decision. The next sections reanalyzes the data from this study set to illustrate both the various components of strategy adaptivity at the global level and individual differences in adaptivity.
D. OVERALL STRATEGY SELECTIONS AREADAPTIVE At the global level, subjects' strategy selections are quite adaptive. Adaptivity is defined as selecting to retrieve when the answer could be retrieved quickly (i.e., the answer time was less than 2.4 s and the answer was correct) and selecting to calculate when the answer could not be retrieved quickly (ie., the answer time was greater than 2.4 s or the answer was incorrect). The majority of the trials in which they do know the answer, they choose retrieval (i.e., a high hit rate), and the majority of the trials in which they do not know the answer, they choose to compute (i.e., a low false alarm rate). Table 1 presents the mean hit rate, false alarm rate, and d ' (Swets, TABLE I RATE, THEMEAN(AND SE) HIT FALSEALARM RATE,AND d' FOR THE Two DAYS
Hit Rate False Alarm Rate d'
Day 1
Day 2
.73 (.05) .33 (.05) 1.18 (.14)
.64 (.05) .24 (.05) 1.21 (.12)
Individual Differences in Adaptivity
-
rn
e!
2
100 90-
-
80
70
-
z
60-
d
50 -
0,
B
4 40
B
121
Day 1 T
-
30 20 10
-
+ Knowanswer
0-
v1
4
90
-
80-
e!
70-
s
60-
8
.W
--It Knowanswer
Fig. 1. The mean select to retrieve and know answer rates within each block (a) for Day 1, and (b) for Day 2.
1986a, 1986b) for each of the two days in the experiment. These levels of d ' are considered quite good when compared to those typically obtained from feeling-of-knowingjudgments in other types of paradigms (e.g., Nelson & Narens, 1990; Schwartz & Metcalfe, 1992).' These levels of accuracy are especially impressive given that the decisions are being made in less than 850 ms (in fact, they are typically made in approximately 500 ms). To illustrate their adaptiveness, we can examine how they select to retrieve in different blocks (of 30 trials) on each day. Fig. l a presents the In this study, subjects made their retrieve-compute strategy selections and produced their answers using the keyboard. In previous studies (e.g., Reder & Ritter, 1992) in which subjects made the strategy selection using the keyboard and simply spoke their answers into a voice key, their d's were even higher (e.g., above 2.0).
122
Christian D. Schunn and Lynne M. Reder
mean rates of selecting the retrieval strategy for each block of Day 1.Figure l a also presents the mean proportion of problems for which the participants knew the answer (i.e., produced the correct answer within 2.4 s, regardless of which strategy decision they made). Fig. l b presents the same information for Day 2. On both days, in the beginning of the experiment, the subjects rarely knew the answers and rarely selected to retrieve. Across blocks on both days, the subjects knew the answers to an increasingly larger proportion of problems and selected to retrieve for an increasingly larger proportion of problems. Although there is a slight bias to select to retrieve more often than when the answer is known (probably reflecting the large point value for on-time retrievals), there was a very high correlation between the mean strategy selection proportions for each block and the mean know rates for each block (3 = .98, p < .0001 and r2 = .93, p < .001 for Day 1 and Day 2, respectively). Not only was there a close correspondence at the aggregate level, but there was also a pretty close correspondence at the individual subject level. Subjects were not all equally adept at memorizing and quickly retrieving the answers. In general, the subjects were quite tuned to their own retrieval success levels. Using the same block sizes (30 trials), the subject strategy selection proprotions for each block were strongly correlated with the subject know rates for each block (? = .70, N = 250, p < .0001 and ? = .79, N = 175,p < .0001 for Day 1 and Day 2, respectively). These subjectlevel correlations assume no separate biases for each individual, making the strong correlations even more impressive. Fig. 2 presents a few individual subject curves (the first four subjects on Day 1). Three of these four subjects were quite tuned in their strategy selections. A later section will explore the issue of whether subjects vary in how adaptive, or how tuned, they are. E. DECOMPOSING ADAFTIVITY
Before turning to the issue of individual differences, the notion of strategy adaptivity must be unpacked further. Past research has identified two separate components of strategy selection, and both components are present in this data set. The first component has to do with learning the base rates of success of a strategy. If an individual were blinded during the strategy selection phase of each trial (i.e., if they had to make the decision without first seeing the problem), by wagering whether they were likely to know the answer they could exhibit strategy adaptivity by simply responding calculate on all early trials (since they did not know the answer to any problems in the beginning), responding retrieve on all trials at the end of a session (since they know the answer to most problems by then), and
Individual Differences in Adaptivity Subject 1 r2=.91
-
Subject 2 r2=.59 100
100 Y) 908 8070602 504 403 302010-
90 80
f
70
3
60 50 40 30 20
3
10 0
0-
1
2 3
Blocks
Subiect 3 r2=.96
4
9
8
5 6 7 8 Blocks
9
10
Subiect 4 r2=.88
80
..
4
100 90 80 70
100 90
8
123
70
60
60 50 40 40
50 40 30 20
30 20 10 10
10 0
0 1
2
3
4
5 6 7 Blocks
8
9
10
1
2 3
4 5
6 7 Blocks
8 9 10
Fig. 2. The mean select to retrieve and know answer rates within each block for the first 4 subjects on Day 1.
responding with a mix of retrieve and calculate in the middle trials. The second component has to do with recognizing which particular problems are very familiar. Because problems were presented at different frequencies (high, medium, and low), not all problems were equally familiar at a given point during the experiment. Subjects could use the familiarity of a particular problem to decide to select to retrieve for the very familiar problems. Although the two components have been given different names by different researchers, this chapter will adopt the terminology developed by Reder (1987; 1988). The first kind of strategy adaptivity is called extrinsic because the adaptivity uses information external to the particular problem, and the second kind is called intrinsic because the adaptivity uses information internal to the particular problem. Variants of these two components can be found in the strategy choice models of Siegler and Shipley (1995) and Lovett and Anderson (1996). The influence of intrinsic adaptivity can be found by examining the operator swap data-problems that look familiar but the answers are un-
Christian D. Schunn and Lynne M. Reder
124
0.6-
I
et
d 0.5-
+ Previous Day Swaps 6
Same Day Training
3
. I
CI
Y
3
0.4-
111
3 0.3-
5
2 0.2a
dw
MediWTl
High
Presentation Frequency Fig. 3. The mean rate of selecting retrieval (and SE) at each frequency level for same and previous day operator-swap test problems and for comparable-time training trials (see text for details) for Day 2. Adapted from Schunn et al. (1997).
known. Fig. 3 presents the mean rates for selecting retrieval for the Day 2 training and operator-swap problems. To make the time periods comparable, training data are only taken from the interval during which swap problems were presented. As the figure shows, there was no effect of switching operators on the same day problems with respect to subjects’ strategy selections-subjects were just as likely to select to retrieve for an operator-swap problem as they were for the original training problem. The effects of frequency of presentation on the previous day operator-switch test problems were in the same direction but weaker. This is to be expected, given that those problems would be much less familiar after a one day delay. These data illustrate how subjects use the familiarity of the problem statement per se to make their retrieve-calculate strategy decisions. The presence of extrinsic adaptivity must be examined jointly with the presence of intrinsic adaptivity because the two covary in the stimuli. This combined analysis was done two ways for the data on both days; statistically and visually (the analyses for each day are presented separately). First, statistically, both frequency of presentation (how many times that problem had been seen before) and trial number (serial position of the trial on that day) were entered into a multiple regression predicting strategy choice on individual trials. For Day 1,both factors were independent predictors (partial r f s of .14, .ll for trial and frequency respectively, N = 7500, p f s < .OOOl). Second, visually, the 300 trials were divided into five blocks of 60
Individual Differences in Adaptivity
125
Day 1
a) 70
1
0
Presentation Frequency U 16-27
-+ 10-15
10
°3
2
1
4
5
3
Blocks
Day 2
b)
1 6070
Presentation Frequency
3 50-
P r%
-U-
16-20
+ 10-15
40-
$
5 302 8 2010 -
I
3
2
4
Blocks
Fig. 4. The mean retrieval rate as a function of presentation frequency and block (a) for Day 1, and (b) for Day 2.
trials, and the presentation frequency range of 1-27 was divided into 1-2, 3-4, 5-9,lO-15, and 16-27 subranges. Fig. 4a presents the mean rate of selecting to retrieve within each block and frequency subrange for Day 1. There are clearly independent effects of both factors on strategy choice: the higher frequency lines are above the lower frequency lines illustrating the intrinsic effect, and the lines have a general positive slope illustrating the extrinsic effect.* The data from Day 2 demonstrate that the increase in bias to select retrieval depends upon the increasing base-rates of knowing the answer. For Day 2, the overall frequency of presentation was lower than for Day 2To calculate the mean for a subrange, the mean of each presentation frequency mean was used-this removed any differential weighting bias across blocks. Correspondingly, for each subrange, only blocks for which all presentation frequencies in the subrange occurred could be included.
126
Christian D. Schunn and Lynne M. Reder
1 and there were many swap problems in the later trials. Thus, there was a much smaller increase in the base rate of knowing the answer (see Figs. 1 and 2). Correspondingly, there was a much smaller predictiveness of trial on strategy selections, whereas the predictiveness of problem frequency was just as large (partial r’s of .04, .19 for trial and frequency, N = 4900, p < .05, p < .0001, respectively). Fig. 4b presents the subrange by block (of 50 trials) decomposition for the Day 2 strategy selection data. There is still an effect of presentation frequency but very little effect of trial. From block 1 to block 2 there is a small increase, but there are equal decreases from blocks 2 to 3 and 4. These changes in bias reflect the introduction of swap problems in blocks 3 and 4-the subjects adjusted their biases to compensate for spurious feeling-of-knowing.Note that the data plotted do not include the swap problems, but merely reflect their influence on the tendency to try retrieval.
F. INDIVIDUAL DIFFERENCES IN ADAFTIVITY Although subjects overall were quite adaptive in their strategy selections, did they vary systematically in how adaptive they are? To investigate individual differences in adaptivity, analyses were conducted on the subject d scores (as defined earlier)-d ’ being a measure of how accurately the subjects chose to retrieve when they actually did know the answer and chose to compute when they did not actually know the answer. The individual difference analyses begin at this aggregate level, collapsing over the intrinsic and extrinsic components of adaptivity. Fig. 5 presents frequency histograms of subject d’s on Day 1 and Day 2 (N = 25). There is a bimodality in the distribution of d ’s for both days. The rough dividing line is a d ’ of 1.OO, with a cluster of subjects with d ‘ s below 1.00 and a cluster of subjects with d’s above 1.00. Note the large range in d’s. There are some subjects with adaptivity levels near chance, and there are other subjects with very high sensitivities (above 2.0). How stable are these individual differences across days? Table 2 presents the correlations of d ’ and p across the two days. In fact, both d ‘ and /3 are quite stable across the two days, although d ‘ is more so than p. Returning to the d ’ threshold of 1.00 in Fig. 5,84% of the subjects are on the same side of the threshold on both days. These stabilities indicate that the individual differences in d ’ cannot be attributed to random noise or insufficient Ns to properly estimate d ’. Is adaptivity in this task simply synonymous with skill in arithmetic? To assess this issue, the subject d I s were regressed against the mean subject know rates (the percentage of trials for which the correct answer was given within 2.4 s), a measure of how quickly subjects were able to learn the
Individual Differences in Adaptivity
a)
401 35
127
Day 1
d‘
d’
Fig. 5. Frequency histogram of subject d ’ s (a) for Day 1, and (b) for Day 2.
TABLE I1 CORRELATIONS BETWEEN d ’
d ‘ Day 1 d ’ Day 2 P Day 1 P Day 2
AND
d ’ Day 1
d’ Day2
1.00 .I6 .44 .38
.I6 1.00 .31 .06
p FOR BOTHDAYS pDay1 .44 .31
1 .oo .54
pDay2 .38 .06 .54 1.00
128
Christian D. Schunn and Lynne M. Reder
arithmetic facts. In fact, d ’ was not correlated with mean know rates on either day ( r = -.05 and r = -23 for Day 1 and Day 2, respectively, p’s > .3), suggesting that adaptivity differences were independent of arithemetic or memorization skill differences. Another worry is that the individual differences are simply due to floor or ceiling effects. That is, it would be hard to demonstrate strategy adaptivity if only one strategy was ever used. For example in the case of Subject 2 in Fig. 2, there appeared to be less adaptivity due to a large bias to select retrieve. To examine this issue, subject d’s was regressed against the mean subject retrieval strategy selection rate (the percentage of trials for which the retrieve strategy was selected). Because both floor and ceiling effects were possible, a quadratic relationship was also assessed. In fact, neither the linear nor quadratic relationships were significant on either days (p’s < .01, p’s > S ) , indicating that the individual differences could not be attributed to floor and ceiling effects-there were some subjects who used both strategies frequently but simply could not select adaptively when to use a particular strategy. Another worry is that the more sensitive subjects were simply the ones who took more time to make their retrieve-compute decisions. However, there was no correlation between subject d ‘ and mean strategy decision time on Day 1 ( r = .02, p < .5), and the correlation was (weakly) in the wrong direction on Day 2 ( r = -.27, p < .25). G. INTRINSIC VERSUS EXTRINSIC INDIVIDUAL DIFFERENCES
The preceding analyses did not establish whether the individual differences in strategy adaptivity were differences in intrinsic adaptivity, extrinsic adaptivity, or both. The following sections attempt to tease apart individual differences on the two components. There are several ways in which individual differences on the two separate dimensions of adaptivity can be measured. The key to these measures is that they take into account the predictiveness of the extrinsic and intrinsic factors for each individual. For example, suppose that for individual A, there was a large difference in memorability of high-frequency problems and low-frequency problems, whereas for individual B, there was only a very small difference. Further suppose that individual A chose to retrieve much more often for the high-frequency problems than the low-frequency problems, whereas individual B showed only a small preference in choosing to retrieve for high-frequency than low-frequency problems. In this case, individual B should not be rated as being less intrinsically adaptive because the particular intrinsic predictor is simply less predictive for that individual. Instead, one might divide the degree of strategy change by the degree of
Individual Differences io Adaptivity
129
retrievability change. This transformation also standardizes the range of scores across different intrinsic predictors. A similar transformation can be applied to extrinsic predictors-divide the degree of strategy change in response to an extrinsic variable by the degree of retrievability change in response to that variable. For both intrinsic and extrinsic measures, a score of one is exactly tuned adaptivity, a score of less than one is undersensitivity (zero being no adaptivity), and a score of more than one is oversensitivity. Another important factor in measuring intrinsic and extrinsic adaptivity is to hold the potential influences of one of them constant while measuring the other. For example, if one examined intrinsic adaptivity in the first two blocks of either day, the influence of extrinsic adaptivity would be very small, as Fig. 4 revealed. This is exactly what was done. For blocks 1 and 2 combined (to increase power), frequencies 1-2 were compared with frequencies 5-9. The difference in retrieval strategy selection rates for the two frequency groups was divided by the difference in know answer rates for the two frequency groups. These ratios ranged from 0 to 1.6 and had a mean of .71 on Day 1, and ranged from 0 to 1.2 with a mean of .51 on Day 2. Over half (57% on Day 1and 75% on Day 2) of the subjects showed undersensitivity (adaptivity < .8), with the remaining subjects mostly closely tuned (.8 Iadaptivity I1.2) and a very few oversensitive (adaptivity > 1.2). The intrinsic adaptivity correlated r = 3 , p < .02 across the two days, suggesting that there was a stable individual difference in intrinsic adaptivity. Nonsignificant linear and quadratic regressions with mean subject retrieval selection rates (PI = -.005, p2 = .00003, p's > .5 for Day 1, and PI = -.003, p2 = -.00006, p's > .5 for Day 2) suggest that the individual differences were not due to floor and ceiling effects in strategy use. However, there were significant correlations with mean subject strategy decision time ( r = .51, p < .02, and r = .44,p < .05 on each day, respectively). Because these correlations are approximately as large as the stability of the individual differences in intrinsic adaptivity, it is likely that those differences are conceived of as differences in how quickly decisions are made. It was difficult to do a similar type of analysis of individual differences in extrinsic adaptivity because, as Figs. 5 and 6 revealed, the influences of intrinsic adaptivity were ubiquitous on both days. The alternative is to do multiple regressions for each subject to assess the independent effects of extrinsic and intrinsic factors in predicting strategy use within each subject. Thus, with this method, one can simultaneously assess individual differences in intrinsic and extrinsic adaptivity. As with the preceding analysis, one must divide by the degree of predictiveness on memorability. Specifically, the @coefficientsfor problem frequency (intrinsic) and trial number (extrinsic) in the multiple regressions on strategy selection for a particular subject
Christian D. Schunn and Lynne M. Reder
130
are divided by the corresponding p-coefficients for the multiple regressions on know answer rates (i.e., answering correctly within 2.4 s) for that subject. Subjects for whom the coefficients in the know answer regression are very small (absolute value < .0005) are excluded because adaptivity to an unpredictive feature is not meaningful. As one would expect, the impact of this was to exclude subjects for only the extrinsic sensitivity measure, the overall less reliable predictor of know rates in this task. Table 3 presents the means and standard deviations for each measure on each day. There was considerably more variance on the extrinsic measure. Although there appears to be less extrinsic adaptivity on Day 2, this is an artifact of different subjects being excluded on different days. There was almost no difference in extrinsic adaptivity values across days for those subjects included on both days, .58 vs. .44,F(1,lO) < 1, MSE = 1.5. The measures of intrinsic and extrinsic adaptivity proved stable across the two days, r = .73,p < .Owl,and r = .67,p < .02, respectively. Moreover, the two measures were independent of one another, r = -.07, r = .12 for each day, p’s > .5, suggesting that the measures were tapping independent constructs. Regressions with mean subject know-answer rates established that intrinsic adaptivity may have been related to arithmetic skill or memorization rates, r = .47, p < .02 and r = .08,p > .5 for each day, respectively. However, extrinsic adaptivity appeared unrelated to know answer rates, r = -.02, p > .9, r = -.23, p > .3. Similarly, quadratic regressions with mean retrieval strategy selection rates revealed that intrinsic adaptivity may be related to floor and ceiling effects, p1= .015, p < .2, p2 = -.00011, p > .3 for Day 1, and PI = .026, p < .05, p2 = -.00032, p < .05 for Day 2. By contrast, extrinsic adaptivity appeared to be also unrelated to mean = .056, p2 = -.001, p’s > .5 for Day retrieval strategy selection rates, 1, and = .016, p2 = -.000067, p’s > .5 for Day 2. Finally, intrinsic TABLE 111
THENUMBER OF SUBJECTS NOT EXCLUDED, THE MEANSTRATEGY AND PERCENTAGE OF SUBJECTS ADAITIVITY, STANDARD DEVIATION, SHOWING UNDERSENSITIVITY. Intrinsic Day 1 Day 2 Extrinsic Day 1 Day 2
N
Mean
St Dev
25 25
.41 .46
.40
15 19
.85 .23
.40 2.1 1.o
%
Undersensitive ( c . 8 ) 84.0 88.0 60.0 78.9
Individual Differences in Adaptivity
131
adaptivity was partially related to mean strategy selection decision time, r = .20, p > .3, and r = .37, p < .1 for each day respectively, whereas extrinsic adaptivity appeared unrelated to decision time, r = .17, p > .5, r = -.22, p > .3 for each day, respectively. In sum, individual differences in intrinsic sensitivities may simply be attributable to differential arithmetic proficiency, large strategy selection biases, and-or speed-accuracy tradeoffs. By contrast, individual differences in extrinsic adaptivity seem NOT to be attributable to these sources, and may instead reflect a more fundamental individual difference factor.
111. Investigating Individual Differences in a More
Controlled Environment One of the problems of analyses of adaptivity individual differences in tasks like arithmetic is the subjects themselves determine the strategy success rates. That is, not only do the subjects vary in terms of which strategies they chose, but they also vary dramatically in terms of how successful are the strategies themselves, and, even worse, the relative successes under different circumstances. The preceding analyses attempted to equate for such differences mathematically by dividing by the predictiveness of the features, but some doubts remain. An alternative approach is to use a task in which the success of the strategies can be controlled entirely by the experimenter, and thus be equated precisely across all individuals. This section reports on a new experiment on individual differences in strategy adaptivity using just such a task, focusing on individual differences in extrinsic adaptivity. Another advantage of the new task is that strategy use is measured implicitly via behavior, and does not require that subjects make a separate, explicit self-report of their strategy use. Finally, the new task is a slower paced task, not requiring rapid strategy selections-it is possible that some of the individual differences in adaptivity for the arithmetic task were due to ineptitude in making rapid key presses. A. THEBUILDING STICKS TASK The Building Sticks Task (BST) (Lovett & Anderson, 1996) is a problemsolving task similar to the classic water jars task (Luchins, 1942). For a given BST problem, subjects must add and subtract an unlimited supply of three different-sized building sticks to create a stick of the desired length (see Fig. 6, “initial state”). BST problems can be solved by one of two strategies. The undershoot strategy involves starting with a building stick that is shorter than the desired stick and then lengthening that stick by
Christian D. Schunn and Lynne M. Reder
132
INITIALSTATE
Possible first moves
Possible second moves
lm r
delete del
...
Ell
b delete b
...
...
L
Fig. 6. Initial and subsequent states in a BST problem. The subjects’ task is to build a current stick (initially length 0) that matches the desired stick in length by adding and subtracting various combinations for the building stick lengths. From Lovett and Anderson (1996).
additional building stick lengths until the desired stick’s length is reached. In contrast, the overshoot strategy involves starting with the building stick that is longer than the desired stick and then shortening that stick by the other building stick lengths. As Fig. 6 shows, subjects implicitly choose between these two strategies in their first step. For example, suppose the desired stick was of length 13 units, and the three sticks a, b, and c, were of lengths 3,17, and 5, respectively. To obtain the desired stick length of 13 units, the subjects might start with stick b of 17 units and remove segments (the overshoot strategy), or the subjects might start with stock c of 5 units and add more segments (the undershoot strategy). In this example, a solution can only be obtained by the undershoot strategy (c + c + a = 5 + 5 + 3 = 13). The overshoot strategy will not work because subtracting lengths a and c from 6 will never lead exactly to a stick of 13 units length. Of course, in other problems the overshoot strategy may be the correct one to use.
Individual Differences in Adaptivity
133
It is very important to note that the subjects were never given the exact numerical lengths of the sticks-the example above was simply used for expository purposes. In the experiment, subjects had to visually estimate the length of each stick. This prevented subjects from being able to solve the task algebraically, and forced them to try a strategy (i.e., make their choice externally) to determine whether or not it would work. Within this task, it is easy to manipulate the base-rates of success of the undershoot and overshoot strategies. Each problem is designed to be solvable by either undershoot or overshoot (but not both) and then the proportion of problems with each solution type is varied across blocks of time. In this way, it is possible to directly control the success rates of each strategy, and thereby more clearly measure individual differences in adaptivity to change success rates (i.e., extrinsic adaptivity).
B. METHODS 1. Subjects
Fifty-six CMU undergraduates participated for course credit, and were randomly assigned to one of two complementary condition^.^
2. Procedure Each subject was given 70 BST problems to solve. To measure extrinsic adaptivity, the base-rates of success of the undershoot and overshoot strategies were manipulated over time. For 10 trials, both strategies were equally successful (i.e., 5 overshoot and 5 undershoot). For the next 30 trials, one strategy was successful on 80% of trials. For the final 30 trials, the other strategy was successful on 80% of trials. Varied across two conditions was the strategy that was more successful first. That is, in one condition (50-8020) the undershoot strategy was successful on 80% then 20% of trials, whereas in the other condition (50-20-80) undershoot was successful on 20% then 80% of trials. The experiment was conducted on a Macintosh IIci, which ran the BST interface, provided initial instructions to subjects, and collected data, including timing data for each mouse click. Another issue of interest is how individual differences in strategy adaptivity relate to explicit awareness of base-rate change. To address this issue, subjects were asked a series of questions at the end of experiment. Specifically, the overshoot and undershoot strategies were described to the subjects, and the subjects were asked, “Did there seem to be any pattern to which strategy worked?” and “Did you notice any changes in the effecBecause of a software bug, there were 42 subjects in the first condition and 14 in the other.
Christian D. Scbunn and Lynne M. Reder
134
tiveness of one strategy or the other as the experiment progressed?” The subjects’ responses were coded for awareness of the base-rate manipulation: a 1 was given if the subject reported awareness of a change and correctly described the direction of the change, and 0 was given otherwise. Only a few subjects had intermediate awareness; apparent awareness of change, but inaccurate in reported direction. These subjects behaved like the unaware subjects and thus were collapsed into that group. Sixty percent of the data was recoded by a second coder, and the inter-rater reliability was 97%. C. RESULTS 1. Overview
The results are broken into three sections. First, there is a presentation of overall extrinsic adaptivity to the manipulations along with a mathematical characterization of aggregate group adaptivity. Second, there is an analysis of individual differences in extrinsic adaptivity. Finally, there is analysis of the relationship between explicit awareness and extrinsic adaptivity.
2. Overall Adaptivity The subjects’ first choices on each trial were used to categorize strategy use into overshoot and undershoot strategies. The 70 trials were divided into blocks of 10 trials. Fig. 7 illustrates the mean undershoot use within each block within each condition, as well as the manipulated success rates of undershoot within each block for each condition. A Block X Condition ANOVA conducted on the mean undershoot use revealed that there was no overall effect of Block, but the interaction of Block by Condition was 50-80-20
-0-
100
-0
904
. 50-20-80
lo< 0 1
2
3
4
5
6
7
Block Fig. 7. Mean (and SE) use of the undershoot strategy within each block in each condition of the BST experiment, along with the manipulated base-rates.
Individual Differences in Adaptivity
135
quite strong, F(6,324) = 14.4, MSE = .03,p < .0001.The 50-80-20 condition showed the expected increase in use of the undershoot strategies from blocks 2 through 4, and then the expected decrease in use of the undershoot in blocks 5 through 7 (analysis of linear trends F(1,82) = 17.4, MSE = .02, p < .0001, and F(1,82) = 32.7, MSE = .02, p < .0001, respectively). The 50-20-80 showed the reverse pattern (analysis of linear trends F(1,26) = 4.0, MSE = .03, p < .06, and F(1,26) = 12.3, MSE = .02, p < ,005, respectively). Thus, on average, the subjects adapted to the base-rate changes. Because the two conditions behaved so similarly, the two conditions were collapsed (to gain greater power for the individual difference analyses) by reversing the values for the 50-20-80 condition (i.e., subtracting the proportions from 1). But before we turn to the issue of individual differences, how might be the group average performance be described? One good approximation is the average success rate of the strategy up until that point, called the global average. For example, by the end of block two, the global average is (50 + 80)/2 = 65. Fig. 8 shows the fit of the global average to the mean subject data. This model with zero free parameters accounts for 93% of the across block variance in the aggregate data. 3. Individual Differences
Most importantly, despite the very good fit of the global average model to the aggregate data, there are large individual differences in strategy adaptivity. Some subjects adapt not at all, some adapted at the level of the global average model, and others adapt much more quickly than the global average model. Using a Monte Carlo simulation (N = lOOO), we can estab100
901 30
-0-
Observed
--...~~~ Global Average
I
I 1 2 3 4 5 6 1 Block
Fig. 8. Mean strategy use (and SE) within each block of the BST experiment, and predictions of the global average model. Copyright 1998 by The American PsychologicalAssociation. Reprinted with permission.
Christian D. Schunn and Lynne M. Reder
136
a
-a 7
3
, ' O ? c ? r ? t Y ' 9
Strategy Adaptivity
Fig. 9. Observed and expected distribution of strategy adaptivity in the BST experiment. Copyright 1998 by The American Psychological Association. Reprinted with permission.
lish the expected variation of the subjects assuming they all had true probabilities of selecting undershoot in a block that corresponded to the mean undershoot use across subjects in each block (i.e., followed the global average model closely). Because there were only 10 trials per block, and each trial produces a binary outcome (retrieve or compute), one would expect a certain amount of variability simply due to sampling noise. Fig. 9 plots the observed distribution of subjects' adaptivity (difference between the mean strategy use in blocks 2-4 and the mean strategy use in blocks 5-7)4, and the expected distribution from the Monte Carlo simulation. The observed distribution is considerably flatter than the expected distribution, indicating more individual differences than could be attributed to sampling noise. Although we would expect 13% of the subjects to have zero or less adaptivity by chance, 25% of the subjects actually fell into this group. At the other extreme, we would expect only 8% of subjects to have greater than .3 adaptivity by chance, in fact 25% of the subjects displayed this high level of adaptivity. Using these cutoff points, the observed distribution of adaptivity did differ statistically from the expected distribution, x2 (df = 2; N = 56) = 3 1 . 6 , ~< .001. These comparisons establish that there are individual differences in extrinsic adaptivity that are not attributable to chance variation. In the subsequent analyses, the subjects whose adaptivity (as defined above) was at or below zero will be called the Nonadaptive subjects. Means for Blocks 2-4 and Blocks 5-7 were used to gain greater power per subject for the individual difference analyses. However, very similar results are obtained when strategy sensitivity is defined as the difference in strategy use between Block 4 and Block 7.
Individual Differences in Adaptivity
137
It is possible that these individual differences in adaptivity are due to floor or ceiling effects in strategy use (Lee,always using undershoot or never using undershoot). However, restricting the subjects to only those who start the first block in the range of 40% to 60% undershoot use, the observed adaptivity remains statistically different from the expected distribution, x 2 (df = 2; N = 37) = 2 4 . 6 , ~ < .001. Moreover, examinations of the distribution of mean percent undershoot across the 70 trials revealed that only 1 of the 16 Nonadaptive subjects was outside of the 20-80% range, and only 4 were outside of the 25-75% range. Therefore most of the Nonadaptive subjects did use both overshoot and undershoot strategies regularly, just not adaptively over time. Another possible explanation for why some subjects may not have been adaptive is that they were simply doing the task too quickly-for lack of motivation or other reasons-to notice the change in base-rates. To investigate this possibility, a Block X Adaptivity (Adaptive-Nonadaptive) ANOVA was conducted on the mean time to make the first move (i.e., clock on the desired first stick and click on its target location). There was < .0001, as the mean a speed-up over blocks, F(6,324) = 7.0, MSE = 1 . 5 , ~ time changed from 5.7 s in block 1 to 4.5 s in block 3. However, there was no effect of Adaptivity, or an interaction with Block, F’s < 1. If anything, the adaptive subjects were slightly faster, 4.6 s versus 5.1 s. Therefore, the Nonadaptive subjects were not simply choosing too quickly-5.1 s is a long time to make two mouse clicks. 4. Relationship of Explicit Awareness to Extrinsic Adaptivity
How did explicit awareness of the base-rate changes (as measured by the debriefing questions) relate to the extrinsic adaptivity? As described earlier, the subjects were divided into those who explicitly noticed the direction of the shift (Aware, N = 18) and those who did not (Unaware, N = 37). An Awareness X Block ANOVA conducted on the mean strategy use in each block revealed a significant interaction, F(6,318) = 3.6, MSE = .02, p < .005.Although both groups showed the same transitions over time, the Aware subjects showed the transition to a greater degree (see Fig. 10). To examine whether the Aware subjects were more likely to show any behavioral transition or whether they simply showed a larger transition, further analyses were conducted. First, a contingency analysis between Awareness (Aware-Unaware) and Adaptivity (Adaptive-Nonadaptive) revealed that the Aware subjects were no more likely to fall in the Adaptive group than the Unaware subjects, 78% versus 70%, respectively), x2 (df = 1; N = 55) < 1. Second, focusing on only the Adaptive subjects, an Awareness (Aware-Unaware) ANOVA conducted on the adaptivity measure (differ-
Christian D. Schunn and Lynne M. Reder
138
100
90
4 *O
B vl
70 60 50 40
30
1
2
3
4
5
6
7
Block Fig. 10. Mean (and SE) strategy use for Aware and Unaware subjects within each block of the BST experiment.
ence in mean strategy use between blocks 2-4 and blocks 5-7) found that the Aware subjects were more adaptive than the Unaware subjects (strategy adaptivity of .32 and .16, respectively), F(1,38) = 10.9, MSE = .02, p < .005. Thus, awareness was associated with larger transitions and not a greater likelihood of making any transition.
D. DISCUSSION As in the arithmetic domain, there were meaningful individual differences in extrinsic adaptivity in the BST domain that could not be attributed to various artifacts. Thus, even when strategy success rates are carefully controlled across individuals, individuals differ significantly in their abilities to change strategy use in response to shifting base-rates of success. The additional contribution of the BST results is that they establish that the individual differences are related to explicit awareness of base-rate change-awareness appears to led to faster or larger changes. The relationship between awareness and strategy adaptivity deserves a cautionary note. Because the relationship found thus far is only correlational, the causality is (quite plausibly) ambiguous. All three logical interpretations are reasonable possibilities; (1) explicit awareness leads to greater shifts in strategy use; (2) greater shifts in strategy use are more likely to lead to explicit awareness after the fact; and (3) some third factor (e.g., working memory capacity or inductive reasoning skill) leads to both greater shifts in strategy use and a greater likelihood of noticing the baserate change.
Individual Differences in Adaptivity
139
IV. Individual Differences in a Complex, Dynamic Task A. STATIC VERSUS DYNAMIC TASKS Although there is now clear evidence that individuals can vary in their extrinsic adaptivity in a few particular tasks, there has been no discussion of whether there are real-world situations in which such individual differences will substantially effect performance. A particular hypothesis is proposed; individual differences in extrinsic adaptivity will be especially important predictors of overall performance in complex, dynamic tasks. Dynamic tasks (contrasted with static tasks) are those tasks in which the environment changes independently of the actions of the individual. For example, in Tower of Hanoi the task never changes unless the individual takes an action. By contrast, in driving a car, the weather, traffic, and roadway conditions can all change independently of the actions of the driver. Of course, the dynamic-static distinction is more of a continuum than a true dichotomy. Thus, the prediction is that the more dynamic the task, the more important differences in extrinsic adaptivity will be for predicting performance. Because most real-world tasks are dynamic, the range of tasks for which such differences will matter is predicted to be quite large. Schunn and Reder (1997; Reder & Schunn, in press) examined individual differences in extrinsic adaptivity in a complex, dynamic task, and those results will be overviewed here. The task they used will be described briefly, followed by a presentation of the results of interest for the current chapter (see Reder & Schunn for the full details). AIRTRAFFIC CONTROLLER TASK B. THEKANFER-ACKERMAN The Kanfer-Ackerman Air Traffic Controller Task (KA-ATC) (Ackerman & Kanfer, 1994; Kanfer & Ackerman, 1989) was designed to simulate dynamic aspects of real air traffic control (e.g., weather changes, consumption of fuel in real time). The object of the KA-ATC task is to accumulate as many points as possible. Points are earned by landing planes (t-50points) and are subtracted by rule violations (-10 points) or plane crashes (-100 points). In the task, subjects must monitor a variety of elements that are displayed on the screen (see Fig. 11): (a) 12 hold pattern positions that are divided into three altitude levels, (b) four runways-two short and two long, one of each running north-south and one of each running east-west, (c) a queue of planes waiting to enter into the hold positions, (d) messages indicating error feedback and changes in runway conditions, wind speed, and direction, and (e) the current score and penalty points. The time pressure in the task is quite severe; plane fuel levels decrease in real time;
Christian D. Schunn and Lynne M. Reder
140 FLTN 233 794 892 888 420
TYPE
FUEL
*
4
*
3
3N 3s 3E 3w 2N
*
3
2s
5
PROP 747 DClO DClO 747
POSN
5
2E
->
127
PROP
*
4
402 822
E l 0
*
PROP
4 5
878
747
5
429
E l 0
*
4
Score
: 200
Landing P t s : 200 Runways
Wind
Penalty P t s : 0
DRY
: 25
-
35 knots from WEST
F l t s i n Queue: . . . . . . .. < F l > to accept
2w 1N
is 1E iw
Fig. 11. The main screen of the Kanfer-Ackerman Air Traffic Controller Task.
weather changes occur approximately every 25 seconds; and planes enter into the queue every 7 seconds. There are six rules governing this task, four of which are particularly pertinent here. First, planes must land in the wind (e.g., use a north-south runway rather than an east-west runway if the wind is coming from the north or south). Second, planes can only land from hold level 1 (the lowest level). Third, the current weather conditions and wind speed determine the runway length required by different plane types (747s always require long runways; DClOs can use short runways if runways are not icy and the wind speed is less than 40 knots; 727s can use short runways only when the runways are dry or wind speed is 0-20 knots, and PROPS can always use short runways). Fourth, only one plane at a time can occupy a runway. Each violation of any of these rules produces a 10-point penalty. The task consists of a sequence of 10-minute trials, with the total number of trials varying from study to study. Each trial begins with planes already in various hold positions and other planes in the queue. At the end of each trial, the subject is given a short, self-timed break. The next trial begins with a new screen display and the cursor at the top of the screen. The task involves three primary subtasks: (1) accepting a plane from the queue into a hold position; (2) moving planes within the hold positions; and (3) landing planes. The analyses presented here focus on the third task,
Individual Differences in Adaptivity
141
specifically on the strategy decision of landing a selected plane on the short or long runways when both are open. There are several advantages to selecting the long runway. First, the long runways are always legal for all plane types. Thus, the probability of making an error is lower. Second, the current wind speed and runway conditions need not be consulted before landing the plane (although wind direction must always be consulted). Third, the rules for landing a plane on the runways need not be retrieved. Fourth, the long runways are closer to the hold positions than the short runways and so require fewer keystrokes. The advantage of using the short runway (when it is legal) is that it keeps the long runway open for the planes that can only land on the long runway under the current wind speed and runway conditions. Because the planes require 15 s to land on a runway and only one plane can be landed on a runway at a time, subjects must maximize the use of both runways in order to maximize the total number of planes landed. In other words, the long runway is a scarce resource that should be used sparingly. The particular operational measure of runway strategies was called OpShort, which is the proportion of times a subject opted to land a plane on the short runway of all the times a plane was landed and both runways were open. This ratio is computed separately for each plane type. Only ratios for DClOs and 727s are computed because 747s can never land on the short runway and PROPs can always land on the short runway. C. A BASE-RATE MANIPULATION IN THE KA-ATC TASK 1. Methods
To measure extrinsic adaptivity, a base-rate manipulation was used in the KA-ATC task much like the base-rate manipulation in the BST domain. Specifically,OpShort success rates were manipulated by varying the proportion of 747s and PROPs in each 10-minute trial. Because 747s can only land on long runways and PROPs can always land on either runway, the proportion of 747s and PROPs drastically affect the importance of using the long runway effectively. When there are many 747s, then it is better to place the DClOs and 727s on the short runway whenever possible. By contrast, when there are few 747s, there is much less pressure to place the DClOs and 727s on the short runway. In this situation, it is better to place those planes on the long runway-landing on the short runway requires more keystrokes, knowing and accessingthe rules for when the short runway is legal, and checking the current wind and weather conditions. Thus, when there are many 747s, OpShort rates should be high, and when there are few 747s, OpShort rates should be low.
142
Christian D. Schunn and Lynne M. Reder
Subjects were given nine 10-minute trials. These nine trials were divided into three blocks of three trials each. The proportion of 747s (and PROPs) were manipulated across blocks. The proportions of 747s over the three blocks were 25%-5%-50%.5 Because the strategy adaptivity measure is defined by plane type, one plane type (DClOs) was set at a constant high level of 40% across all three blocks to insure sufficient Ns for each subject on at least one plane type. The frequency of PROPs was set to be 55% minus the frequency of 747s (i.e., 30%-50%-5%), thereby completing the manipulation of the scarcity of the long runways. To contrast how well adaptivity differences predict overall performance in the task with the predictiveness of more traditional measures, and to understand what factors were related to adaptivity, the subjects were given an individual ability battery. The individual ability battery used was a subset of the CAM4 (Cognitive Abilities Measurement) (Kyllonen, 1993, 1994, 1995) battery. The CAM battery provides a broad range of tests that are plausibly related to adaptivity in strategy use, and has been used to predict learning and performance in a large number of training environments (e.g., Shebilske, Goettl, & Regian, in press; Shute, 1993). Eleven CAM tests were used, including multiple tests of associative learning, procedural learning, processing speed, working memory, and inductive reasoning. 2. Results
a. Overall Adaptivity To examine whether the subjects, overall, were able to adapt to the base-rate manipulation in the context of a complex task, an ANOVA was conducted on OpShort with Block as a withinsubjects factor. The effect of Block was quite strong, F(2,174) = 43.0, MSE = .04, p < .0001, with subjects adapting in the predicted pattern of medium-low-high (see Fig. 12). Thus, people are able to show extrinsic adaptivity even in the context of a very complex, dynamic task. b. Individual Differences in Adaptivity Although subjects on average adapted to the base-rate manipulation, the subjects did vary significantly in their adaptivity. Only 30% of the subjects followed the full pattern of adaptivity in OpShort use across the three blocks; medium, low, high (i.e., block 3 > block 1 and block 2 < block 1 ), and only 75% adapted at all to the largest transition from block 3 to block 2 (i.e., block 3 > block 2). Another measure of adaptivity is the extent of change from blocks 2 to 3
’
A second control condition with a different order (25%-50%-5%) was used to insure that the results were not peculiar to one particular order, nor simply due to changes that would have occurred naturally as a function of practice with the task (i.e., independent of the manipulation). The control condition simply performed as expected and so will not be discussed here.
Individual Differences in Adaptivity
143
0'71 0.6
Bldckl
Blick2
Bldck3
Fig. 12. The mean proportion (and SE) OpShort within each block of three trials in the KAATC base-rate manipulation (25%-5%-50%747s). Adapted from Schunn and Reder (1997).
(i.e., the difference in mean OpShort use in those two blocks). Even in those who showed any adaptivity at all on the second transition, there was significant variability in OpShort use. A third of the adaptive participants adapted less than .2, and a third adapted .5 or more. c. Correlations with Performance and the CAM Battery Do these differences in extent and rate of adaptivity relate to task performance? Because adapting strategy use may require cognitive resources, performance on other aspects of the task may suffer. Therefore, it is not necessarily true that differences in OpShort adaptivity will be related to differences in overall task performance. Moreover, the runway landing decision is only a very small part of the overall task. To examine this relationship, extent of adaptivity as measured by the rise in OpShort from blocks 2 to 3 was regressed against mean block score. Adaptivity proved to be quite strongly correlated to overall score, r = .69, p < .0001. For comparison, the largest correlation between the 11 CAM ability measures and score was only .56. Thus, the general trait of adaptivity, as measured on only one of the myriad of strategy decisions that must be made in this complex task, is a very important predictor of overall task performance. What factors predict extent of adaptivity? To assess which measures were independent predictors of adaptivity, a stepwise regression was conducted including all 11 CAM test scores. Two measures entered into the stepwise regression (multiple 9 = .38), in order of predictiveness: working memory ( r = S 5 , p < .005), and the inductive reasoning ( r = S 3 , p < .Ol). Fig. 13 illustrates the overall pattern of correlations among CAM, adaptivity, and overall task performance.
144
Christian D. Schunn and Lynne M. Reder
Fig. 13. Best correlations in the KA-ATC base-rate manipulation between extent of adaptivity, ability measures, and overall task score.
The relationship between the CAM measures and adaptivity are interesting because they provide information about what processes underlie strategy choice and strategy adaptivity. What possible mechanisms could explain these relationships? Inductive reasoning could play a role in adaptiveness in at least two ways. First, inductive reasoning skill may be related to being able to notice shifting patterns in the environment, as in the BST task. Second, inductive reasoning might be related to being able to quickly understand the relationship between a strategy and its effect, or diagnosing when a strategy is no longer appropriate. Relatedly, higher working memory capacity may involve an increased ability to keep information, specifically base-rate information, in mind while simultaneously performing the task. Under this account, ability to retain the recent set of outcomes would predict how quickly the pattern of change can be detected, and hence how fast one could adapt. Inductive reasoning ability, in contrast, would predict whether, given a pattern, the individual understood what strategy to adopt for best performance with the new pattern.
D. MICRO-LEVEL EXTRINSIC ADAP~VITY Another kind of measure of extrinsic adaptivity focuses at a more microlevel: adaptivity to the success of the strategy in the previous attempt. Even when base-rates of success remain roughly constant, the individual must use success and failure feedback to discover which strategies and what levels of strategy use are optimal in the current task. To investigate this kind of extrinsic adaptivity, and possible individual differences therein, Schunn and Reder also reanalyzed unpublished KA-ATC data collected by Phillip Ackerman.6 In this data set, each 10-minute trial was roughly the same-what varied from trial to trial was the actions taken by the subjects and their knowledge and skills as they learned to perform the complex task. The data set also included individual ability scores from The data were taken from the Kanfer-Ackerman CD-ROM Database0 (1994).
Individual Differences in Adaptivity
145
22 ability tests that were centered around six factors (Perceptual Speed, Movement Speed, Memory, Verbal, Reasoning, and Psychomotor factors). 1. Results a. Overall Adaptivity To examine whether subjects were adaptive in their OpShort use at the micro-level, the OpShort data were analyzed as a function of whether the previous attempt to land that plane type on the short runway had been successful (i.e., had not resulted in an error). If subjects were adaptive, then they should reduce their tendency to use the short runway when that landing attempt resulted in an error previously and they should increase their tendency to use the short runway for that plane type when that action was previously successful. Only data from the first nine trials were used because error rates were very low after the ninth trial. The analyses also distinguished situations in which the short runway was legal for the selected plane type and those in which it was illegal-the definition of OpShort required only that the two runways were open. Any observed adaptivity may reflect several types of strategy shifts. For example, it may lead subjects to learn the rules more completely following an unsuccessful attempt and then be less likely to use the short runway in illegal situations. Alternatively, it may reflect a simple shift in tendency to use the short runway that would have had equal impact in legal and illegal situations. Fig. 14 suggests that this second alternative is what occurred, and ANOVA results supported this view. There were main effects of legal0.5 -
-
0.4
0.3-
/
L
+ Illegal
0
3 0
Unsuccessful Successful Previous attempt Fig. 14. Mean OpShort for legal and illegal short runway opportunities as a function of the success of the previous attempt to land that plane type. From Schunn and Reder (1997).
146
Christian D. Schunn and Lynne M. Reder
ity, F(1,42) = 76.2, MSE = .05, p < .Owl, and success of the previous attempt, F(1,42) = 20.2, MSE = .Ol,p < .0001, but no hint of an interaction F(1,42) < 1. Thus, subjects do adapt at the micro-level, producing shifts in the bias to use a strategy as a result of success and failure feedback. b. Individual Differences in Adaptivity Because evidence has been found for micro-level strategy adaptivity for OpShort, the issue of individual differences can be examined. In fact, there were significant individual differences in adaptivity, as defined by the difference in OpShort use following successes and OpShort use following failures (collapsing across legal and illegal situations). The modal adaptivity level, accounting for 23% of subjects, was -.05 to +.05 (i.e., no adaptivity). Moreover, this mode did not include the mean adaptivity level across all subjects (.13). Approximately 21% and 24% of the subjects showed adaptivity levels that were two and three or more times as high as the mean, respectively. Thus, a significant proportion of subjects showed no evidence of strategy adaptivity to success rates, whereas many subjects showed very strong strategy adaptivity to success rates.
c. Correlations with Performance and the CAM Battery How do these individual differences in strategy adaptivity relate to performance in the task? Because the primary goal (and primary determinant of score) in the KA-ATC task is landing planes, OpShort adaptivity was correlated against the mean number of planes landed. OpShort adaptivity correlated positively with planes landed, r = S O , p < .02. By contrast, of the 22 individual difference tests administered by Ackerman, the 4 best predictors of planes landed correlated in the .44 to .47 range. Using a stepwise regression on planes landed with the 22 individual difference tests and the Adaptivity measure as possible predictors, only two independent factors emerged (multiple 9 = .29): OpShort Adaptivity /3 = 11.4, p < .05, and the Patterns test /3 = 0.29, p < .05. Thus, adaptivity appears to predict performance directly and not through indirect correlations with other determinants of performance. These results provide further evidence that the individual differences in adaptivity are a very important performance predictor and they demonstrate that the individual differences are systematic and cannot be attributed to noise. Another issue of interest is whether adaptivity can be predicted from the psychometric ability tests. When the 22 psychometric scores were entered into a stepwise regression predicting OpShort adaptivity, only one factor entered: Letter Sets, r = .45, p < .02, a test that loads heavily on reasoning and general intelligence. The results are similar to those found in the KA-ATC base-rate manipulation study in that they also point to reasoning ability as an important correlate of strategy adaptivity. The overall patterns of correlations among adaptivity, ability scores, and task perfor-
Individual Differences in Adaptivity
147
Fig. 15. Best correlations in the KA-ATC micro-level adaptivity study between extent of adaptivity, ability measures, and number of planes landed.
mance are illustrated in Fig. 15. These correlations provide another piece of evidence that the individual differences in adaptivity were not simply due to chance variation. E. DISCUSSION Overall, subjects were able to demonstrate extrinsic adaptivity, even in a complex dynamic task. However, across several KA-ATC studies, in both college (Ackerman data set) and noncollege populations (base-rate manipulation data set), using both macro- and micro-level measures, there were significant individual differences in extrinsic adaptivity, with many individuals showing no evidence of adaptivity. By measuring adaptivity on one small strategy decision in a complex task, it was possible to predict overall task performance with greater accuracy than with all of the traditional individual difference measures that were tried. These results suggest that strategy adaptivity differences must have manifested themselves in many other strategic aspects of the overall task. The results also suggest that strategy adaptivity is a very important component of performance in complex, dynamic tasks. These studies using the KA-ATC task also found that inductive reasoning and working memory capacity were correlated with individual differences in extrinsic adaptivity. These correlations provide new insights into the processes of strategy choice and strategy adaptivity. For example, they suggest that strategy adaptivity involves noticing patterns and keeping significant amounts of information in working memory-two features that are not part of current models of strategy choice (e.g., Logan, 1988;Lovett & Anderson, 1996; Reder, 1987; Schunn et al., 1997; Siegler & Shipley, 1995).
V. General Discussion Previous accounts of individual differences in performance have tended to focus on either parametric differences in an underlying cognitive architec-
148
Christian D. Sehunn and Lynne M. Reder
ture (e.g., attentional capacity, working memory capacity, g, etc.) or strategy differences. This chapter has explored a possibility that unifies the previous two approaches: that there is an architectural parameter that controls strategy use, or more specifically,that individuals may have the same strategies but are differentially able to select the most appropriate strategy for the given situation. Reder and Schunn (in press; Schunn & Reder, 1997) explored this idea within the KA-ATC task. This chapter extends this work both in the sophistication of the analyses and the breadth of the tasks: examination of individual differences in three very different types of tasks (rapid arithmetic, simple problem solving, and complex problem solving) produced evidence of significant variability in strategy adaptivity in all three cases. Although the individual differences were all investigated among normal, nonexpert, adult populations, these kinds of adaptivity differences might also be studied in all the other areas of individual differences (e.g., child development, aging, expertise, giftedness, schizophrenia, etc.). The prediction from the findings presented here is that populations that display working memory or inductive reasoning differences will also display adaptivity differences. In fact, many of these other areas of individual differences have already hypothesized group differences in working memory, and thus there are many places to test this prediction. Moreover, the differences in performance due to adaptivity differences is likely to be most pronounced in dynamic tasks. Methodologically,this chapter described five different methods for establishing the meaningfulness of individual differences in strategy adaptivity. First, observed individual differences were compared to chance levels of variation due to sampling noise, as predicted by a Monte Carlo simulation. Second, the stability of individual differences was measured across consecutive days in the same task. Third, the individual differences were shown to be strongly correlated to overall task performance, even in situations in which base-rates of success are not directly manipulated. Fourth, the individual differences were compared to existing ability measures and found to be correlated to several of them. Finally, many correlational analyses were conducted to show that the individual differences were not due to simple artifacts (e.g., floor and ceiling effects in strategy use, overall skill in the task domain, and speed-accuracy tradeoffs). Two kinds of strategy adaptivity were considered in the arithmetic domain: intrinsic and extrinsic. Although there appeared to be individual differences in both kinds of adaptivity, the intrinsic adaptivity differences appeared to be attributable to various artifacts-differential task proficiency, large strategy selection biases, and speed-accuracy tradeoffs. The extrinsic adaptivity differences did not have this problem, and were further
Individual Differences in Adaptivity
149
established in the BST and KA-ATC tasks. Of course, individual differences in intrinsic adaptivity have not been ruled out altogether-it still might exist in other population comparisons or become more apparent in other tasks. Lovett and Schunn (1997) found that although people can learn not to pay attention to certain features (i.e., intrinsic adaptivity is malleable), they also found significant variation across individuals in intrinsic adaptivity under a given payoff schema. The distinction between intrinsic and extrinsic adaptivity is an important one, and future attempts to measure individual differences in adaptivity should continue to distinguish these two types of adaptivity. There are many conceptions of adaptivity to found in the psychological literature. This chapter has focused on adaptivity in strategy use. It is an open question as to how strategy adaptiveness might relate to other kinds of adaptiveness. For example, it may be correlated with individual differences in the ability to adapt to instructions (Reder, 1987; Shebilske, Goettl, & Regian, in press), in the ability to select and change representations (Lovett & Schunn, 1997; Schunn & Klahr, 1996; Schunn & Lovett, 1996), or in the ability to adaptively control attention (Gopher, 1982,1996; Gopher & Kahneman, 1973). Another question is how strategy adaptivity relates to metacognition. Although the metacognitive view is related to the strategy adaptiveness view explored here, there are several important differences. The metacognitive view (e.g., Case, 1985; Flavell, 1979; Kuhn, 1988; Sternberg, 1985) postulates that some individuals use less sophisticated strategies because they have poor metacognitive knowledge of why and when different strategies are effective. Although this view seems quite plausible, empirical research has found that there is at best a weak relationship between metacognitive knowledge and the adaptiveness of strategy selections (e.g., Cavanaugh & Perlmutter, 1982; Schneider & Pressley, 1989). Base-rate adaptivity and individual differences in base-rate adaptivity has also been studied in a much more explicit kind of task-text-based probability problems that explicitly present both base-rate and case-specific information and require subjects to predict the probability that one alternative will succeed. Studies using this paradigm have typically found baserate neglect (e.g., Ginossar & Trope, 1987; Lyon & Slovic, 1976;Tversky & Kahneman, 1973). That is, people’s predictions frequently do not adequately take into account the overall probabilities (or base-rates) of the various alternatives, and overweight the influence of case-specific information. However, various manipulations of the task wording can lead to greater use of base-rate information (cf. Koehler, 1996, or Lovett & Schunn, 1997 for a review). Most importantly for this chapter, Stanovich and West (1998) studied individual differences in base-rate neglect and other reasoning
150
Christian D. Schunn and Lynne M. Reder
fallacies observed on text-based problems, and found that cognitive ability measures explain some of the variance of people’s performance on these tasks. It may be interesting to explore how these individual differences in the ability to reason explicitly about base-rates may be related to individual differences in extrinsic strategy adaptivity. A final remaining issue is the cross-task stability of individual differences in strategy adaptivity. More generally, the issue is whether the individual differences in strategy adaptivity are specific to particular tasks or types of tasks or whether they are general across many types of tasks. This chapter presented data showing stability within a task over time, similar kinds of individual differences in three very different tasks, and correlations with standard ability tests, all of which suggest that cross-task stability would be found if it were measured. However, no direct data were presented on stability across tasks. Although this issue is not yet resolved, it is important to note that the interestingness of individual differences in adaptivity does not hinge on cross-task stability: Even if the individual differences were specific to particular tasks, they nevertheless remain important features of task performance, and they provide insight into the mechanisms of strategy choice and strategy adaptivity.
REFERENCES Ackerman, P. L. (1988). Determinants of individual differences during skill acquisition: Cognitive abilities and information processing. Journal of Experimental Psychology: General, 117(3), 288-318. Ackerman, P. L. (1989). Individual differences and skill acquisition. In P. L. Ackerman, R. J. Sternberg, & R. Glaser (Eds.), Learning and individual differences (pp. 165-217). New York: Freeman and Company. Ackerman, P. L., & Kanfer, R. (1994). CD-ROM Database, Data Collection Program, and Playback Program. Carver, C . S., & Scheier, M. F. (1992). Perspectives onpersonality (2nd ed.). Needham Heights, MA: Simon & Schuster. Case, R. (1985). Intellectual development: A systematic reinterpretation. New York: Academic Press. Cavanaugh, J. C., & Permutter, M. (1982). Metamemory: A critical examination. Child Development, 53, 11-28. Chi, M. T. H., Bassok, M., Lewis, M., Reimann, P., & Glaser, R. (1989). Self-explanations: How students study and use examples in learning to solve problems. Cognitive Science, 13, 145-182. Chi, M. T. H., Feltovich, P. J., & Glaser, R. (1981). Categorization and representation of physics problems by experts and novices. Cognitive Science, 5, 121-152. Dunlosky, J., & Connor, L. T. (1997). Age differences in the allocation of study time account for age differences in memory performance. Memory & Cognition, 25(5). 691-700. Ericsson, K. A., & Polson, P. G. (1988). A cognitive analysis of exceptional memory for restaurant orders. In M. T. H. Chi, R. Glaser, & M. J. Farr (Eds.), The nature ofexpertise. Hillsdale, NJ: Erlbaum.
Individual Differences in Adaptivity
151
Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cognitivedevelopmental inquiry. American Psychologist, 34, 906-91 1. Fry, A. F., & Hale, S. (1996). Processing speed, working memory, and fluid intelligence: Evidence for a developmental cascade. Psychological Science, 7(4), 237-241. Geary, D. C., & Brown, S. C. (1991). Cognitive addition: Strategy choice and speed-ofprocessing differences in gifted, normal, and methematically disabled children. Developmental Psychology, 27(3), 398-406. Ginossar, Z., & Trope, Y. (1987). Problem solving in judgment under uncertainty. Journal of Personality and Social Psychology, 52, 464-473. Gopher, D. (1982). A selective attention test as a predictor of success in flight training. Human Factors, 24(2), 173-183. Gopher, D. (1996). Attention control: Explorations of the work of an executive controller. Cognitive Brain Research, 5, 23-38. Gopher, D., & Kahneman, D. (1973). Individual differences in attention and the prediction of flight criteria. Perceptual and Motor Skills, 33, 1335-1342. Greenfield, P. M., & Lave, J. (1982). Cognitive aspects of informal education. In D. A. Wagner & H. W. Stevenson (Eds.), Cultural perspectives on child development. San Francisco, CA: Freeman. Hunt, E., Joslyn, S., & Sanquist, T. (1996). Evaluating individual differences in response to emergency situations. ONR Tech Report. Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99(1), 122-149. Kail, R. (1988). Development functions for speeds of cognitive processes. Journal of Experimental Child Psychology, 45, 339-364. Kanfer, R., & Ackerman, P. L. (1989). Motivation and cognitive abilities: An integrative/ aptitude-treatment interaction approach to skill acquisition. Journal of Applied Psychology-Monograph, 74, 657-690. Koehler, J. J. (1996). The base rate fallacy reconsidered Descriptive, normative, and methodological challenges. Behavioral and Brain Sciences, 19, 1-53. Kuhn, D. (1988). Cognitive development. In M. H. Bornstein & M. E. Lamb (Eds.), Developmental psychology: An advanced textbook (2nd ed.). Hillsdale, NJ: Erlbaum. Kyllonen, P. C. (1993). Aptitude testing inspired by information processing: A test of the four-sources model. The Journal of General Psychology 120(3), 375-405. Kyllonen, P. C. (1994). Cognitive abilities testing: An agenda for the 1990s. In M. G. Rumsey, C. B. Walker, & J. H. Harris (Eds.), Personnel Selection and Classification. Hillsdale, NJ: Erlbaum. Kyllonen, P. C. (1995). CAM: A theoretical framework for cognitive abilities measurement. In D. Detterman (Ed.), Current topics in human intelligence: Volume IV, Theories of intelligence. Norwood, NJ: Ablex. Larkin, J. H., McDermott, J., Simon, D., & Simon, H. (1980). Expert and novice performance in solving physics problems. Science, 208, 140-156. LeFevre, J., Greenham, S., & Waheed, N. (1993). The development of procedural and conceptual knowledge in computational estimation. Cognition & Instruction, II(2), 95-132. LeFevre, J., Sadesky, G. S., & Bisanz, J. (1996). Selection of procedures in mental addition: Reassessing the problem-size effect in adults.Journal of Experimental Psychology: Learning, Memory, and Cognition, 22, 216-230. Lemaire, P., & Siegler, R. S. (1995). Four aspects of strategic change: Contributions to children’s learning of multiplication. Journal of Experimental Psychology: General, 124(1), 83-97.
152
Christian D. Schunn and Lynne M. Reder
Logan, G. D. (1988). Toward an instance theory of automatization. Psychological Review, 95(4), 492-527. Lovett, M. C., & Anderson, J. R. (1996). History of success and current context in problem solving: Combined influences on operator selection. Cognitive Psychology, 31 168-217. Lovett, M. C., & Schunn, C. D. (1997). Task representations, strategy variability and baserate neglect. Paper submitted for publication. Lovett, M. C., Reder, L. M., & Lebiere, C. (1996). Modeling individual differences in a digit working memory task. Paper presented at the International Conference on Memory. Padova, Italy. Lovett, M. C., Reder, L. M., & Lebiere, C. (in press). Modeling working memory in a unified architecture: The ACT-R perspective. To appear in A. Miyake & P. Shah (Eds.), Models of Working Memory. Cambridge University Press. Luchins, A. S. (1942). Mechanization in problem solving. Psychological Monographs, 54. Lyon, D., & Slovic, P. (1976). Dominance of accuracy information and neglect of base rates in probability estimation. Acta Psychologica, 40, 287-298. Miller, P. H., & Seier, W. L. (1994). Strategy utilization deficiencies in children: When, where, and why. In H. W. Reese (Ed.), Advances in child development and behavior (Vol. 25). New York: Academic Press. Miyake,A., Carpenter, P. A., & Just, M. A. (1995). Reduced resources and specific impairments in normal and aphasic sentence comprehension. Cognitive Neuropsychology, 12,651-679. Nelson, T. O., & Narens, L. (1990). Metamemory: A theoretical framework and some new findings. In G. H. Bower (Ed.), The psychology of learning and motivation (pp. 1-45). New York: Academic Press. Reder, L. M. (1989). The role of elaborations in memory for prose. Cognitive Psychology, 11, 221-234. Reder, L. M. (1982). Plausibility judgments versus fact retrieval: Alternative strategies for sentence verification. Psychological Review, 89, 250-280. Reder, L. M. (1987). Strategy selection in question answering. Cognitive Psychology, 19, 90- 137. Reder, L. M., & Ritter, F. E. (1992). What determines initial feeling of knowing? Familiarity with question terms, not with the answer. Journal of Experimental Psychology: Learning, Memory, & Cognition, 18, 435-451. Reder, L. M., & Schunn, C. D. (in press). Bringing together the psychometric and strategy worlds: Predicting adaptivity in a dynamic task. To appear in D. Gopher & A. Koriat (Eds.), Cognitive regulation of performance: lnteractiori of theory and application. Attention and Performance XVll. Reder, L. M., Wible, C., & Martin, J. (1986). Differential memory changes with age: Exact retrieval versus plausible inference.Journal of Experimental Psychology: Learning, Memory, & Cognition, 12(1), 72-81. Saccuzzo, D. P., Johnson, N. E., & Guertin, T. L. (1994). Information processing in gifted versus nongifted African American, Latino, Filipino, and White children: Speeded versus nonspeeded paradigms. Intelligence, I9(2), 219-243. Salthouse, T. A. (1994). Aging associations: Influence of speed on adult age differences in associative learning. Journal of Experimental Psychology: Learning, Memory, & Cognition, 20(6), 1486-1503. Schneider, W., & Pressley, M. (1989). Memory development between 2 and 20. New York: Springer-Verlag. Schooler, C., Neumann, E., Caplan, L. J., & Roberts, B. R. (1997). A time course analysis of Stroop interference and facilitation: Comparing normal individuals and individuals with schizophrenia.Journal of Experimental Psychology: General, 126(l), 19-36.
Individual Differences in Adaptivity
153
Schunn, C. D., & Klahr, D. (1996). The problem of problem spaces: When and how to go beyond a 2-space model of scientific discovery. Symposium talk presented at The 18th Annual Conference of the Cognitive Science Society (pp. 25-26). Schunn, C. D., & Lovett, M. L. (1996). Representation and individual differences in baserate neglect. Poster presented at 37thAnnual Meeting of the Psychonomic Society, Chicago, IL (November, 1996). Schunn, C. D., & Reder, L. M. (1997). Another Source of Individual Differences: Strategy Adaptivity to Changing Rates of Success. Paper submitted for publication. Schunn, C. D., Reder, L. M., Nhouyvanisvong,A., Richards, D. R.,& Stroffolino,P. J. (1997). To calculate or not calculate: A source activation confusion (SAC) model of problemfamiliarity’s role in strategy selection. Journal of Experimental Psychology: Learning, Memory, & Cognition, 23(1), 3-29. Schwartz, B. L., & Metcalfe, J. (1992). Cue familiarity but not target retrievability enhances feeling-of-knowingjudgments. Joural of Experimental Psychology-Learning, Memory, & Cognition, 18,1074-1083. Shapira, N., & Kushnir, T. (1985). Aging and information seeking: Patterns in sampling of sucrose solutions. International Journal of Aging & Human Development, 20(2) 103-1 12. Shebilske, W., Geottl, B., & Regian, J. W. (in press). Individual and group protocols for training complex skills in laboratory and applied settings. To appear in D. Gopher & A. Koriat (Eds.), Cognitive regulation of performance: Interaction of theory and application. Attention and Performance XVII. Shute, V. J. (1993). A macroadaptive approach to tutoring. Journal of Artijical Intelligence in Education, 3(1), 61-93. Siegler, R. S. (1983). Five generalizationsabout cognitive development.American Psychologist, 38, 263-217. Siegler, R. S . (1987). Strategy choices in subtraction. In J. Sloboda and D. Rogers (Eds.), Cognitive processes in mathematics, (pp. 81-106). Oxford Oxford Books. Siegler, R. S. (1988). Strategy choice procedures and the development of multiplication skill. Journal of Experimental Psychology: General, 117, 258-275. Siegler, R. S. (1996). Emerging minds: Theprocess of change in children’s thinking. New York: Oxford University Press. Siegler, R. S., & Jenkins, E. (1988). How children discover new strategies. Hillsdale, NJ: Erlbaum. Siegler, R. S., & Lemaire, P. (1997). The choicelno-choice method: Applications to assessing the adaptiveness of older and younger adults strategy choices in multidigit multiplication. Journal of Experimental Psychology: General, 126, 71-92. Siegler,R. S., & Shipley, C. (1995). Variation, selection, and cognitive change. In G. Halford & T. Simon (Eds.), Developing cognitive competence: New approaches to process modeling, (pp. 31-76). New York: Academic Press. Siegler, R. S., & Shrager, J. (1984). Strategy choices in addition and subtraction: How do children know what to do. In C. Sophian (Ed.), Origins of cognitive skills, (pp. 229-293). Hillsdale, NJ: Erlbaum. Snow, R. E., Kyllonen, P. C., & Marshalek, B. (1984). The topography of ability and learning correlations. In R. Sternberg (Ed.), Handbook of human intelligence. New York: Cambridge University Press. Spearman, C. (1904). The proof and measurement of association between two things. American Journal of Psychology, 15, 72-101. Stanovich, K. E., & West, R. F. (1998). Who uses base rates and P(D/-H)? An analysis of individual differences. Memory & Cognition, 26, 161-179.
154
Christian D. Schunn and Lynne M. Reder
Sternberg, R. J. (1977). Intelligence, informationprocessing,and analogical reasoning. Hillsdale, N J Erlbaum. Sternberg, R. J. (1985). Beyond IQ: A triarchic theory of human intelligence. New York: Cambridge University Press. Sternberg, R. J., & Grigorenko, E. L. (1997). Are cognitive styles still in style? American Psychologist, 52(7), 700-712. Swets, J. A. (1986a). Form of empirical ROCs in discrimination and diagnostic tasks: Implications for theory and measurement of performance. Psychological Bulletin, 99, 181 -198. Swets, J. A. (1986b). Indices of discriminationor diagnosticaccuracy:Their ROCs and implied models. Psychological Bulletin, 99, 100-1 17. Tversky, A., & Kahneman, D. (1982). Evidential impact of base rates. In D. Kahneman, P. Slovic, & A. Tversky, (Eds.), Judgment under uncertainty: Heuristics and biases. Cambridge University Press. Wagner, D. A. (1978). Memories of Morocco: The influence of age, schooling and environment on memory. Cognitive Psychology, 10, 1-28. Zbrodoff, N. J. (1995). Why is 9 plus 7 harder than 2 plus 31 Strength and interference as explanations of the problem-size effect. Memory & Cognition, 23(6), 689-700.
GOING WILD IN THE LABORATORY Learning about Species Typical Cues Michael Domjan
I. Introduction The twentieth century has witnessed the vigorous pursuit of two grand traditions in the study of behavior, behavioral psychology and ethology. Although both traditions originated in the work of Darwin, they have progressed pretty much independently of one another. Behavioral psychology, as exemplified by the work of Thorndike, Pavlov, Hull, and Skinner, has been concerned with discovering general laws of learning through detailed investigation of a few laboratory preparations. The experimental preparations (e.g., a rat pressing a response lever for food) were designed to reveal universal features of behavior rather than species specific solutions to unique ecological problems. In contrast, ethology has been concerned with how organisms behave in their natural habitat, how species differ in their solutions to the diverse challenges of living, and how those species differences may have evolved. These concerns have led ethologists to focus on species typical responses rather than general behavioral processes. To the extent that learning has attracted the attention of ethologists, they have investigated specialized forms of learning, such as imprinting or the learning of bird song, rather than general-process classical and instrumental conditioning. Behavioral psychology and ethology managed to progress with few skirmishes between them until the 1960swhen a number of striking phenomena THE PSYCHOLOGY OF LEARNING AND MOTIVATION, VOL. 38
155
Copyright B 1998 hy Academic Press. All rights of reproduction in any form reserved. 0079-7421/98 $25.00
156
Michael Domjan
(such as taste-aversion learning) were discovered that seemed to challenge the general-process approach of behavioral psychology. These phenomena were variously called “adaptive specializations in learning” and “biological constraints on learning” (Hinde & Stevenson-Hinde, 1973; Seligman & Hager, 1972; Rozin & Kalat, 1971; Shettleworth, 1972). The phenomena seemed to reflect the intrusion of species typical responses and specialized adaptations into learning experiments. Efforts to dismiss constraints on learning as artifactual or due to improper controls (Bitterman, 1975) were not successful.Therefore, some kind of accommodation between behavioral psychology and ethology had to take place. But, it was not clear how. Shettleworth (1983) suggested a peaceful coexistence, pointing out that behavioral psychology and ethology contribute different but complementary information about behavior. Behavioral studies serve to identify mechanisms of learning, whereas ethological investigations are more apt to reveal the biological functions of behavior. After reviewing the extensive literature on constraints on learning (in volume 13 of this annual series), I advocated modifying the level at which general theories are formulated without abandoning the general-process approach (Domjan, 1983). In a companion paper, Galef and I suggested that the general-process approach be supplemented by an ecologically informed comparative methodology (Domjan & Galef, 1983). A more radical solution was advocated by Johnston (1981), who proposed abandoning traditional laboratory studies of learning altogether. He suggested that new studies of learning be rooted in naturalistic observations. According to Johnston, the ecological validity of a learning phenomenon has to be established through field research before the phenomenon is brought into the laboratory for examination (see also Miller, 1985). About 15 years have passed since these various approaches were proposed to deal with the tensions between behavioral psychology and ethology. During the intervening years, the tension has subsided as both camps have developed more respect for each other’s work. The pursuit of general mechanisms of learning has continued unabated, as have ethological investigations of behavior. Although much of the recent progress in each field has occurred independently of the other, there have also been some tentative but clear exceptions to that. We have been reminded that conventional learning preparations were in fact designed with careful attention to the species typical responses of the rats and pigeons that commonly serve in these experiments (Hailman, 1985; Timberlake, 1990). In addition, some concerted efforts have been made to integrate behavioral psychology with ethology. These efforts have occurred at the level of theory, at the level of the problems that are selected for investigation, and at the level of methodology.
Learning about Species Typical Cues
157
At the level of theory, concepts and theoretical approaches that have been prominent in ethology are turning up with increasing frequency in analyses of data from learning laboratories. Hollis (1982, see also Hollis, 1997) provided an analysis of Pavlovian conditioning in terms of its adaptive significance and biological function. The ethological distinction between appetitive and consummatory behavior has become the focus of some learning investigations (e.g., Hillard, Domjan, Nguyen, & Cusato, 1998),and ethological theories of optimal foraging have invaded analyses of laboratory schedules of reinforcement (Shettleworth, 1988). Perhaps the most comprehensive effort to integrate ethology and behavioral psychology at the level of theory is Timberlake’s behavior systems approach to the analysis of how animals learn about food (Timberlake, 1983, 1994; Timberlake & Lucas, 1989). The behavior systems approach views feeding as consisting of a series of responses (general search, focal search, food handling and consumption) organized by a hierarchy of control mechanisms (modes and modules). Different components of the system are sensitive to different temporal and stimulus parameters. These factors in turn determine how learning is manifest in the feeding system. The basic concepts of the behavior systems approach have been also extended to analyses of learning related to defensive behavior (Fanselow, 1994, 1997) and reproduction (Domjan, 1997). The integration of behavioral psychology with ethology is also evident in the learning phenomena laboratory investigators are electing to examine. Increasingly, behavioral psychologists are turning their attention to problems such as foraging, spatial navigation, learning and memory in foodcaching birds, and sexual behavior (Domjan & Holloway, 1998; Gallistel, 1990; Kamil & Sargent, 1981; Sherry, 1988; Shettleworth, 1988; Shettelworth & Krebs, 1982), topics that were previously investigated primarily by ethologists. The third level at which behavioral psychology is becoming integrated with ethology is at the level of methodology. Ethological methods are cropping into behavioral psychology both in how behavior is being measured and in the stimuli that are used in the experiments. Instead of relying on automated recording of isolated responses, behavioral psychologists are recognizing the value of directly observing their subjects and recording a variety of their activities (Timberlake & Silva, 1994). And, in contrast to the reductionism exemplified by the Skinner box, some investigators have developed experimental situations that attempt to replicate in the laboratory some of the complexities of an animal’s natural environment. Mellgren (1982), for example, distributed feeding stations on the tops of poles in a large room to simulate a foraging environment, and Kamil and Balda (1985)
158
Michael Domjan
designed a floor with 180 potential food storage sites to study food-caching in Clark’s nutcrackers. The focus of the present chapter is on methodological integration of behavioral psychology and ethology. I will review experiments conducted primarily in my laboratory that have used species typical cues rather than arbitrary lights and tones as conditioned stimuli. The experiments illustrate that such stimuli can be profitably used in laboratory studies of learning. Because species typical stimuli in the laboratory are the same as those the animals encounter in their natural environment, our hope is that the learning mechanisms and phenomena illuminated by these experiments can be easily generalized to the wild. Which features of the ecological niche of a species are relevant to foraging or food-caching can be difficult to identify and replicate in the laboratory because these behaviors occur over a large and complex terrain in the wild. To avoid such complications, we have turned our attention to learning about social situations. Bringing natural stimuli into the laboratory is fairly convenient in instances of social behavior-behavior involving the interactions of an animal with other members of its species (conspecifics). Animals interact with conspecifics in the course of feeding, territorial defense, parental behavior, and, of course, sexual behavior. Although intraspecific social behavior is controlled by a variety of inanimate environmental cues, social behavior also involves reacting to the stimuli that one organism provides another. In fact, social cues are an essential feature of social behavior. Social cues can be easily brought into the laboratory by arranging for one organism to serve as the conditioned stimulus (CS) in the control of the behavior of another organism.
II. Use of a Conspecific as a Conditioned Stimulus Perhaps the first study in which one organism was used as the CS for another was performed by Timberlake and Grant (1975). They investigated appetitive conditioning with food as the unconditioned stimulus (US). Mildly hungry laboratory rats were periodically presented with a pellet of food. Rather than signaling the food with a brief tone or a light, as is customary in conventional learning experiments, Timberlake and Grant devised an apparatus that permitted them to introduce a stimulus rat into the experimental chamber just before the subject received its food pellet. With repeated pairings of the stimulus rat with food, the experimental subjects came to perform conditioned responses directed towards the stimu-
Learning about Species Typical Cues
159
lus rat. The conditioned responses involved approaching and nosing the stimulus rat. At the time of its publication, the Timberlake and Grant experiment was important because it provided evidence against Pavlov’s stimulussubstitution hypothesis. The conditioned responses of approaching and nosing the stimulus rat were different from the unconditioned responses of biting and chewing that were elicited once the food US was presented. The experiment also provided a method in which one animal served as the CS for another animal. On first impression, it may seem a bit strange to use an animal as a CS in a food-conditioning experiment. However, subsequent studies have shown that such a paradigm can provide valuable information about the role of social cues in the feeding system of various species (Timberlake, 1983). The Timberlake and Grant study also provided a technique that can be applied to the study of learning in other forms of social behavior. My research collaborators and I have been using the Timberlake and Grant technique in some of our experiments on sexual conditioning in the domesticated quail, Coturnix juponicu. In a sexual conditioning paradigm, a CS is presented to an experimental subject shortly before the subject receives the opportunity to copulate with a conspecific. In the tradition of laboratory studies of Pavlovian conditioning, early studies of sexual conditioning employed arbitrary auditory and visual cues as CSs. Farris (1967), for example, presented a buzzer to male Japanese quail just before permitting them to copulate with a female. Within about 20 conditioning trials, the males were observed to perform conditioned courtship responses to the buzzer. (For a similar study with a red light CS conducted with the blue gourami, a fish species, see Hollis, Cadieux, & Colbert, 1989). In the wild, sexual encounters are not likely to be signaled by an arbitrary buzzer or light. The relevant cues may be special calls, sexual odors, or displays performed by the potential sexual partner. Such stimuli can be difficult to identify and replicate in the laboratory. However, in all sexual encounters, one participant first detects the other at a distance before the two approach one another and get close enough to make physical contact and copulate. This sequence can be modeled in the laboratory by presenting a potential sexual partner as the CS shortly before permitting the experimental subject to copulate with the stimulus animal. We first used such a procedure in studies of the development of social proximity behavior in male Japanese quail. In one experiment (Nash, Domjan, I% Askins, 1989, Experiment l), each male quail was housed in a large experimental chamber (91 cm wide, 122 cm deep, and 76 cm high), with a female housed in a smaller adjoining
160
Michael Domjan
compartment (see Fig. 1). Ordinarily the male and female could not see one another. However, the wall separating the two compartments had both a window and a door that could be opened during conditioning trials. At the start of each conditioning trial, the window (a vertical slit 1.3 cm wide and 15.2 cm high) was first opened allowing the male to see the female. After about 25 min, the door between the male and female compartments was also opened, and the male invariably copulated with the female. No more than one conditioning trial was conducted each day. The development of conditioned responding was measured by recording how much time the male spent near the window to the female compartment when the window was opened on each conditioning trial. As conditioning proceeded, the subjects spent increasing amounts of time near the window. The asymptote of learning was reached within 10 trials. At asymptote, the subjects were spending an average of 75% of their time near the open window. This behavior is illustrated in Fig. 2. Once the social proximity response has been acquired, it persists for two weeks (and perhaps longer) without further sexual reinforcement (Domjan & Hall, 1986). Various techniques, including tests with taxidermic models, have shown that the behavior is primarily controlled by the visual features of the female; the female’s calls or movements are not necessary to stimulate the male’s approach to the window (Domjan & Hall, 1986; Domjan & Nash, 1988). The stability and reliability of the social proximity behavior has encouraged its use in both hormonal and genetic analyses of
Fig. 1. Schematic of the apparatus used to the study the acquisition of social proximity behavior in male Japanese quail.
Learoing about Species Typical Cues
161
Fig. 2. A male quail looking through a window at a female on the other side, after visual access to the female had been paired with copulatory opportunity. (Drawing by Susanna Douglas, based on a photograph.)
sexual anticipatory behavior (Balthazart, Reid, Absil, Foidart, & Ball, 1995; Domjan, 1987; Hamilton, Domjan, & Mills, 1998). OF LEARNING A. MECHANISMS
In the study described above, the window through which the males could see a female was located on the door that provided them with copulatory access to the female. Furthermore, the door was opened after the males received visual access to the female and approached and remained near the window. These factors permitted both spatial learning and instrumental conditioning. Opening the door to the female’s compartment after the males approached and remained near the window could have instrumentally reinforced the window-approach response and also permitted the males to learn where the female with whom they could copulate was located. Although both spatial learning and instrumental conditioning are plausible explanations, both were ruled out by a study conducted by Domjan, Akins, and Vandergriff (1992, Experiment 1).In this experiment, a female
162
Michael Domjan
compartment was attached to both the left and the right walls of the experimental chamber, making the apparatus symmetrical. Each female compartment had a door and a narrow window that opened to the male’s area, as in the previous studies. On each training day, the window to one of the female compartments was opened for about 25 min, and some time later, the door to one of the female compartments was also opened permitting the male to copulate with that female. In these respects this experiment was similar to the first one I described. However, there were two important differences. The first change was that copulatory opportunity was always provided 2-3 h after the window had been opened each day. This was done to make sure that approaching and remaining near the window would not be instrumentally reinforced by the opportunity to copulate with a female. (Instrumental conditioning is unlikely if the reinforcer is delayed 2-3 h after the instrumental response.) The second change involved varying the relationship between the side where the window was opened and where the female was released. This change was designed to test the role of spatial learning. Two independent groups of males were compared. Each male consistently had either the window on the right side or the left side opened during each training trial. For group Same, the female door that was opened each day was on the same side as the window that had been opened earlier that day. In contrast, for group Opposite, the female door that was opened was always on the side opposite where the window had been opened. Thus, the procedure for group Same permitted subjects to associate the window with the location of the female with which they copulated. The procedure for group Opposite did not permit such spatial learning. The results of the experiment are summarized in Fig. 3. We recorded how often the subjects were near the opened window and how often they were near the closed window. Note that the subjects hardly spent any time near the closed window. This was the case even for group Opposite, who received access to a female on the closed window side. Both groups spent much more time near the opened window, irrespective of whether that window was on the same side as the door that provided them with access to a female or on the opposite side. Statistical analyses failed to reveal any significant differences between the two groups. However, the subjects increased their time near the opened window during successive trials of the experiment. These results provide no evidence that the side on which males received opportunity to copulate with a female influenced their acquisition of the social proximity response. Thus, the results failed to support a spatial learning interpretation. The results are also difficult to explain in terms of instrumental reinforcement of the social proximity behavior be-
Learning about Species Typical Cues
s
163
lool
8c 805 ‘
3
60-
z F
. 40-
+ Opposite open Sameclosed OppGsite Closed
* 20
0 5
10
15
20
Trials Fig. 3. Mean percentage of time male quail spent near windows located on the left and right side of the male’s chamber. One of the windows was opened each day, while the other remained closed. For independent groups, the opened window was on the Same side as the door that was used to provide the male with access to a female or on the Opposite side of the male’s chamber. (Redrawn from Domjan et al., 1992, reprinted by permission.)
cause sexual reinforcement was provided 2-3 h after the social proximity behavior occurred, and the window was closed during the delay interval. Given that we have ruled out spatial learning and instrumental conditioning, what is responsible for the acquisition of the social proximity behavior? Two mechanisms of learning appear to be involved. The first is direct Pavlovian conditioning of species typical cues by sexual reinforcement. Each time a male copulates with a female, the male first encounters the visual cues of a female at a distance. Exposure to those cues is then paired with more proximal tactile stimulation and sexual reinforcement as the male and female come together and copulate. Through these experiences, visual cues of the female are presumed to become associated with sexual reinforcement, and this increases subsequent responsiveness to those female cues. The second mechanism involves potentiation of the effectiveness of species typical cues in eliciting components of sexual behavior. This potentiation can be triggered by exposure to sexually conditioned arbitrary or contextual stimuli. I will discuss studies involving species typical cues as CSs in the following sections and will address the potentiation of responding to species typical cues in Section IV. B. LEARNING TO DISCRIMINATE THE SEXOF CONSPECIFICS If the acquisition of social proximity behavior reflects a Pavlovian association between female cues and sexual reinforcement, then this phenomenon should exhibit common features of Pavlovian conditioning such as stimulus
164
Michael Domjan
generalization, discrimination learning, and discrimination reversal learning. We have examined these issues in the context of investigating how male quail learn to distinguish between males and females of their own species-how they learn to discriminate the sex of conspecifics. Male and female quail are not easy to tell apart. Males are a bit smaller than females. The feathers on their torso are similar, but the head and neck feathers of males are darker and less spotted than the head and neck feathers of females. Perhaps because of the modest morphological differences between the sexes, male quail have been found to respond similarly to other males and females in both copulatory and aggressive behavior tests (Sachs, 1966; Schlinger, Palter, & Callard, 1987; Wilson & Bermant, 1972). Lack of differential responding to the sex of conspecifics has been also observed in other avian and mammalian species, usually related to limited social experience (Noble & Vogt, 1935; Nyby, Bigelow, Kerchner, & Barbehenn, 1983). 1. Stimulus Generalization and Discrimination Learning
Similar levels of responding to male and female quail appears to be an example of stimulus generalization from one sex to the other. As I described earlier, if males are presented with visual access to a female just before a chance to copulate with the female, the males will come to approach and remain near the female visual cues. Once this social proximity behavior has been established to female cues, the substitution of a male stimulus bird behind the window will elicit similar responses (Nash, Domjan, & Askins, 1989). Thus, social proximity behavior conditioned to female cues strongly generalizes to male cues. Can males ever tell male and female quail apart, and how do they learn to do so? With sufficient social experience, male quail do come to respond differentially to the sex of conspecifics. As other forms of discrimination learning, discrimination between stimulus males and stimulus females can be achieved by differential reinforcement. The subjects have to be exposed to female cues paired with copulatory reinforcement and male cues presented in the absence of copulatory opportunity (Nash et al., 1989).Discriminative performance improves with increased numbers of female-reinforced trials and with increased exposure to male stimuli in the absence of reinforcement (see Nash et al., 1989). The stimulus generalization of sexual learning, and subsequent discriminative performance, are illustrated by an experiment by Domjan and Ravert (1991). On each of the first 19 days of this experiment, the subjects received a reinforced trial that consisted of exposure to a female stimulus bird behind a window, followed by the opportunity to copulate with that female. Two
Learning about Species Typical Cues
165
,
test trials were then conducted, one with a female bird behind the window and the other with a male stimulus bird. The results of these tests are summarized by the first two bars in each panel of Fig. 4.The behavior of the male subjects was measured in four different ways. We measured the total percentage of time the subjects spent near the window, the number of times they entered the criteria1 zone in front of the window, how long
loor 80
FEMALE MALE
-
w $60-
1
z w
40-
z 20
L
-
l/i 8 2 1
0-
0
A
B
A
B
0
0
A
U A B
Fig. 4. Responses of male quail subjects to a stimulus female or a stimulus male that is visible through a narrow window. Tests were first conducted after 19 presentations of a female stimulus bird paired with copulatory reinforcement (Test Series A) and then again after 16 exposures to male stimulus birds in the absence of sexual reinforcement (Test Series B). (Based on Domjan & Ravert, 1991.)
166
Michael Domjan
they spent in the criterial zone per entry, and locomotor activity in the criterial zone. Notice that during the first pair of tests, the subjects responded similarly to male and female stimulus birds. These results illustrate that the responding that had developed to the stimulus females substantially generalized to the stimulus males. After the first test series, the subjects were given nonreinforced exposure to male stimulus birds. On each of the next 16 days (not shown), a male stimulus bird was placed in the side compartment, and the window was opened for 75 min. At the end of the 75 min, the window was closed and the subjects remained in their chambers until the next day without sexual reinforcement. The effects of these nonreinforced exposures to the male stimulus birds were assessed in tests with male and female stimulus birds at the end of the experiment, the results of which are summarized by the bars to the right of each panel of Fig. 4. This time strong differential responding occurred. The subjects spent a much larger percentage of their time near the window when a stimulus female was present behind the window than when a stimulus male was on the other side. The stimulus female stimulated fewer entries into the criterial zone than the stimulus male, but the subjects spent much more time in the criterial zone on each entry. The stimulus female also elicited less locomotion in the criterial zone. The various responses that were measured during the tests conducted at the end of the experiment indicate that when a female was visible on the other side of the window, the subjects approached the window and stayed there. The subjects also approached the window when a stimulus male was visible, but when they saw a stimulus male they did not remain near the window but kept moving about. Thus, only the stimulus female came to elicit social proximity behavior. Other experiments have shown that nonreinforced exposure to male stimuli is necessary for the development of this discriminative performance (see Nash et al., 1989). 2.
Learning to Discriminate between Different Kinds of Females
The results of the above study suggest that a preference to remain near a female rather than a male stimulus bird develops not because of an inherent preference for females over males but because of differential sexual reinforcement. If this is true, then it should be also possible to condition a discrimination between females of different types. It should be possible to condition subjects to respond more to one kind of female than another. Nash and Domjan (1991) tested this prediction using two types of female quail. Some of the stimulus females had the normal brown plumage that is characteristic of the species (N-females). The other stimulus quail were blonde (B-females). The blonde quail were of normal size, but their plum-
Learniog about Species Typical Cues
167
age was primarily a smooth buff color. The male subjects that served in the experiment had the characteristic brown plumage and had been kept in mixed-sex brooders with other brown quail until 30 days of age. Therefore, the subjects should have become imprinted to the normal brown plumage (Gallagher, 1976). The experiment started with a phase of sexual reinforcement. On each of 16 days of this phase, the subjects were first exposed to visual cues of a blonde female behind the window, followed by the opportunity to copulate with this female. Fig. 5 (left panel) shows that as these trials progressed, the subjects spent increasing amounts of time near the window. Thus, as usual, copulation established social proximity behavior. During the next phase of the experiment, the subjects received nonreinforced exposures to normal brown stimulus females. Six such exposure trials were conducted, followed by nonreinforced test trials with blonde and normal stimulus females. The middle and left panels of Fig. 5 show the results of these phases of the experiment. Social proximity behavior decreased the first time normal brown stimulus females were visible behind the window (see El, middle panel of Fig. 5). Responding to the normal brown females declined with subsequent nonreinforced presentations of the normal stimulus females, both during
100
-
3
g
Opportunitytrials
i i
Exposure i trials i
Tests r
80-
2
5 pc
60-
w z
2 I-
40-
-8
iL;
20-
0
" " " " ' ~ " ' ~ '
168
Michael Domjm
the exposure phase (middle panel of Fig. 5 ) and during the test trials conducted at the end of the experiment (right panel of Fig. 5). The final outcome was that the male subjects responded much more vigorously to the blonde stimulus females than to the normal females (see right panel of Fig. 5). The discriminative responding that developed between the blonde and normal stimulus females was similar to what we had previously observed between normal stimulus females and stimulus males (See Fig. 4). Thus, this study failed to provide any evidence that male quail have an inherent predisposition to respond preferentially to normal brown females. 3. Learning to Respond More to Males than to Females
In the above study, male quail were conditioned to respond more to blonde stimulus females than to normal brown females. If differential responding is governed primarily by differential sexual reinforcement, then it should be possible to condition male quail to respond more to stimulus males than to stimulus females. Such an outcome should result from a procedure in which stimulus males are paired with sexual reinforcement and stimulus females are presented in the absence of copulatory opportunity. This prediction was tested in a second experiment by Nash and Domjan (1991). The procedural design was the same as in the study of copulatory reinforcement provided by blonde female quail. However, in this case copulatory reinforcement was provided by male stimulus birds. To insure that the male stimulus birds would tolerate being mounted, they were functionally castrated by putting them on a schedule of restricted lighting. Reduced gonadal function in the stimulus males was confirmed by measures of their cloaca1 gland (Sachs, 1967). On each of the first 20 days of the experiment, visual exposure to a male stimulus bird behind the window was followed by the opportunity to copulate with that male. As was expected, the experimental subjects came to spend increasing amounts of time near the window as a result of these procedures (see left panel of Fig. 6). During the next phase of the experiment, the stimulus males were replaced with normal females, and exposure to the stimulus females occurred in the absence of copulatory opportunity. The results of these nonreinforced female exposures are summarized in the middle panel of Fig. 6. At first, the subjects responded to the stimulus females almost as much as they responded to the stimulus males at the end of phase 1, indicating generalization of the behavior from stimulus males to stimulus females. However, as expected, responding to the stimulus females declined with further nonreinforced exposures. A series of nonreinforced tests with stimulus males and stimulus females was conducted at the end of the experiment. The results of these tests
Learning about Species Typical Cues
r
Opportunity trials
Exposure trials
169 Tests
6,
2:L iL 1
2
3
4
5
8
7
8
9 10
El E2 €3 E4 €5 EB E7 EB
M1 F2
M3 F4
BLOCKS OF TWO TRIALS
Fig. 6. Percentage of time spent near a window by male quail. During the first phase of the experiment (left panel), exposure to functionally castrated stimulus males was paired with the opportunity to copulate with these males. During the second phase (middle panel), exposure to a normal stimulus female was provided in the absence of copulatory opportunity. During the test phase (right panel), the subjects received nonreinforced exposures to stimulus males (M) and stimulus females (F). (Based on Nash & Domjan, 1991.)
confirmed that the subjects had learned to respond more to stimulus males than to stimulus females (see right panel of Fig. 6). However, in this case the degree of discriminative responding was not as marked as what we had observed in previous studies (compare Fig. 6 with Fig. 4). This impression was confirmed by data from a second group of subjects that was tested concurrently with the subjects whose data are shown in Fig. 6 (see Nash & Domjan, 1991, Experiment 2). The second group received the identical procedure, except that normal female stimulus birds were paired with sexual reinforcement in phase 1 and male stimulus birds were presented in the absence of copulatory opportunity in phase 2. The second group showed greater differential responding to the reinforced versus nonreinforced stimulus birds than the first group. The above results suggest that it is harder to get subjects to favor responding to stimulus males than it is to get them to favor responding to stimulus females. Thus, although there is considerable plasticity in sexual discrimination learning, this plasticity appears to be superimposed on a bias to approach and remain near stimulus females. This bias is also evident in discrimination reversal training and retention tests (see Nash & Domjan, 1991, Experiment 2). The mechanisms responsible for the bias of male quail to favor responding to female stimuli have yet to be identified. One possibility is that the bias reflects differences in the quality of sexual reinforcement derived from
170
Michael Domjan
copulating with females as compared with functionally castrated males. Copulation with functionally castrated males may not be as reinforcing. Alternatively, the effect may reflect a stimulus relevance or belongingness relation (LoLordo, 1979). The cues provided by a stimulus female may be more “relevant” to sexual reinforcement than the cues provided by a stimulus male.
111. Do Species Typical Cues Have Special Properties as
Conditioned Stimuli? The studies reviewed in the previous section showed that the cues provided by stimulus females and stimulus males can be used as positive and negative discriminative stimuli for sexual reinforcement. Furthermore, most of the results that we obtained with the species typical discriminative stimuli were comparable to the results of more conventional studies of discrimination learning. Are there also some special (perhaps unique) properties of species specific cues as conditioned stimuli? We have addressed this question in studies of learning involving limited female cues and have found three ways in which such cues appear to have privileged access to control sexual behavior. First, species typical cues as conditioned stimuli are more effective in supporting conditioned copulatory behavior than arbitrary conditioned stimuli. Second, responding to species typical cues is more apt to become sensitized by stimulus exposure than responding to arbitrary stimuli. Third, the conditioning of species typical cues is more difficult to disrupt with a blocking procedure than the conditioning of arbitrary stimuli. OF COPULATORY BEHAVIOR A. CONDITIONING
Copulatory behavior in the domestic quail consists of the male grabbing the female in the back of her head or neck with his beak. The male then mounts by placing both feet on the female’s back and makes cloacal contact by arching his back and pressing his cloacal opening against the female’s. Cloaca1 contacts are accompanied by spreading of the wings for balance and a rapid “shivering” movement that concludes in a brief period of immobility. For copulatory behavior to become conditioned, a threedimensional object that the male can grab and mount has to be used as the CS. We have found that in addition to that, it is very helpful for the CS to include some of the species typical features of a female quail. In particular, subjects are much more likely to mount a CS object that includes the taxidermically prepared head of a female, with some of the neck feathers.
Learning about Species Typical Cues
171
In our initial efforts to condition copulatory behavior in male quail, we used two different types of CSs (Domjan, O'Vary, & Greene, 1988). In one experiment, the CS was a small yellow stuffed toy (a Pound Puppy 0); in another experiment, the CS was exposure to a female quail whose appearance had been altered by attaching two large bright orange feathers to her shoulders. As usual, the conditioning procedure consisted of providing access to the CS immediately before giving the males an opportunity to copulate with a normal female bird. The subjects came to approach the CS as a conditioned response in both experiments. However, only the embellished female CS came to elicit conditioned copulatory responses. The two CSs that were used in the above study differed in numerous respects, any one of which might have been responsible for their contrasting effectiveness. To better isolate the stimulus features required for the conditioning of copulatory behavior, in subsequent studies we used CSs that differed systematically in terms of the species typical features they contained (see Domjan, Huber-McDonald, & Holloway, 1992). Two such stimuli are shown in Fig. 7. Both of these objects were made of terrycloth filled with a soft polyester fiber. The objects are similar in size and are shaped so that a male can easily grab, mount, and make cloaca1 contact with them. The primary difference is that one object includes a taxidermically prepared head and some of the neck feathers of a female quail. Cusato and Domjan (1998) recently conducted separate conditioning experiments with each of the objects shown in Fig. 7. One trial was con-
Fig. 7. Two CS objects made of blue terrycloth filled with soft polyester fiber. Both are shaped to permit a male to grab and mount the objects. In addition, one of the CSs includes a taxidermically prepared female head and partial neck feathers. (From Cusato & Domjan, 1998, reprinted by permission.)
172
Michael Domjan
ducted each day, and two groups served in each experiment. For group Paired, conditioning trials consisted of a 30-sec presentation of the CS, followed immediately by an opportunity for the subjects to copulate with a normal female bird. For group Unpaired, the same number of CS exposures and copulatory opportunities were provided, but for these subjects copulatory opportunity always occurred 25-30 min before their exposure to the CS. Ten conditioning trials were conducted, with a nonreinforced test trial presented after every two copulation trials. The results of the test trials are summarized in Fig. 8. Four different responses to the CS objects were measured: time spent near the CS, and the frequency of grab, mount, and cloacal contact responses. One noteworthy aspect of the results is that the CS object that included cues of a female’s head and neck elicited relatively little behavior unconditionally. The Unpaired group showed a modest level of approach to the head-CS but made virtually no grab, mount, and cloacal contact responses.
Fig. 8. Comparison of sexual conditioned responses to a CS object that either includes the cues of a female’s head and neck or lacks these species typical features. The panels show mean time spent near the CS and mean frequency of grab, mount, and cloacal contact responses during nonreinforced test trials conducted for subjects that received the CS paired (P) or unpaired (U) with copulatory opportunity. (Based on Cusato & Dornjan, 1998.)
Learning about Species Typical Cues
173
As expected, Paired subjects came to approach and remain near the CS significantly more than Unpaired subjects. Furthermore, this effect was evident whether or not the CS object included female head and neck cues. The inclusion of species typical cues in the CS was not necessary for the development of the conditioned approach response. However, the female head cues were important for the conditioning of the various copulatory responses. Conditioned grab, mount, and cloaca1 contact responses were much more likely to develop if the CS included cues of a female head than in the absence of these species typical cues. But, even the head-CS had to be paired with sexual reinforcement to elicit substantial levels of copulatory behavior. Thus, the development of conditioned copulatory behavior required both the use of the head-CS and pairings of this CS with sexual reinforcement.
B. SENSITIZATION OF RESPONDING TO A CS THAT INCLUDES FEMALE CUES The above findings illustrate one way in which female cues appear to have privileged access to control the sexual behavior of males. Once paired with copulatory reinforcement, limited female cues are more apt to elicit conditioned copulatory responses. Another way that female cues appear to be special is that mere exposure to such stimuli increases responding to them. That is, responding to female cues becomes sensitized with repeated exposures. Cusato and Domjan (1998) recently found this effect in other aspects of their research with the terrycloth objects that are displayed in Fig. 7. Separate sets of male quail were tested with the stimulus object that contained female head cues and the object that was made entirely of terrycloth. Within each of these sets of animals, one group received extensive exposure to the stimulus object (without access to a live female or other copulatory opportunity) prior to the test session. Fifty-nine exposures of the stimulus object (15 min each time) were provided over a period of 17 days. The remaining subjects did not get this pre-test exposure and encountered the stimulus object for the first time during the test session. (These nonpreexposed subjects subsequently served in the experiments whose results were summarized in Fig. 8.) The findings are summarized in Fig. 9. Mere exposure to the terrycloth object with the taxidermic female head elevated both approach and copulatory responses to the object. The increased responding was statistically significant for all response measures except the frequency of grabs. In contrast, only a modest increase occurred in time spent near the stimulus object that lacked species typical features (and this increase was not statistically significant). No copulatory responses were elicited by the no-head object whether or not the object had been preexposed.
174
Michael Domjan
Fig. 9. Comparison of responding to a terrycloth object that either included the cues of a female’s head and neck or lacked these species typical features. The panels show mean time spent near the terrycloth object and mean frequency of grab, mount, and cloaca1contact responses during a test trial for subjects that previously received extensive nonreinforced preexposure to the test stimulus (PX) or no prior exposure (NX). (Based on Cusato & Domjan, 1998.)
C. RESISTANCE TO THE BLOCKING OF SEXUAL CONDITIONING
A third respect in which species typical features appear to have special properties in sexual conditioning is that the conditioning of such cues cannot be blocked by a previously conditioned CS. The blocking paradigm was developed originally by Kamin in the fear conditioning of rats (Kamin, 1969) and stimulated major theoretical developments in animal learning (e.g., Mackintosh, 1975; Rescorla & Wagner, 1972) because it dramatically demonstrated that CS-US contiguity is not sufficient for the establishment of conditioned responding. The blocking paradigm involves two phases of training. During the first phase, subjects receive enough pairings of a conditioned stimulus (CS1) with the US to establish strong conditioned responding to CSl. In the next phase, CS1 is presented simultaneously with a new cue, CS2, and this stimulus compound is paired with the US. The focus of interest is the extent
Learning about Species Typical Cues
175
to which conditioned responding develops to CS2. The blocking effect is said to occur if the presence of the previously conditioned CS1 disrupts (or blocks) the development of conditioned responding to CS2. Koksal, Domjan, and Weisman (1994) explored the blocking effect in sexual conditioning using terrycloth objects with and without female cues in the role of CS2. During the first phase of each experiment, the Blocking group received an audiovisual cue (a flashing green light accompanied by a buzzer) for 30 s, followed immediately by the opportunity to copulate with a female bird. Fifteen such phase 1 trials were conducted, which was sufficient to condition the audiovisual cue to asymptote. In the next phase of the experiment, the audiovisual CS1 was presented together with a terrycloth object, and the compound was paired with copulatory opportunity. In one experiment, the terrycloth object lacked a female head; in another experiment, CS2 included the female head. Four test trials with the terrycloth object presented alone were conducted during this phase, one after the first four compound conditioning trials, and the remaining three after an additional four compound conditioning trials. In addition to the Blocking group, two control groups were included in the experiments. The control groups received the same CS1-CS2 conditioning trials during the second phase as the subjects that received the blocking procedure. However, the control groups were not conditioned to respond to CS1 prior to the compound conditioning trials. Control-1 received unpaired exposures to CS1 and the US in the first phase, and Control-2 was not exposed to either CS1 or the US in the first phase. The results of the experiments are summarized in Fig. 10 in terms of how much time the subjects spent near the terrycloth object (CS2) during test trials conducted in phase 2. When the terrycloth object lacked species typical features (left panel of Fig. lo), the presence of the previously conditioned audiovisual cue blocked the development of conditioned approach responding. The Blocking group spent significantly less time near the tenycloth object than the control groups. In contrast, when CS2 included the stimuli of a female’s head (right panel of Fig. lo), all three groups of subjects spent substantial periods near the CS. This time, the previously conditioned audiovisual cue did not block the conditioning of the terrycloth CS. Thus, inclusion of species typical features in the CS object made that object resistant to the blocking effect. (For analogous results in fear conditioning, see LoLordo, Jacobs, & Foree, 1982.)
IV. Potentiation of Responding to Species Typical Cues In the studies reviewed in the previous sections, species typical cues were used by incorporating them into the CS that was paired with copulatory
Michael Domjan
176
"1
-0-
BLOCKING
+ CONTROL-1 + CONTROL-2 1
2
3
4
1
2
3
4
TEST TRIALS
Fig. 10. The development of conditioned approach responding to a CS object that either lacked species typical features (left panel) or included cues of a female's head (right panel). During the conditioningtrials, the CS object was presented simultaneouslywith an audiovisual cue and paired with copulatory opportunity. For the Blocking groups, the audiovisual cue was previously paired with the sexual US. For Control-1, the audiovisual cue was previously presented unpaired with.the sexual US. For Control-2, the audiovisualcue was not previously presented. (Based on Koksal et al., 1994.)
opportunity. Another way in which species typical cues may be used in laboratory analyses of learning is by incorporating them into a probe stimulus that is used to examine learning about other kinds of conditioned stimuli. A. POTENTIATION BY CONDITIONED CONTEXTUAL CUES
Early evidence that the presence of species typical cues can reveal the existence of sexual learning to other cues was provided by a study with laboratory rats (Zamble, Hadad, Mitchell, & Cutmore, 1985).Conditioning trials consisted of placing the rats in a plastic tub with wood shavings on the floor. After exposure to this CS, the animals were moved to a mating arena in which they were exposed to a sexually receptive female rat on the other side of a wire-mesh barrier. Eight such conditioning trials were conducted. The conditioning of sexual arousal to the contextual CS was not measured in terms of any changes in behavior elicited directly by the context. Rather, during the test trial subjects were moved to the mating arena after exposure to the contextual CS, and this time the wire-mesh barrier separating the male from the female was removed. Exposure to the conditioned contextual cues reduced the latency of the males to ejaculate when they were permitted to copulate with the female. Thus, during the test trial, the species typical cues provided by the female served to reveal the conditioned properties of the contextual CS (see also Zamble, Mitchell, & Findlay, 1986).
Learning about Species Typical Cues
177
Female cues are also effectiveprobe stimuli for revealing the conditioned properties of contextual cues in the sexual conditioning of male Japanese quail. In one study (Domjan, Akins, & Vandergriff, 1992, Experiment 2), male quail spent alternate days in two different types of experimental chambers. On some days, the birds were housed in chambers whose walls and ceiling were painted white, whose floor was covered with sand, and that had dried tree branches that reached from floor to ceiling and side to side. On alternate days, the birds were moved across the hall and placed in chambers that were shorter, whose walls were made of plywood covered with a clear sealant, and whose floors were made of wire mesh suspended above a drop tray. Opportunity to copulate with a female bird was provided periodically in one of the contexts (counterbalanced across subjects) by releasing the female from under a hood where she had been placed. Copulatory opportunity was never provided in the alternate context. After 20 trials, the subjects were tested with a female bird presented behind a window. The test trial was the first time the window was opened during the experiment, and in this experiment the female behind the window had never been released into the male’s compartment. Nevertheless, subjects spent significantly more time in front of the window in the context in which they previously received copulatory opportunity than in the alternate context. Thus, the conditioned contextual cues increased the responsiveness of the male subjects to cues provided by a female bird on the other side of the window. Contextual potentiation of responding to female cues is a highly robust phenomenon. First, the potentiation effect can increase not only approach and proximity behavior elicited by female cues but also copulatory behavior (Domjan, Greene, & North, 1989). Second, the female cues need not be provided by a live female bird. A terrycloth object with a taxidermically prepared female head and some neck feathers can also serve as an effective probe stimulus (Domjan et al., 1989). Finally, the effect can be evident following just one context-conditioning trial (Hilliard, Nguyen, & Domjan, 1997). Figure 11shows the results of an experiment in which independent groups of male quail were placed in a distinctive experimental chamber for 0, 2, or 4 min before a female bird was placed with them for 5 min to permit copulation (Hilliard et al., 1997, Experiment 2). Males in the Unpaired control group received the same duration of exposure to the contextual cues as the 4-min Paired group. However, for the Unpaired controls, copulatory opportunity was provided in their home cages more than 1 h before or after exposure to the contextual CS. The results of the single context conditioning trial were assessed the next day. For the test trial, the subjects were returned to the experimental chamber and exposed for the first time
Michael Domjan
178
'"1
Fig. 11. Mean time spent near a probe stimulus that included the head and some neck feathers of a female quail after one previous exposure to the test context that was followed immediately (Group 0) or 2 or 4 min later (Group 2 and Group 4) or not at all (group Unpaired) by the opportunity to copulate with a live female. (From Hilliard, Nguyen, & Domjan, 1997, reprinted by permission.)
to a probe stimulus that had been placed in the middle of the chamber. The probe stimulus was a terrycloth object with a female quail head, such as that shown in Fig. 7. The results summarized in Fig. 11 show that groups 0, 2, and 4,which had received a single exposure to the contextual cues of the experimental chamber paired with copulatory opportunity, spent much more time near the female probe stimulus than subjects in the Unpaired control group. The duration of context exposure (0, 2, or 4 min) before introduction of the female during the conditioning trial did not affect the outcome of the experiment. One-trial learning effects have been reported previously in appetitive conditioning paradigms (e.g., D'Amato & Buckiewicz, 1980). However, one-trial appetitive conditioning is not a common finding and appears to require special test procedures. The probe test used by Hilliard et al. (1997) provides an especially sensitive index of appetitive conditioning in the sexual behavior system. One-trial learning has been obtained more frequently in fear conditioning and taste-aversion learning (e.g., Fanselow, 1986; Garcia, Kimeldorf, & Koelling, 1955). The conditioning procedure used by Hilliard et al. (1997)
Learning about Species Typical Cues
179
is similar to the context conditioning procedure that has been investigated in fear conditioning (e.g., Fanselow, 1986). In both procedures, subjects receive a US after their first exposure to a novel context. However, the conditioning effect obtained by Hilliard et al. (1997) differs from context fear conditioning in an important respect. Fear conditioning is disrupted by presentation of the US within a few seconds of the animal’s placement in the experimental context (Fanselow, 1986, 1990; Westbrook, Good, & Kiernan, 1994). In contrast, Hilliard et al. (1997) did not observe less conditioning in subjects for whom a female was introduced immediately upon exposure to the contextual cues (group 0) than in subjects that were exposed to the contextual cues for 2 or 4 min before the female appeared. Further research is required to determine whether this difference is due to special properties of the sexual conditioning system or to procedural differences between the fear and sexual conditioning experiments. B. POTENTIATION BY PUNCTATE CONDITIONED STIMULI The presentation of a discrete sexually conditioned stimulus can also facilitate responding to species typical cues. The first clear evidence of this was provided by Zamble et al. (1985, Experiment 3) in an experiment with Long-Evans rats.’ A flashing light and a plastic toy fish served as the CS for different subjects. After exposure to the CS during the conditioning trials, the subjects were moved to a mating arena that provided exposure to a receptive female on the other side of a wire-mesh screen. Learning was evaluated by permitting the subjects to copulate with a female after exposure to the light or toy fish CS. As was the case when contextual cues served as the CS (Zamble et al., 1985, Experiment 1; Zamble et al., 1986), exposure to the light or toy object decreased the latency of the subjects to ejaculate when they were permitted to copulate with the female. We obtained similar results in an experiment with quail conducted around the same time (Domjan, Lyons, North, & Bruell, 1986). In this experiment, a red light served as the CS. Each conditioning trial consisted of presentation of the red light, followed 30 s later by the opportunity to copulate with a female. Birds in a control group received copulatory opportunities unsignaled. After three trials, males that received the Pavlovian signal initiated copulation with the female significantly faster than males for whom access In two recent publications (Hollis, 1997; Hollis et al., 1997), the claim was made that the first report that a sexually conditioned stimulus can increase the responsiveness of males to female cues was provided by Fams in his Ph.D. dissertation (Farris, 1964). 1 have been unable to verify these claims in my own reading of Farris’s dissertation. Hollis and I have discussed the discrepancy in our interpretations at great length. Unfortunately, Farris did not provide enough information about the copulatory behavior of his conditioned and control subjects to help resolve our differences of opinion.
180
Michael Domjan
to the female was not announced by the red light. Thus, as in the studies with rats, the punctate CS increased the effectiveness of female cues in eliciting male copulatory behavior. Facilitation of responding to species typical cues by the presence of a punctate CS has also been demonstrated in the blue gourami, a fish species (Hollis et al., 1989). A red light again served as the CS. During the conditioning trials, presentation of the red light was paired with visual access to a female for subjects in the Pavlovian group. For subjects in a control group, the CS and US presentations were unpaired. After 12 days of training (which included 120 conditioning trials), a test session was conducted during which the male subjects were permitted to interact with a female after exposure to the CS. The Pavlovian signal significantly increased the courtship behavior of the subjects and decreased the aggressive responses they directed towards the females. Thus, as in the other examples, exposure to the punctate CS increased the effectiveness of the female cues in eliciting sexual behavior in the male subjects. In all of the above examples, males served as subjects and females provided the species typical cues. In a recent study, the role of the subjects was reversed (Gutierrez & Domjan, 1997). In this experiment, female quail received conditioning trials in which a punctate CS was paired with the opportunity to copulate with a male. This procedure resulted in little behavior directed towards the CS. However, females for whom the CS reliably preceded the presentation of a male were more sexually receptive than females in a control group once a stimulus male was introduced to provide copulatory opportunity. The CS increased the frequency and duration of squatting responses performed by the female in the presence of a stimulus male. Squatting in female quail is a response indicative of sexual receptivity. C. SIGNIFICANCE OF THE POTENTIATION OF RESPONDING TO SPECIES TYPICAL CUES The potentiation effects described in this section are important both methodologically and functionally. The methodological significance of the phenomena is that they provide a highly sensitive technique for measuring the associative properties of a CS. With the potentiation effect, the associative properties of a CS can be detected after just one conditioning trial and under circumstances in which the conditioned stimuli do not elicit any discernible behavior directly. We have found this to be especially useful in studying the associative properties of contextual cues and in studying the conditioning of female quail. Contextual cues that have been paired with copulatory opportunity do not produce clearly discernible changes in behavior. For example, we have been unsuccessful in detecting changes in locomo-
Learning about Species Typical Cues
181
tor behavior in a sexually conditioned context. However, the conditioned properties of contextual stimuli are readily evident in the potentiation of responding to species typical cues. Similarly, our efforts to identify sexual learning in female quail were of limited success until we employed the potentiation paradigm (Gutierrez & Domjan, 1997). The functional significance of the potentiation effect is that it provides a mechanism for increased reproductive success in organisms exposed to an arbitrary CS. By potentiating responding to the species typical cues provided by a potential sexual partner, the presentation of an arbitrary CS can increase the efficiency and effectiveness of the subject’s sexual behavior. Evidence of this increased reproductive functioning has been obtained in both quail and blue gouramis. As I noted earlier, Hollis et al. (1989) found that exposure to a sexual CS reduces aggressive displays and facilitates the emergence of courtship and reproductive behavior when a male blue gourami is permitted to interact with a female. Dramatic recent evidence indicates that this facilitation of sexual behavior persists for several days and greatly increases the number of offspring that are produced. Males that received a control procedure produced on average fewer than 50 offspring. In contrast, males allowed to court a female after exposure to a sexually conditioned red light produced on average more than 1000 offspring (Hollis, Pharr, Dumas, Britton, & Field, 1997). Studies with quail have shown that the increased responsiveness to female cues that results from exposure to an arbitrary sexually conditioned CS enables a male to copulate with a female more quicklyand thus gain an advantage in sexual competition with other males (Gutierrez & Domjan, 1996). The potentiation of responding to female cues is also accompanied by the release of greater volumes of semen and greater numbers of spermatozoa (Domjan, Blesbois, & Williams, 1998). Thus, sexual conditioning to stimuli that are initially unrelated to sexual behavior can nevertheless significantly alter an organism’s reproductive success by increasing its responsiveness to the natural species typical cues that usually regulate sexual behavior. V. Conclusion A persistent nagging problem for investigations of animal learning conducted with traditional experimental techniques has been the extent to which the results of such studies have ecological validity. Historically, the ecological validity of traditional laboratory experiments has been defended with the use of plausibility arguments rather than with careful comparison of the laboratory stimuli and procedures with the events an organism is
182
Michael Domjan
likely to encounter in its natural environment. Laboratory studies cannot be expected to stimulate natural conditions perfectly, because some degree of artificiality is an inherent feature of all experimental work. However, the ecological validity of laboratory procedures is easier to defend if those procedures include stimuli and event sequences that are likely to occur in nature. The experiments described in this chapter provide a case study of such research. The sexual behavior of male quail is controlled primarily by the visual features of a female. The shape and plumage of a female quail in the laboratory is much like its shape and plumage in the wild. Therefore, laboratory investigations of learning about female cues are likely to reveal how female cues control sexual behavior in the wild. The present studies have shown that the species typical cues provided by a conspecific can be profitably used in studies of what quail (and presumably other animals) learn from opportunities to copulate with other animals of their species. The association of female cues with sexual reinforcement results in a tendency of males to approach and remain near females, whereas exposure to male cues in the absence of sexual reinforcment results in reduced time spent near stimulus males. There is considerable (but not complete) flexibility in this social proximity learning. If male cues are paired with copulatory opportunity and female cues are presented without sexual reinforcement, the birds learn to favor approaching and remaining near stimulus males. However, preferential conditioned responding to stimulus males is not as marked nor as persistent over time as preferential conditioned responding to females. In many respects, learning about species typical cues is similar to learning about arbitrary stimuli in conditioning experiments. Conventional learning phenomena such as acquisition, extinction, and stimulus discrimination are evident with CSs that include species typical features. However, species typical cues also have some special properties. Responding is more likely to increase with mere exposure to species typical cues than with mere exposure to an arbitrary CS. The addition of species typical features to a stimulus object facilitates the conditioning of copulatory responses to that object, and makes the object resistant to the blocking effect (the disruption of conditioning that occurs when a new CS is presented in compound with a previously conditioned CS). The addition of species typical features to a stimulus object also makes that object much more effective as a probe stimulus for detecting sexual associations conditioned to an arbitrary punctate stimulus or to contextual cues. The inclusion of female cues in a CS object may facilitate sexual learning in males by several different mechanisms. One possibility is that male quail have a predisposition to associate female cues with sexual reinforcement.
Learning about Species Typical Cues
183
Female cues may be especially relevant to sexual USs. Alternatively, female cues may facilitate sexual conditioned responding because males encounter these cues while they copulate with a female in sexual conditioning experiments. A third possibility is that species typical cues are especially salient for male quail and attract more of their attention than arbitrary CSs. If this is the case, then species typical cues should be more easily conditioned regardless of the type of US that is used. Future research is required to evaluate these possibilities. ACKNOWLEDGMENTS The preparation of this chapter and the recent research with quail that is described were supported by National Institutes of Health Grant MH39940. I wish to thank Susanna Douglas for her help in preparing the figures and Brian Cusato for his comments on an earlier draft of the manuscript. Correspondence should be addressed to Michael Domjan, Department of Psychology, University of Texas, Austin, Texas 78712. (Email:
[email protected]).
REFERENCES Balthazart, J., Reid, J., Absil, P., Foidart, A., & Ball, G. F. (1995). Appetitive as well as consummatory aspects of male sexual behavior in quail are activated by androgens and estrogens. Behavioral Neuroscience, 109, 485-501. Bitterman, M. E. (1975). The comparative analysis of learning. Science, 188, 699-709. Cusato, B., & Domjan, M. (1998). Special efficacy of female species typical cues in sexual conditioning: Tests with a CS preexposure design. Learning and Motivation, in press. D’Amato, M. R., & Buckiewicz, J. (1980). Long-delay, one-trial conditioned preference and retention in monkeys (Cebus apella). Animal Learning & Behavior, 8, 359-362. Domjan, M. (1983). Biologicalconstraints on instrumental and classicalconditioning: Implications for general process theory. The Psychology of Learning and Motivation, 17,215-277. Domjan, M. (1987). Photoperiodic and endocrine control of social proximity behavior in male Japanese quail (Coturnix coturnix japonica). Behavioral Neuroscience, 101, 385-392. Domjan, M. (1994). Formulation of a behavior system for sexual Conditioning. Psychonomic Bulletin & Review, 1, 421-428. Domjan, M. (1997). Behavior systems and the demise of equipotentiality: Historical antecedents and evidence from sexual conditioning. In M. E. Bouton & M. S. Fanselow (Eds.), Learning, motivation, and cognition (pp. 31-51). Washington, D C American Psychological Association. Domjan, M., & Galef, B. G., Jr. (1983). Biological constraints on instrumental and classical conditioning: Retrospect and prospect. Animal Learning & Behavior, 11, 151-161. Domjan, M., & Hall, S. (1986). Determinants of social proximity in Japanese quail (Coturnix coturnix japonica): Male behavior. Journal of Comparative Psychology, 100, 59-67. Domjan, M., & Holloway, K. S. (1998). Sexual learning. In Comparative psychology: A handbook. New York: Garland, in press. Domjan, M., & Nash, S. (1988). Stimulus control of social behavior in male Japanese quail, Coturnix coturnix japonica. Animal Behaviour, 36, 1006-1015.
184
Michael Domjan
Domjan, M., & Ravert. R. D. (1991). Discriminating the sex of conspecifics by male Japanese quail (Coturnix coturnix japonica). Journal of Comparative Psychology, 105, 157-164. Domjan, M., Akins, C., & Vandergriff, D. H. (1992). Increased responding to female stimuli as a result of sexual experience: Tests of mechanisms of learning. Quarterly Journal of Experimental Psychology, 45B, 139- 157. Domjan, M., Greene, P., & North, N. C. (1989). Contextual conditioning and the control of copulatory behavior by species-specific sign stimuli in male Japanese quail. Journal of Experimental Psychology: Animal Behavior Processes, 15, 147-153. Domjan, M., Huber-McDonald, M., & Holloway, K. S. (1992). Conditioningcopulatory behavior to an artificial object: Efficacy of stimulus fading. Animal Learning & Behavior, 20, 350-362. Domjan, M., Lyons, R., North, N. C., & Bruell, J. (1986). Sexual Pavlovian conditioned approach behavior in male Japanese quail (Coturnix coturnix japonica). Journal of Comparative Psychology, 100, 413-421. Domjan, M., O’Vary, D., & Greene, P. (1988). Conditioning of appetitive and consummatory sexual behavior in male Japanese quail. Journal of the Experimental Analysis of Behavior, 50, 505-519. Fanselow, M. S. (1986). Associative vs. topographicalaccounts of the immediate shock-freezing deficit in rats: Implications for the response selection rules governing species-specific defensive reactions. Learning and Motivation, 17, 16-39. Fanselow, M. S. (1990). Factors governing one-trial contextual conditioning. Animal Learning & Behavior, 18, 264-270. Fanselow, M. S. (1994). Neural organization of the defensive behavior system responsible for fear. Psychonomic Bulletin & Review, 1, 429-438. Fanselow, M. S. (1997). Species-specificdefense reactions: Retrospect and prospect. In M. E. Bouton & M. S. Fanselow (Eds.), Learning, motivation, and cognition (pp. 321-341). Washington, D C American Psychological Association. Farris, H. E. (1964). Behavioral development, social organization,and conditioningof courting behavior in the Japanese quail Coturnix coturnix japonica. Ph. D. Dissertation, Michigan State University. (University Microfilms, Dissertation #65-1738, Ann Arbor, Michigan.) Farris, H. E. (1967). Classical conditioningof courting behavior in the Japanese quail, Coturnix coturnix japonica, 10, 213-217. Gallagher, J. E. (1976). Sexual imprinting: Effects of various regimens of social experience on the mate preference in Japanese quail (Coturnix coturnix japonica). Behaviour, 57,91-115. Gallistel, C. R. (1990). The Organization of learning. Cambridge, MA: MIT Press. Garcia, J., Kimeldorf, D. J., & Koelling, R. A. (1955). Conditioned aversion to saccharin resulting from exposure to gamma radiation. Science, 122, 157-158. Gutierrez, G., & Domjan, M. (1996). Learning and male-male sexual competition in Japanese quail (Coturnix japonica). Journal of Comparative Psychology, 110, 170-175. Gutierrez, G., & Domjan, M. (1997). Differences in the sexual conditioned behavior of male and female Japanese quail (Coturnix japonica). Journal of Comparative Psychology, 111, 135-142. Hailman, J. P. (1985). Historical notes on the biology of learning. In T. D. Johnston & A. T. Pietrewicz (Eds.), Issues in the ecological study of learning (pp. 27-57). Hillsdale, NJ: Erlbaum. Hamilton, M. B., Domjan, M., & Mills, A. (1998). Sexual behavior of male Japanese quail genetically selected for fear and sociality. Psychobiology, in press. Hilliard, S., Domjan, M., Nguyen, M., & Cusato, B. (1998). Dissociation of conditioned appetitive and consummatory sexual behavior: Satiation and extinction tests. Animal Learning & Behavior, 26, 20-33.
Learning about Species Typical Cues
185
Hilliard, S., Nguyen, M., & Domjan, M. (1997). One-trial appetitive conditioning in the sexual behavior system. Psychonomic Bulletin and Review 4, 237-241. Hinde, R. A., & Stevenson-Hinde, J. (Eds.). Constraints on learning. New York Academic Press. Hollis, K. L. (1982). Pavlovian conditioning of signal-centered action patterns and autonomic behavior: A biological analysis of function. Advances in the study of behavior, 12, 1-64. Hollis, K. L. (1997). Contemporary research on Pavlovian conditioning: A “new” functional analysis. American Psychologist, 52, 956-965. Hollis, K. L., Cadieux, E. L., & Colbert, M. M. (1989). The biological function of Pavlovian conditioning: A mechanism for mating success in the blue gourami (Trichogaster trichoprerus). Journal of Comparative Psychology, 103, 115-121. Hollis, K. L., Pharr, V. L., Dumas, M. J., Britton, G. B., & Field, J. (1997). Classical conditioning provides paternity advantage for territorial male blue gouramis (Trichogasrer trichopterus). Journal of Comparative Psychology, 111, 219-225. Johnston, T. D. (1981). Contrasting approaches to a theory of learning. Behavioral and Brain Sciences, 4, 125-173. Kamil, A. C., & Balda, R. P. (1985). Cache recovery and spatial memory in Clark’s nutcrackers (Nucifraga colurnbiana). Journal of Experimental Psychology: Animal Behavior Processes, 11, 95-111. Kamil, A. C., & Sargent, T. D. (Eds.). (1981). Foraging behavior: Ecological, ethological, and psychological approaches. New York: Garland. Kamin, L. (1969). Predictability, surprise, and conditioning. In B. A. Campbell & R. M. ,, Church (Eds.), Punishment and aversive behavior. New York: Appleton-Century-Crofts. Koksal, F., Domjan, M., & Weisman, G. (1994). Blocking of the sexual conditioning of differentially effective conditioned stimulus objects. Animal Learning & Behavior, 22, 103-111. LoLordo, V. M. (1979). Selective associations. In A. Dickinson & R. A. Boakes (Eds.), Mechanisms of learning and motivation (pp. 367-398). Hillsdale, NJ: Erlbaum. LoLordo, V. M., Jacobs, W. J., & Foree, D. D. (1982). Failure to block control by a relevant stimulus. Animal Learning & Behavior, 10, 183-192. Mackintosh, N. J. (1975). A theory of attention: Variations in the associability of stimuli with reinforcement. Psychological Review, 82, 276-298. Mellgren, R. L. (1982). Foraging in simulated natural environments: There’s a rat loose in the lab. Journal of the Experimental Analysis of Behavior, 38, 93-100. Miller, D. B. (1985). Methodological issues in the ecological study of learning. In T. D. Johnston & A. T. Pietrewicz (Eds.), Issues in the ecological study of learning (pp. 73-95). Hillsdale, NJ: Erlbaum. Nash, S., & Domjan, M. (1991). Learning to discriminate the sex of conspecifics in male Japanese quail (Coturnix coturnix japonica): Tests of “biological constraints.” Journal of Experimental Psychology: Animal Behavior Processes, 17, 342-353. Nash, S., Domjan. M., & Askins, M. (1989). Sexual-discrimination learning in male Japanese quail (Coturnix coturnix japonica). Journal of Comparative Psychology, 103, 347-358. Noble, G. K., & Vogt, W. (1935). An experimental study of sex recognition in birds. Auk, 52, 278-286. Nyby, J., Bigelow, J., Kerchner, M., & Barbehenn, F. (1983). Male mouse (Mus musculus) ultrasonic vocalizations to female urine: Why is heterosexual experience necessary? Behavioral and Neural Biology, 38, 32-46. Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and nonreinforcement. In A. H. Black & W. F. Prokasy
186
Michael Domjan
(Eds.), Classical conditioning 11: Current research and theory (pp. 64-99). New York: Appleton-Century-Crofts. Rozin, P., & Kalat, J. W. (1971). Specific hungers and poison avoidance as adaptive specialuations of learning. Psychological Review, 78, 459-486. Sachs, B. (1966). Sexual aggressive interactions among pairs of quail (Coturnk coturnix japonica). American Zoologist, 6, 559. Sachs, B. (1967). Photoperiodic control of the cloaca1 gland of the Japanese quail. Science, 1.57, 201-203. Schliger, B. A,, Palter, B., & Callard, G. (1987). A method to quantify aggressiveness in Japanese quail (Coturnix c. japonica). Physiology & Behavior, 40, 343-348. Seligman, M. E. P., & Hager, J. L. (Eds.) Biological boundaries of learning. New York Appleton-Century-Crofts. Sherry, D. F. (1988). Learning and adaptation in food-storing birds. In R. C. Bolles & M. D. Beecher (Eds.), Evolution and learning (pp. 79-95). Hillsdale, NJ: Erlbaum. Shettleworth, S. J. (1972). Constraints on learning. Advances in the Study of Behavior, 4, 1-68. Shettleworth, S. J. (1983). Function and mechanism in learning. In M. D. Zeiler & P. Harzem (Eds.), Advances in analysis of behaviour. Vol. 3: Biological factors in learning (pp. 1-39). Chichester, England Wiley. Shettleworth, S. J. (1988). Foraging as operant behavior and operant behavior as foraging: What have we learned? The Psychology of Learning and Motivation, 22, 1-49. Shettleworth, S. J., & Krebs, J. R. (1982). How marsh tits find their hoards: The roles of site preference and spatial memory. Journal of Experimental Psychology: Animal Behavior Processes, 8,354-375. Timberlake, W. (1983). The functional organization of appetitive behavior. Behavior systems and learning. In M. D. Zeiler & P. Harzem (Eds.), Advances in the analysis ofbehaviour. Vol. 3. Biological factors in learning (pp. 177-221). Chichester, England: Wiley. Timberlake, W. (1990). Natural learning in laboratory paradigms. In D. A. Dewsbury (Ed.), Contemporary issues in comparative psychology (pp. 31-54). Sunderland, M A Sinauer. Timberlake, W. (1994). Behavior systems, associationism,and Pavlovian conditioning. Psychonomic Bulletin & Review, I , 405-420. Timberlake, W., & Grant, D. (1975). Auto-shaping in rats to the presentation of another rat predicting food. Science, 190, 690-692. Timberlake, W., & Lucas, G. A. (1989). Behavior systems and learning: From misbehavior to general principles. In S. B. Klein & R. R. Mowrer (Eds.), Contemporary learning theories: Instrumental conditioning theory and the impact of biological constraints on learning (pp. 237-275). Hillsdale, NJ: Erlbaum. Timberlake, W., & Silva, F. J. (1994). Observation of behavior, inference of function, and the study of learning. Psychonomic Bulletin & Review, I , 73-88. Westbrook, R. F., Good, A. J., & Kiernan, M. J. (1994). Effects of the interval between exposure to a novel environment and the occurrence of shock on the freezing response of rats. Quarterly Journal of Experimental Psychology, 478, 427-446. Wilson, M. I., & Bermant, G. (1972). An analysis of social interactions in Japanese quail (Coturnix coturnix japonica). Animal Behaviour, 20, 252-258. Zamble, E., Hadad, G. M., Mitchell, J. B., & Cutmore, T. R. (1985). Pavlovian conditioning of sexual arousal: First- and second-order effects. Journal of Experimental Psychology: Animal Behavior Processes, 11, 598-610. Zamble, E., Mitchell, J. B., & Findlay, H. (1986). Pavlovian conditioning of sexual arousal: Parametric and background manipulations. Journal of Experimental Psychology: Animal Behavior Processes, 12, 403-411.
EMOTIONAL MEMORY The Effects of Stress on “Co01” and “Hot” Memory Systems Janet Metcalfe W. Jake Jacobs
The memory processes used and the systems that dominate under unemotional laboratory conditions-the conditions under which much of the psychological data on human memory has been obtained-may be crucially different from those that dominate when a person experiences acute stress or even trauma. A variety of findings suggest that there may be two rather different but interacting memory systems.A “cool” hippocampally centered cognitive system encodes complex, bound representations of episodes replete with spatio-temporal and contextual details. In addition, though, there appears to be a separable “hot” amygdala-centered emotional system, which processes unbound, fragmentary emotional trigger stimuli connected to conditional fear responses. Experimental evidence bearing on the issue of memory under stress stems from two distinct traditions within psychology: research in cognitive psychology investigating human learning and memory, and a tradition focused on animal learning and physiological psychology. Studies within the former tradition have usually been conducted under stress-free or lowstress conditions. In contrast, studies in the latter tradition often explicitly use stressors (food deprivation-to allow food to be used as reinforcers, foot shock, etc.) as part of the manipulation, and may also use pharmacological and lesion manipulations that are rare in human studies (although, as THE PSYCHOLOGY OF LEARNING AND MOTIVATION, VOL. 38
187
Copyright Q 1998 by Academic Press. All rights of reproduction in any form reserved. W79-7421/98S25.M)
188
Janet Metcalfe and W. Jake Jacobs
will be noted shortly, several important cognitive neuroscience studies with brain-damaged patients provide some converging human data). Of course, the existence of two different research subdisciplines does not necessarily imply that there are different mental systems involved, but we suggest that in this case the data now indicate that there may indeed be two memory systems. The data, that will be reviewed later, suggest that the hot and cool systems select different aspects of the same nominal stimulus. The results of processing in these two systems are superimposed at the level of consciousness to yield a cognitive-emotional representation in which the coolsystem components form a cognitively informed spatio-contextual web and the hot system superimposes a distinct enhancement of the emotionally significant fragments. Although the hot and cool systems interface at the level of conscious awareness, both systems may have behavioral implications (semantic priming in the cool system; fear conditioning in the hot system, for example) prior to awareness. The joint cognitive-emotional representation is encoded and may be entered into memory, with the weighting of emotional fragments or of contextual-informational content encoded in the representation depending on the hot and cool system activation, respectively. This activation, in turn, depends upon the stress levels experienced by the person. Data are reviewed here that suggest that as stress levels increase, the “hot” emotional-amygdala system becomes increasingly activated. In contrast, although the “cool” episodic-hippocampal system is responsive at low levels of stress, at traumatic levels it is disregulated. We will illustrate this conceptualization by showing how it explains memory phenomena that fall under the rubric of weapon focus and the fragmentation of memory under conditions of traumatic stress.
I. Hot and Cool Memory Systems One of the clearest examples of the distinction we outline here between a “hot” and a “cool” system has been given by Bechara, Tranel, Damasio, Adolfs, Rockland, and Damasio (1995). They contrast the behavior of a patient WC1606, who has extensive bilateral hippocampal damage resulting from severe ischemic anoxia with that of patient SM046, who has a selective bilateral amygdala lesion secondary to the rare Urbach-Wiethe disease. Both patients were given a series of trials in which four monochromatically colored slides were presented. The blue slides were followed by a startling aversive blaring boat horn delivered at 100dB. WC1606, the hippocampally impaired patient, was unable to produce any evidence of memory for the pairing of the blue slide with the horn; that is, relational knowledge about
Emotional Memory
189
the event was missing. However, he showed a normal change in the galvanic skin response to the blue slide, indicating normal conditioning. In contrast, SM046’s relational memory for the events was normal-she could report that the blue slide was followed by the boat horn-but she showed no conditioning. This kind of dissociation between fear conditioning, on the one hand, and episodic memory, on the other, suggests that two separable systems may be involved. The cool system-the first system that will be discussed here-has been extensively investigated by cognitive psychologists. Decades of experimental research in combination with detailed computational models have helped to specify the processes, operations, and representations that are needed to allow people to remember events. A. CHARACTERIZATION OF THE COOLSYSTEM: BINDING IN EPISODIC MEMORY The cool system encompasses the nonemotional or cognitive functions that are needed to perceive, encode, store, monitor, update, evaluate, and retrieve events in memory. Complex thinking and self-control processes also rely on the cool system. Anatomically, it involves the hippocampus and frontal lobes, as well as other cortical areas involved in nonemotional perceptual and memory processing. Although the cool system encompasses many operations that are of crucial importance in other contexts-novelty monitoring, semantic priming, problem solving, metacognition, control processes, rehearsal, retrieval, planning, self-reflective monitoring, categorization, comprehension, verbal labeling, and so on-here we will emphasize one particular cool-system operation-episodic binding (see Metcalfe, Cottrell, & Mencl, 1992; Metcalfe, Mencl, & Cottrell, 1994). The binding operation is essential in ensuring that memory for events is episodic-that the elements that initially co-occurred are remembered as having been together in the unique combinations that specify particular events. The reason for this particular focus is that the hippocampus appears to be particularly affected by high stress levels. Along with a number of other researchers (Metcalfe et al., 1992,1994;Chalfonte &Johnson, 1994; Rudy & Sutherland, 1989,1994), we suggest that a critical operation that the hippocampus performs is episodic memory binding.
1. Effects of Hippocampal Damage Human patient studies (Kroll, Knight, Metcalfe, Wolf, & Tulving, 1996) suggest that the hippocampus is implicated in binding (viz., the encoding and memory of the configuration of the elements of an event, Metcalfe, 1993, 1997; Metcalfe et al., 1992, 1994). Kroll et al. (1996) showed
190
Janet MetcaUe and W. Jake Jacobs
patients with unilateral or bilateral damage to the hippocampus and some surrounding structures a series of stimuli-words, visual-spatial designs, and line drawings of faces. Subjects were asked to make old or new judgments about the old stimuli, completely new stimuli, and also about conjunction stimuli in which all of the elements were old, and in which the objects or words made sense or were structurally correct, but which were not in the configuration that had originally occurred. For example, if subjects saw VALLEY and BARTER in the list, they could be tested, in the conjunction condition with BARLEY. Similarly, when they studied the line drawings of faces, the eyes and nose of one studied face would be conjoined with the face outline, mouth, and hair of another, to form the conjunction stimuli. Although people with unilateral damage to the hippocampus (and some surrounding structures) were close to normal on the entirely old or new probes (see also, Mishkin & Murray, 1994,for a similar result with primates), they showed a particular impairment on the conjunction stimuli requiring the binding of the elements of an event. Left hippocampally damaged patients showed impairment on the verbal conjunctions, and both left and right hippocampally damaged patients showed impairment on the faces and visual-spatial design stimuli. The people with hippocampal damage were more likely than controls to call these new configurations old, indicating that they were responding to the oldness of the elements but that they had difficulty in binding together these elements. These data, then, are consistent with the idea that this central structure in the cool memory system binds the elements of an event into a coherent episode. Converging evidence on binding in the hippocampus also comes from experiments with rats. For example, rats with bilateral hippocampal lesions fail to remember the location of a platform hidden under the surface of opaque, milky water (Morris, Garrud, Rawlins, & O’Keefe, 1982), or the location of rewards in a radial arm maze-spatial tasks that may require memory for relational information. (Note that Abrahams, Pickering, Polkey, & Morris, 1996 have shown that patients with unilateral damage to the right hippocampus also show spatial impairments.) Although both hippocampal and intact rats can discriminate stimulus A and stimulus B, intact rats can also concurrently discriminate a stimulus consisting of a conjunction between A and B, whereas hippocampal rats cannot (see Barnes, 1988). A number of other tasks that seem to require conjunctive coding and learning show impairments and sometimes cannot be learned at all in rats with hippocampal damage. For example, tasks such as negative patterning (a discrimination between A + , B+, and AB-), transverse patterning (a discrimination among A+ vs. B-, B + vs. C-, and C+ vs. A-), biconditional discriminations (AB+, CD+, AC-, DB), and contextual conditioning show impairment with hippocampal damage in rats. However,
Emotional Memory
191
hippocampally damaged rats can still perform some nonlinear problems such as neutral discriminations (i.e., discriminations between AC+, B+, AB -, and C-), and some biconditional discriminations. Although some models, such as CHARM (Metcalfe et al., 1994), make no predictions and no claims about performance on these learning tasks, others, such as that of Sutherland and Rudy (1989), are disconfirmed by these findings, causing Rudy and Sutherland (1995) to propose a modified binding model. Single-cell recording studies have isolated neurons known as “place cells” in the CA1 region of the hippocampus, which become active when the animal is in a specific spatial location (O’Keefe & Dostrovsky, 1971). As Barnes (1988) showed, the CA1 place neurons of animals in a room with various sound sources, odors, and numerous objects (such as curtains, cotton wool, a polyethylene tube, levers, a black-and-white striped board, a polygraph, and fans) continued to fire normally even when individual items were removed from the room. In contrast, when the configuration of the objects, or the conjunctive relations among them, was altered-by scrambling their positions-these cells stopped firing. Barnes (1988) concluded that these cells code conjunctive relations. If binding is one of the critical functions of the hippocampus, it should be susceptible to failure if the hippocampus were to become dysfunctional, whether because of lesion (as in the above cited study by Kroll et al., 1996) or because of stress (as will be elaborated in more detail later in this article). Memory for the elements of an episode may persist intact under such conditions. However, the configuration of an event-what features went together in what arrangement-should be lost, leading to an impairment of memory that might be rather subtle. Most recognition memory tests fail to evaluate binding. Typically, old items are pitted against new items. Usually, though, none of the new items are reconfigurations consisting of old features. A person could perform perfectly on such a recognition memory task without ever having bound the features of an episode together. Furthermore, in a real-world setting, even in the face of binding failure, people’s feelings of familiarity, providing the basis for recognition memory, could stem from activation of the elements in isolation. Sometimes these elements could have occurred together. However, equally, they could have occurred in different events, provoking feelings of familiarity (and recognition judgments) attributable to elements from different episodes. In recall tasks, elements of different episodes could, in principle, be overtly reported as if they came from a single episode, if the story so formed was sufficiently plausible. Alternatively, though, recall in the absence of binding might appear fragmented, because the parts, but not their relationships to one another, might be remembered. We will later report on several clinical cases that suggest such fragmentation of memory
192
Janet Metcnlfe nnd W. Jnke Jacobs
under stress. First, though, we will outline the characteristics of the other system-the hot system that appears to function rather well under conditions of stress.
B. CHARACTERIZATION OF THE HOTSYSTEM: EMOTIONAL MODULATION, FEARCONDITIONING, AND TRIGGER FEATURE AMPLIFICATION Several researchers have proposed that the second memory system, an amygdala-centered system, handles certain aspects of fear-related perception, responding and learning (Damasio, 1994; Davis, 1992; LeDoux, 1989) and assigns emotional meaning and feeling tone to stimuli (Posner & Raichle, 1994; Halgren, 1992; Weiskrantz, 1956). This second system may have great importance in allowing us to understand the special qualities of emotional memory. 1. Emotional Impairment
Aggleton (1992) noted that amygdala lesion has been used as psychosurgery, to decrease the emotional responsivity, aggressiveness,and impulsivity of psychiatric patients-sometimes with apparent success, although the presurgical abnormalities of such patients prohibit rigorous and precise assessment. Lee, Arena, Meador, and Smith (1988) showed a decrease in autonomic responsiveness in one such patient, and suggested that this diminution could be associated with reductions in episodes of aggression. Halgren (1981), in his review of the literature, indicated that people with damage to the amygdala often exhibit a loss of fear and a general decrease in emotional tension. A tameness, indifference, and loss of emotional responsiveness has been shown to occur in amygdalectomized monkeys (Kluver & Bucy, 1939; Weiskrantz, 1956).These impairments may result in a loss of the modulation of social behavior. For example, wild monkeys with uncus and amygdala lesions (Dicks, Myers, & Kling, 1969) exhibited disrupted social behavior, rarely initiated social interactions with their troop, and remained socially isolated. Aggleton and Passingham (1981), Kling (1972), Rosvold, Mirsky, and Pribram (1954), and Weiskrantz (1956) have reported similar responses. In the Dicks et al. (1969) study, the isolation, the lack of submissive and aggressive behavior, and the other consequences of amygdala lesion, had devastating consequences for the affected animals, who, with the exception of two juveniles who survived and rejoined the social group, eventually died. Downer (1961) studied monkeys with lesions placed in the optic chiasm, one amygdala, and a forebrain commissure, so that each hemisphere received information only from the contralateral eye. These animals exhibited
Emotional Memory
193
species-typical defensive displays and appeared wild when viewing fearprovoking objects with the eye connected to the intact amygdala. They failed to exhibit defensive responses and appeared tame when viewing the same objects with the eye connected to the damaged amygdala, indicating an emotional component to amygdala functioning. SM046, the amygdala patient studied by Bechera et al. (1995), had emotional impairments, and mildly inappropriate behavior as well. She also had difficulty discerning some kinds of emotional facial expressions.’ She rated expressions of anger, surprise, and especially fear as being less intense than did control subjects, and showed a severe recognition-of-expression impairment that was specific for fear (Adolphs, Tranel, Damasio, & Damasio, 1995). She had some other emotional abnormalities, and exhibited inappropriate behaviors on occasion. She was less able than normals to assess combinations of emotional expressions. Although she was able to draw most facial emotional expressions, such as happiness or even anger, when asked to draw the facial expression of fear, with reluctance and with repeated prompting, she drew a full-body picture of a crawling baby with its hair standing on end. She explained that she did “. . . not know what an afraid face would look like” (Adolphs et al., 1995, p. 5887). She was able to use words like afraid or frightened appropriately (though unemotionally) in text and speech, and she was normal in grouping together words such as “afraid,” “scared,” “worried,” “terrified,” and “alarmed.” Adolphs et al. (1995) suggested that although SM046 may have knowledge about fear, she does not experience it in the normal way. Some electrical stimulation studies have been conducted in humans as a part of the brain mapping needed prior to surgery. Stimulation of the amygdala elicits autonomic fear reactions and automatism (Feindel & Rasmussen, 1991; Gloor, Olivier, & Quesney, 1981), and people report feelings of fear. Chapman, Schroeder, Geyer, Brazier, Fager, Popper, Solomon, and Yakovlev’s (1954) patients reported a variety of emotional responses including fear, anxiety, and feeling both “weird” and “terrific.”
2. Effects on Fear Conditioning In contrast to the hippocampal patient, SM046 showed severe impairments in fear conditioning and responding (Bechera et al., 1995). As noted above, Hamann, Stefanacci, Squire, Adolphs, Tranel, Damasio, and Damasio (1996) reported normal recognition of facial expression in two patients who had bilateral amygdala and temporal lobe lesions that were sustained well into adulthood, rather than in childhood, as had been the case for SM046. The authors suggested that the age at which a lesion to the amygdala is sustained may be critical, noting that a fourth patient, D. R., who, like SM046 suffered an amygdala lesion in childhood (along with damage to the basal ganglia), also showed impaired recognition of facial expression.
194
Janet Metcalfe and W. Jake Jacobs
SM046 could report the pairing of the blue slide and the loud noise, but failed to produce the conditioned skin response. Unlike people with hippocampal damage, SM046 had no difficulty remembering new information at generic and specific levels-including new objects and the identity of new faces. She could also keep track of and remember temporal relations. Her cross-modal associations were unimpaired (provided these were not tested with methods requiring appropriate processing of a reward that entails emotional processing, Lee, Meador, Smith, Loring, & Flanigan, 1988). But she did not condition to fear. Converging data concerning the function of the amygdala (and related structures such as the bed nucleus of the stria terminalis and the periaquiductal gray) in processing emotional and especially fear evoking stimuli come from many studies conducted on rodents (Davis, Rainnie, & Cassell, 1994; LeDoux, 1993, 1995; Davis, 1992). Many studies have shown that lesions to the amygdala impair conditional fear responses (Davis, 1992; Kapp, Whalen, Supple, & Pascoe, 1992; LeDoux, 1993), fear learning (Davis, 1992), fear-potentiated startle (Hitchcock & Davis, 1986), learned deceleration of heart rate (Gentile, Jarrell, Teich, & Schneiderman, 1986), and pain sensitivity (Helmstetter, 1992). 3. Emotional Fragmentary Stimuli LeDoux (1995) has shown that the emotional system initially processes fragmentary stimulus features arriving from thalamus to the amygdala through a single synapse. It is thought that these innately determined behavioral triggers release species-specificbehavior such as orientation, fear, and defensive responses (Davis, 1992; Fanselow, 1994), and serve as the basis for classically conditioned fear learning (Davis, 1992), as well as reward learning (Gaffan, 1992). Jacobs and LoLordo (1977,1980) have shown that classical conditioning is easier for particular kinds of stimulus features such as looming stimuli, sudden noises, or abrupt movements-stimuli that appear to be species specific, survival specific, and fragmentary rather than representationally replete. Studies using single cell recordings in response to complex sensory stimulation have shown that individual neurons in the amygdala are responsive to stimuli with clear emotional value (B. L. Jacobs & McGinty, 1972; O’Keefe & Bouma, 1969; Rolls, 1986). For example, B. L. Jacobs and McGinty (1972) recorded responses from particular amygdala neurons in cats while exposing the animals to either simple stimuli, such as clicks, tones, white noise, etc., or complex stimuli, such as dog barks, whistles, cat food, and hand and voice threats by the experimenter. Ten of the 57 cells studied responded selectively to particular complex emotional stimuli-a
Emotional Memory
195
dog bark, the clang of a garbage can cover, the experimenter threatening the cat with a hand, a rat moving about the chamber, a cat howl. C. INTERACTION BETWEEN
THE
HOTAND COOLSYSTEMS
Although several researchers have indicated that there may be two systems, only the McGaugh-Cahill group has concentrated on the interaction between them (see, Cahill & McGaugh, 1996). They have proposed that the amygdala acts as a modulator of other structures, including both the hippocampus and the caudate nucleus. Activation of the amygdala serves to turn up the gain on the other memory-relevant structures in which the processing may be quite selective in nature. This function is similar to that given in psychological theories of emotion in which arousal is the underlying state that determines the extent of felt emotion, but the nature or content of the emotion experienced involves cognitive labeling (Schachter & Singer, 1962). The amygdala itself, though, in this view, is seen as unselectivefunctioning as a modulator rather than a structure enacting particular learning functions, and highlighting particular features. The modulation hypothesis was based, in part, on a study by Packard, Cahill, and McGaugh (1994) in which rats received D-amphetamine injections into the hippocampus, the caudate nucleus, or the amygdala. Rats were given either a spatial learning task-in which the hippocampus is particularly implicated-or a cued water maze task, in which the platform is clearly visible, a task in which the caudate nucleus is implicated. When the hippocampus was injected, performance on the spatial task improved. When the caudate was injected, performance on the visibly cued task improved. When the amygdala was injected, performance on both tasks improved. We propose that, rather than being global, the amplification from the hot system is usually selective on particular activated hot features. It seems likely that the entire amygdala would be excited only rarely, and thus the situation given in the injection experiment would be atypical (although, if it did occur-perhaps for pharmacological reasons-then a global modulation as posited above should result). More typically, though, those amygdala neurons responsive to the particular situation the animal or person faced, such as those that fire to the dog bark or the cat howl (illustrated in the single cell experiments), or looming or threatening species-specificstimuli (illustrated in the experiments on preferential conditioning) are likely to become selectively activated. These specifically activated amygdala neurons should selectively influence corresponding features at the level of conscious awareness, and selectively amplify such content in memory. The person’s conscious state, and the memory trace that is thereby encoded, could be conceptualized as a superposition of the processing of the
196
Janet Metcalfe and W. Jake Jacobs
hot-amygdala and the cool-hippocampal systems. The hot system, in the present framework, highlights the species-specific or learned fragments that are highly emotional in nature, along with the feeling states-such as fear-that accompany such fragments. The cool system is emotionally and informationally more neutral-providing the background events and configurations, that are bound up into a spatio-temporally coherent narrative frame. The resultant representational array is contextually coherent (to the extent that the cool system is active at time of encoding) with the hot fragments emphasized and the feeling states included. Though the component elements stem from two different systems, systems that can be separable under some special circumstances (such as lesion to the amygdala or the hippocampus), a unified conscious state is the result. In normal people, we suggest that the two components thoroughly interact in a superimposed manner so that they are inextricably combined-providing contextual information, learned responses and feelings and emotions in a coherent integration.
D. RELATION OF THE HOT-COOL FRAMEWORK TO OTHER VIEWS The hot system-cool system framework proposed here could be related to a number of other views of human memory. For the sake of clarity, some of the distinctions between this view and others will be briefly outlined. First, we will outline the CHARM model-a model that has been applied to much of the human memory data from laboratory studies. Then we will contrast the hot-cool view to the implicit-explicit, the proceduraldeclarative and the taxon-locale views of memory. 1. CHARM
The model of human memory called the CHARM model (composite holographic associative recall recognition memory) (Metcalfe Eich, 1982,1985; Metcalfe, 1990,1991,1993aand b, 1994,1996,1997),like most other memory models emerging from cognitive psychology, has been directed exclusively at the hippocampal-temporal lobe (and frontal lobe) system, which we here call the cool system. No hot-system functioning has been included in this model, and thus, it gives a summary of the “cool” human memory system in isolation from fear conditioning, and the amplification of fear-related features proposed here. Within the domain of laboratory studies and information processing in relatively neutral emotional states, the model does a good job of accounting for a large body of the data on human memory, including paired associate learning, the Osgood similarity surface, prototype formation, the independence of B and C responses in the A-B A-C paradigm (and does not suffer from the catastrophic interference that so plagues
Emotional Memory
197
other models in that paradigm, but rather gives the appropriate B-C tradeoff function, Metcalfe Eich, 1982), encoding specificity effects, levels of processing effects, effects of elaboration, the fan effect (Metcalfe Eich, 1985), von Restorff effects, release from proactive inhibition, novelty effects, spacing effects, primacy effects, the p300 event-related potential, habituation effects (Metcalfe, 1993 a and b, 1994), and a template for understanding many of the effects of different kinds of amnesia (Metcalfe 1996). Interestingly, the model can account for studies on implanting information in the eyewitness testimony paradigm including both the standard and the modified test procedures (Metcalfe, 1990; Metcalfe & Bjork, 1991). These results are sometimes taken as the source of people’s judgments that they have recovered repressed memories-though no such construct as repression is operant in the model. The results from these paradigms can be explained without taking account of any special features or alterations in function that might result from emotional processing. However, that is not to say that no such special processes occur-only that the standard laboratory procedures do not normally evoke them. The model components that are needed to allow recognition, recall, and rehearsal include a perceptual-lexical processing and pattern identification component, consciousness or working memory, an association formation operation (which in the model is given by the mathematical operation of convolution), composite memory storage, a novelty monitoring and control feedback loop, and a retrieval operation (which in the model is given by the mathematical operation of correlation). It is assumed that events that impinge upon the senses are perceptually analyzed and may be characterized as patterns of features, or as vectors, and that, in the case of verbal input, may be configured as lexical “units.” Plasticity may occur in this pre-episodic system (consistent with the views of other theorists such as Mishkin & Petri, 1984;Moscovitch, 1982; O’Keefe & Nadel, 1978; Schacter, 1987; Squire, 1994; Tulving, 1985; who distinguish among a core episodic memory system and other more peripheral systems that underlie skill effects, priming effects, comprehension, motor and linguistic competence and learning). A well-established phenomenon in the amnesia literature is that repetition and associative priming measured by lexical decision, fragment completion, homophone spelling, free association, and word identification tasks may be spared even though recall, and sometimes recognition memory-episodic memory tasks-are impaired (Buschke, 1965, 1984; Gardner, Boller, Morienes, & Butters, 1973; Graf, Shimamura, & Squire, 1985; Graf, Squire, & Mandler, 1984; Jacoby & Witherspoon, 1982; Moscovitch, 1982; Schacter, 1987; Shimamura, 1986). This result is consistent with the structure of the model in which the perceptual system is prior to the episodic system.
198
Janet Metcalfe and W. Jake Jacobs
At a particular level of representation phenomenological awareness emerges. This level of representation is a necessary but not sufficient condition for entry into the episodic memory system. Its insufficiency is apparent insofar as amnesics have unimpaired conscious awareness-suggesting that they can represent events at this level-and yet are unable to store and retrieve such events in memory. Its necessity is shown by data indicating that events that are not consciously processed (i.e., do not reach this level) are not retrievable from episodic memory (although, as the implicit memory literature suggests, they may influence behavior without being rememberable as events that happened to the individual). In CHARM two items, represented as vectors, may be associated, or an item may also be associated with itself. Interitem associations are needed to allow one item to provide a retrieval cue for another item; autoassociations underlie recognition memory, allowing an item to retrieve a representation of itself. The operation used for association formation-convolution-is a binding operation whereby all of the various parts of a representation become linked or related to one another (see, Metcalfe et al., 1992,1994). Convolution is assumed to be enacted by the hippocampus. If the hippocampus becomes dysfunctional, either because of stress or anatomical insult, we should be able to understand and analyze the implications by investigating the predictions related to the operation of convolution. Multiple associations are stored by being superimposed in a composite memory trace. A novelty monitoring and control circuit assesses the novelty of the incoming association with respect to the preexisting composite trace, computes a global familiarity value, and sends a signal to adjust the weighting on the incoming event as a function of its novelty (Metcalfe, 1992) and produces an adaptive system responsive to novel and relatively unresponsive to familiar events. This novelty monitor may be responsible for familiarity-based feeling-of-knowing judgments (Metcalfe, 1993a and b, 1994; Metcalfe, Schwartz, & Joaquim, 1993; Reder & Ritter, 1992; Schwartz & Metcalfe, 1992), release from proactive inhibition (Cermak, Butters, & Morienes, 1974;Winocur, Kinsbourne, & Moscovitch, 1981) von Restorff effects, p300s (or sometimes the “late positively,’’ Sutton, Braren, Zubin, & John, 1965; Picton, 1992), some non-rehearsal based primacy effects (see Metcalfe & Murdock, 1981), habituation effects, and spacing effects (Metcalfe, 1996). When a retrieval cue is given, it is perceptually analyzed and used to retrieve a representation from the composite trace. The retrieval operation of correlation is the inverse operation to convolution. The item that is retrieved from memory is in a form available to phenomenological awareness. Retrieval has the characteristic, in this model, of being redintegrative: a part of an event is sufficient to retrieve the whole. Similarity effects
Emotional Memory
199
fall out automatically. Furthermore, everything that was associated with a particular retrieval cue will be retrieved by that cue-resulting in the superimposed retrieval of multiple events that allows the model to account for such diverse phenomena as episodic classification learning and generalization (Metcalfe Eich, 1982), similarity effects (Metcalfe Eich, 1985) interference in the A-B A-C paradigm (Metcalfe Eich, 1982), and the effects of misleading suggestions on eyewitness testimony (Metcalfe, 1991). The model takes account of certain kinds of content. For example, it is sensitive to the similarity structure between and within items. It also responds to novelty. However, it does not take into account the biological importance of particular environmental cues or features. It gives no special priority to cat screeches or looming stimuli, to shows of dominance or gestures of threat-the kinds of trigger features that are selectively processed by the amygdala. The model, as it has been developed in the past, is informationally neutral-not giving preferential treatment to emotional input either representationally or in terms of the processes enacted. A consequence of this is that the model cannot currently account for the data from studies in which a particular emphasis is given to emotional features, for example, weapon focus studies. It also cannot account for changes in memory as a result of stress. Although the model, then, has explained a wide range of data, and has considerable generality, it is strictly a coolsystem model. Because of the obvious importance of emotional modulation, its addition is essential to allow the model to more accurately reflect ecologically valid human memory behavior. Such a modification is currently in progress (Metcalfe & Jacobs, 1998). 2.
The Implicit Memory-Explicit Memory View
Schacter (1987) and Graf and Schacter (1985) have proposed an important distinction between implicit and explicit memory that is relevant to, but different from, the one between hot and cool. Explicit tasks are those that are enacted consciously as remembering tasks, and that involve conscious awareness about remembering. Implicit tasks can benefit (or be harmed) by learned information about which the subject has no awareness. Such tasks do not require conscious awareness about remembering. To give an example, Hayman and Tulving (1989) provided subjects with a list of words to study-later to be tested either implicitly or explicitly. At the time of the test subjects were given a number of letters, or a fragment, from each of the words on the list. The instructions given to elicit explicit memory were to complete the fragment by remembering the word from the list. The implicit-memory task given to subjects was to complete each fragment by saying the first thing that came to mind that would form a word. Amnesics can often perform normally on implicit tasks, but not on explicit tasks.
200
Janet Metcalfe and W. Jake Jacobs
Some implicit-memory tasks, such as fear conditioning, would be classified to be hot. Others, such as categorical learning or semantic priming tasks, would be considered to be cool. Furthermore, both the hot and the cool system make a contribution to autobiographical recall-usually thought to be an explicit task. The potential extent of the contribution of each to retrieval of autobiographical memories, in the present view, depends on the degree of activation of the hot and the cool systems at time of encoding. Unlike in the implicit-explicit distinction, we do not consider either the hot or the cool components necessarily to be unconscious (or, indeed, conscious). A person can be fully aware of feeling states and emotional responses, just as he or she can be fully aware of the informational content of memories. Similarly, the emotional aspects of an event could be weak enough to allude consciousness, just as informational components of an event could be activated to a level below the threshold of consciousness. Although we recognize the importance of consciousness in people’s mental life (Tulving, 1985), the conscious-nonconscious distinction is different from the hot-cool distinction. 3. Procedural or Declarative
Cohen and Squire (1982, and see Squire, 1994) have proposed that human memory may be divided into declarative-or “knowledge that” systemsand procedural-or “knowledge how” systems. Although some declarative knowledge tends to be cool, there are other kinds of cool knowledge (such as lexical knowledge, semantic knowledge, or mathematical knowledge) as well. Insofar as procedures are active, some aspects of the responses typical of the hot system (such as flight from a fear-provoking situation) are procedural. However, there are many procedures-such as problem solving, or riding a bicycle, or enacting mirror drawing-that are unemotional, and hence would not be considered to be hot. 4. Taxon or Locale
Following O’Keefe and Nadel (1978), Jacobs and Nadel (1985) suggested a distinction between a locale (or hippocampal place learning) system, and the taxon systems, responsible for all other “taxonomic” and cue learning. The hot-system-cool-system framework differs from the taxon-locale view insofar as the locale system is only one aspect of the cool system. We consider that other subsystems (metacognitive, strategic, and control systems, short-term memory systems, semantic learning, language learning, mathematical learning, and comprehension systems) are cool. The hot system-centered on the amygdala, and possibly the orbital and dorsolateral
Emotional Memory
201
frontal regions-might be considered to be a taxon system. But in the O’Keefe and Nadel framework there are a number of other learning systems, such as those involved in object recognition, semantic learning, language learning, and classification,which they would consider taxon systems but which we would consider “cool.” Even so, to the extent that O’Keefe and Nadel (1978) actually focused on the function of the hippocampus itself, their work is of significance for the present framework. Jacobs and Nadel (1985) drew out two related conjectures important for the hot-cool framework. The first is that the hippocampus is relatively late in developing, whereas taxon systems are fully functioning at birth. The developmental difference between the hippocampus and the amygdala has received considerable empirical support (Rudy, 1993). The second conjecture is that the hippocampus responds differently to stress than do other brain regions. Jacobs and Nadel (1985) suggested that the hippocampus becomes dysfunctional at extremely high levels of acute stress. The physiological data for this conjecture were, in 1985, somewhat tentative. However, as will be reviewed in the next section of the paper, the ensuing 13 years of research seem to have borne out this hypothesis. Jacobs and Nadel (1985) used the combination of these ideas to explain how childhood phobias may reemerge in adulthood under conditions of acute stress. They assumed that when the hippocampal system was fully functioning, it suppressed the processing of the more primitive taxon systems. Thus, if a person experienced an intensely fear-provoking stimulus or situation in early childhood, it would be processed by the taxon system, but would later be suppressed once the hippocampal system was functioning. However, under conditions of acute stress, when the hippocampal system becomes dysfunctional, the taxon systems would again dominate processing and the fears and phobias of childhood could resurface.
11. Stress
The stressfulness of a particular event depends on the event itself, the person’s interpretation of the event, the person’s coping skills, psychological state, and personality or stable intrapersonal “if-then” relations, and his or her current physiological state (Cohen & Wills, 1985; Lazarus & Folkman, 1984; Mischel & Shoda, 1995). Viewing a violent film, being exposed to a loud noise, jumping from an airplane, or witnessing a crime may, of course, be experienced in different ways, from mildly arousing to highly stressful. Despite these modulating factors, there is some agreement concerning the overall ordering of the stressfulness of events for most people (Holmes & Rahe, 1967). There are some events-such as being
202
Janet Metcalfe and W. Jake Jacobs
severely beaten-that are for most people traumatic, and others-like surgery-that cause a physical stress response even when the person is unconscious. Stress can be defined as any event that upsets the body’s homeostatic state-regardless of whether the precipitating event or situation is physical or psychological. Autonomic responses-increased heart rate, dry mouth, glucocorticoid release, epinephrine and norepinephrine release, and other responses that prepare the person for fight or flightare markers. Most laboratory studies (see Christianson, 1992; Egeth, 1994, for review) and even naturalistic studies (c.f., Yuille & Cutshall, 1989) examine the effects on memory of only mild to moderate stressors. As van der Kolk and Fisler (1995) have noted, even when exposed to a movie depicting actual executions, normal college students do not exhibit post-traumatic stress symptoms. Most laboratory studies are much milder than these films. Trauma anchors the high end of the stress dimension. Classen, Koopman, and Spiegel(l993) described trauma as: “ . . .an abrupt physical disruption in ordinary daily experience, often with loss of control over the body. Natural disasters, war, rape, and other frightening experiences cause discontinuities in both physical and psychological experience. Trauma renders a person helpless as the world suddenly becomes unpredictable, threatening, and assaultive” (p. 179). In this section, we review evidence suggesting that a major stressor may impact differentially upon the hot system, and particularly the amygdala, and the cool system, and particularly the hippocampus. It appears that increasing stress, at all except the very highest levels, may result in facilitated processing in the hot system. In contrast, at high levels stress results in cool-system, and particularly hippocampal, disregulation. A. STRESS AND THE COOL SYSTEM
A stressor initiates a cascade of events that impacts upon both peripheral and brain physiology. The component of primary interest, with respect to the cool system in the present context, involves activation of the adrenal glands, which causes a release of glucocorticoids into the bloodstream. The glucocorticoids cross the blood-brain barrier in which they have a wide influence on neural activity, the most important of which, from the current perspective, is activity in the hippocampus. The hippocampus contains a high density of neurons sensitive to glucocorticoids (de Kloet, Oitzl, & Joals, 1993). Many neurons in the hippocampus, and especially in the CA1 region, have two types of receptors that bind glucocorticoids (i.e., corticosteroids): mineralcorticoid and the glucocorticoid receptors (McEwen, 1994), though the latter also have receptors more diffusely located throughout
Emotional Memory
203
the brain (see, Joals & de Kloet, 1992). One mechanism by which stress may influence hippocampal processing is via the mineralcorticoid and the glucocorticoid receptors. Some data indicate that increasing occupation of the mineralcorticoid receptors enhances neuronal excitability, whereas, increasing occupation of the glucocorticoid receptors in combination with the occupation of the mineralcorticoid receptors seems to decreasg the excitability of the hippocampal neurons (Cahill & McGaugh, 1996;Joals & de Kloet, 1990). The details of the interaction of these receptors and, thus, the exact workings of the mechanism mediating stress-related memorial effects in the hippocampus is thought to be complex, and is actively being investigated (see Roozendaal, Portillo-Marquez, & McGaugh, in press, for review). At low plasma levels of corticosteroids, few of either the mineralocorticoid or glucocorticoid receptors are occupied and hippocampal neurons are difficult to excite (Joals & de Kloet, 1989; Kerr, Huggett, & Abraham, 1994). As the levels of corticosteroids increase the mineralocorticoid receptors become occupied and the excitability of the hippocampus increases. As the glucocorticoid levels increase further, though, both mineralocorticoid and glucocorticoid receptors become occupied resulting in a decrease in the eqcitability the neurons (Cahill & McGaugh, 1996; de Kloet et al., 1993; Joals & de Kloet, 1990; Rey, Carlier, & Soumireu-Mourat, 1987; R e d & de Kloet, 1985). These processes are relatively slow (Kirschbaum, Diedrich, Gehrke, WAst, & Hellhammer, 1991). As is shown in Fig. 1, the net result of the increasing glucocorticoid levels in the hippocampus, with the two different receptor sensitivities depending upon overall glucocorticoid level, is an inverted U-shaped function that relates stress levels to hippocampal neuronal excitability.
c
0 .c m > .c
Hippampu-bsad SYrtSm
Amygdals-bued
syacn
Fig. 1. Idealized stress-response curves of the hot and cool systems.
204
Janet Metcalfe and W. Jake Jacobs
The extent of activation of the neurons in the hippocampus appears to be closely related to memory. Diamond, Bennett, Fleshner, and Rose (1992) have shown that changing levels of glucocorticoids affect synaptic plasticity via long-term potentiation (LTP) in the hippocampus, which is closely linked to memory. For example, when LTP in the hippocampus is blocked by the use of selective antagonists, rats’ spatial memory in a water maze is impaired (Kerr et al., 1994; Sakimura, Kutsudwada, Ito, Manabe, Takamyama, Kushiya, Yagi, Aizawa, Inoue, Sugiyama, & Mishina, 1995). Similarly, in genetically fyn mutant mice who have impaired LTP, spatial learning is also impaired (Grant, O’Dell, Karl, Stein, Soriano, & Kandel, 1992). Stressors such as restraint and inescapable tail shock are associated with both high-level release of glucocorticoids and a decrease in LTP in the rat hippocampus (Foy, Foy, Levine, & Thompson, 1990; Foy, Stanton, Levine, & Thompson, 1987; Shors, Seib, Levine, & Thompson, 1989). For example, Foy et al. (1987) restrained rats for thirty minutes during which time they presented mild inescapable shock to the tail. Compared to controls, these rats showed impaired LTP. Furthermore, Pavlides, Watanabe, and McEwen (1993) showed that when corticosteroids were administered to rats systemically even in the absence of an environmental stressor, LTP in the hippocampus was impaired. These data, taken together, suggest that the glucocorticoids produce both physiological effects at the level of synaptic plasticity, and measurable decrements in the rats’ cool-system memory. In summary: the physiology of the hippocampus under acute stress mirrors the standard Yerkes-Dodson (1908) function. If the memory encoding capabilities of the hippocampus follow the physiological inverted U-shaped function outlined above in humans, as the animal experiments described above indicate, then cool-system memory encoding should be impaired under conditions of stress.
B. STRESS AND THE HOTSYSTEM The aspect of the stress reaction that is of primary interest with respect to the hot system involves the activation of the locus coeruleus and the resultant release of norepinephrine into the amygdala and other structures. This occurs rapidly. The rate of synthesis and metabolism of norepinephrine in the central nucleus of the amygdala is increased by stress (Pacak, Palkovits, Kopin, & Goldstein, 1995), and the amygdala becomes increasingly responsive because of this increase in norepinephrine up to very high levels. Treatments that either increase noradrenergic transmission or decrease the action of compounds that inhibit noradrenergic transmission (e.g., opi-
Emotional Memory
205
ate and GABA transmission) have been shown to result in an increase in excitability of the amygdala (McGaugh, Introini-Collison, Cahill, Castellano, Dalmaz, Parent, & Williams, 1993). Under either of these conditions, fear is conditioned more easily (Davis et al., 1994). In contrast, treatments that increase the excitability of inhibitory neurons efferent to the amygdala decrease the ease of fear conditioning (Helmstetter & Bellgowan, 1994). When propranolol, which blocks the action of norepinephrine, was infused directly into the amygdala of rats, it inhibited fear conditioning (Introini-Collison, Nagahara, & McGaugh, 1989). These several lines of evidence indicate that increasing concentrations of norepinephrine in the amygdala increases learning and retention of conditional fear responding (McGaugh et al., 1993). The generally facilitative effects of stress on hot system function, suggested in Fig. 1, does have an upper limit (Laing, Juler, & McGaugh, 1986), but we suggest that it is at extremely high levels-higher than those that cause disregulation of the cool system. Although we have been unable to find any single experiment that has examined both the conjectured impairment of spatial learning-a cool-system function, and the improvement in classical conditioning-a hot-system function, at the same level of stress, Shors, Weiss, and Thompson (1992) have shown that classical conditioning is enhanced under stressful conditions of inescapable tail shock that in other studies has revealed impaired instrumental “cool” system learning. As they note: “These results stand in contrast to the learning deficits typically observed and suggest that stress can enhance the acquisition of discrete conditioned responses” (p. 537). It seems likely, then, that the disruption of functioning in the cool system occurs at a lower, and possibly much lower, level of stress than any disruption of the hot system. C. INTERACTION OF THE Two SYSTEMS UNDER STRESS The effects of stress on memory have often been considered controversial and confusing (Christianson, 1992; Egeth, 1994; Kassin, Ellsworth, & Smith, 1989) with stress sometimes enhancing (e.g., Bohannon, 1988; Yuille & Cutshall, 1989), impairing (e.g., Loftus & Burns, 1982; Peters, 1988), or doing nothing to alter memory (Cutler, Penrod, & Martens, 1987; Hosch & Bothwell, 1990). Such complicated effects may result in part because of the superimposed interaction of two systems that have differential processing specialties as well as different vulnerabilities to stress. As stress increases, the cool-memory system at first becomes increasingly responsive but then, as the level continues to grow, becomes less responsive until, at traumatic levels of stress, it becomes dysfunctional. The overall pattern is an inverted U-shaped function. In contrast, the hot system be-
206
Janet Metcalfe and W. Jake Jacobs
comes increasingly responsive to increasing levels of stress in a monotonic manner up to and including very extreme levels, breaking down only at extremely high levels. The balance between the fear responsive features of the hot system and the integrated spatio-contextual web of the cool system changes as a function of the level of acute stress. In Fig. 1, which shows the response of the hot system and the cool system to stress, we have divided the conjoint responses into three sections making three types of predictions: A, B, and C. Type A predictions occur at low to moderate levels of stress when processing in both systems is facilitated by increasing levels of stress. There is a point, though, at which the hippocampal system begins to decrease in function with increasing stress, while the amygdala system continues to be facilitated, resulting in type B predictions. Finally, at traumatic levels of stress, dominance shifts from the cool system to the hot system. It is with this shift in the usual dominance pattern that we get type C predictions-the predictions relevant to memory for trauma. D. IMPLICATIONS OF THE HOT-COOL FRAMEWORK FOR MEMORY UNDER STRESS 1. Memorial Effects of Increases in Stress at Low Levels:
Type A Predictions As is shown in Fig. 1, in the region of the stress-response curve labeled
“A,” increasing the stress or arousal level, at low levels of ambient arousal should result in enhanced processing in both the cool and the hot systems. Studies that show an effect of a general increase in memory as a function of stress-both for the emotionally balanced aspects of an event (most salient to the hot system), and for the surrounding spatio-temporal context (most salient to the cool system)-are rare. However, there are some. Often, in these studies, rather than having a separate factor that controls arousal independently of the to-be-remembered materials, arousal or stress level is manipulated by the same stimuli that are the emotional target events. For example, Heuer and Reisberg (1990) manipulated the arousal level by using emotionally neutral or emotionally stressful slides embedded in the center of a sequence of neutral slides. The slide sequence depicted a mother taking her son to watch the father at work. In the emotional version the father was a surgeon who performed an operation, which was displayed graphically. In the unemotional version the son also watched the father at work, but in this case the father was an auto mechanic. Subjects were tested at a two-week delay on both recognition and recall. Those exposed to the emotional story had better recall and recognition, both for the emotional and the unemotional parts of the slide sequence.
Emotional Memory
201
A similar study was conducted by Christianson (1984). In this case, the slide sequence involved a mother leaving a house with her seven-year-old son, and the pair getting into a cab, in the neutral condition, or the child being injured in a dreadful auto accident, in the emotional condition. Then, in both conditions, the mother made a phone call and returned home. Recall was tested after either 10 minutes or two weeks. At the 10-minute retention interval, there was no difference in recall between the emotional and unemotional conditions on the unemotional material that preceded and followed the emotional slide sequence. The emotional central sequence was recalled better than the unemotional central sequence, however. When testing occurred at a two-week delay, the unemotional central sequence, in the control condition, was forgotten, as might be expected. Interestingly, though, there was no forgetting of the analogous emotional central sequence, nor was recall of the surrounding unemotional slides impaired. The largest enhancing effect, then, occurred when testing was delayed. Burke, Heuer, and Reisberg (1992) mentioned three additional independent studies that show similar results: increased recall of both the emotional events and the nonemotional surrounding events at low levels of stress, though their own data do not show this pattern, as will be discussed later. The finding that the observed enhancement with stress is most likely to be found after a delay is of interest because enhancement effects attributable to hippocampal potentiation by low levels of glucocorticoids should not be apparent immediately insofar as the release and the uptake of the hormones is relatively slow (Stansbury & Gunnar, 1994; Jacobson & Sapolsky, 1991). If the testing occurs immediately after the event, the facilitating effect of the low levels of glucocorticoids may not yet be evident. Christianson’s (1984) and Heuer and Reisberg’s (1990) studies both showed some enhancement of memory with low stress at the two-week delay. Overall, then, these results-both the finding of overall enhancement at low levels of stress and the finding that the enhancement is especially pronounced at a delay of testing-are consistent with the predictions of the cool-systemhot-system framework. 2. Memorial Effects of Increases in Stress at Moderate to High Levels: Type B Predictions Increasing stress at moderate to high levels should result in impaired processing in the cool system coupled with enhanced processing in the hot system. The amygdala-based emotional trigger stimuli should become more dominant, and there should be concomitant decreases in the memory for the nonemotional spatio-temporal context processed by the hippocampus. This pattern conforms to the observations of Easterbrook (1959); it is also
208
Janet Metcalfe and W. Jake Jacobs
consistent with the effect sometimes dubbed “weapon focus,” though, as we shall argue shortly, it is unlikely that all of the effects in so-called weapon focus studies are attributable to moderate to high levels of stress. Easterbrook (1959) proposed that the degree of cue utilization narrows with increasing stress. In contrast, the present framework points to particular kinds of features that become increasingly dominant under stress: those preferred by the amygdala system. It is not simply any cues, but fearprovoking cues, or those that have been so conditioned, that are highlighted in the cool-system-hot-system framework. Central stimuli may be remembered preferentially as a function of stress, but it is not their “centrality” per se that is important, but rather the fact that they are preferred by the hot system as fear evoking or emotional. In most experimental studies, these two are confounded. A pistol may be a central stimulus in a crime scene. But, it is also fear provoking. Thus, in some cases the pattern of results predicted by the hot-cool framework is similar to that predicted by the Easterbrook hypothesis, though the reasons for the similar predictions are different. There are many illustrative studies in the experimental literature that conform to the type B predictions. For example, in a naturalistic study, Larson’s (1992) subjects recorded news events and everyday occurrences in a diary for nine months. Memory for the content of upsetting events was better than would be expected based on a comparison of memory performance on all events recalled at the delay interval approximating that of the upsetting event in question. Memory for the context in which such upsetting events were learned, however, was worse than memory for the analogous learning context for events that were unemotional. Nearly all weapon-focus studies show the type B pattern. However, whether such results, at least in laboratory studies, should necessarily be ascribed to the effects of moderate to high stress on memory is questionable. In a typical “weapon focus” laboratory study, Loftus, Loftus, and Messo (1987) presented participants with a series of slides portraying an event in a restaurant. Half of the participants saw a person point a pistol at the cashier; the other half saw the same individual hand the cashier a check. Memory for peripheral detail was less accurate in the weapon condition than in the neutral condition. Memory for the weapon itself was better than memory for the check. Although this is the pattern of data expected when the cool system is beginning to be negatively affected by stress, the experiment did not appear to be particularly stressful, and so a different cause may be implicated. The result was probably due to selective attention at time of study. As the authors reported, subjects’ eye fixations clustered on the pistol. Subjects may have failed to notice the surrounding context.
Emotional Memory
209
According to the cool-system-hot-system framework, attentional focusing itself may be directed by the selective amygdala responses. It has been demonstrated that through conditioning, formerly neutral stimuli (such as a weapon) may function to release primitive attentional, emotional, and behavioral repertoires (Gallagher & Holland, 1994; LeDoux, 1989). Thus, a stimulus that the individual has learned is dangerous is treated as any primitive releasing stimulus. The other aspects of the scene, not having previously undergone this conditioning, would not be singled out for special processing by the hot system. The enhanced amygdala response to the weapon, even under conditions of low stress, may direct the attentional system to give it special priority at the expense of the surrounding context. There is an important theoretical, and possibly practical, distinction, then, between perceiving the context but not remembering it because of hippocampal impairment due to high stress, and not seeing the context in the first place because one’s attention was not focused upon it. Maass and Krohnken (1989) report a study similar to that of Loftus et al. (1987) in which the subjects were approached by an experimenter who held either a syringe or a pen. Memory for the syringe was better than memory for the pen. Recall of most peripheral details was worse in the syringe condition than in the pen condition. Interestingly, though, individuals exposed to the syringe more accurately recalled details for the hand that held the experimental object than did those exposed to the pen. This result seems more consistent with a focused attention explanation than with the type B prediction, though it is possible that the hand might have been a conditioned amygdala-preferred stimulus. In a follow-up to Heuer and Reisberg (1990), Burke, Heuer, and Reisberg (1992) obtained measurable physiological stress-level differences with their emotional versus neutral slide presentations and, thus, their results are clearly relevant to the stress-related predictions of the present framework. They showed participants a series of slides accompanied by a tape-recorded narration that described the mother who took her young son to visit the father at work. Again, the father was either a surgeon or an auto mechanic. The test was immediate. The arousal-stress story produced a memory advantage for the emotional central events, and, as has been found in many other studies (Steblay, 1992), resulted in a small memory disadvantage for the background details. Cahill, Prins, Weber, and McGaugh (1994) used a modification of the Burke et al. (1992) paradigm to investigate the physiological underpinnings of the stress-induced memory advantage for the emotional material. They tested the hypothesis that arousal-associated enhanced memory for emotional material resulted from an activation of adrenergic stress hormone systems both during and after an emotional experience. Cahill et al. (1994)
210
Janet Metcalfe and W. Jake Jacobs
administered either propranolol or a placebo to subjects one hour before viewing the to-be-remembered slides. Propranolol, a medication commonly used for hypertension, blocks activation of P-adrenergic receptors, especially those in the amygdala (McGaugh & Introini-Collison, 1988). People in the placebo condition recalled more of the emotionally charged material than did the people who received propranolol. On the emotionally neutral material, the memory performance of the two groups was indistinguishable. These results indicate that blocking the effects of epinephrine and norepinephrine (P-adrenergic systems) results in a selective inhibiting effect on the hot system, and obviates the “emotional bias.” Many studies with animals show that increasing the levels of epinephrine and norepinephrine produces a selective enhancing effect on fear conditioning and other hot-system memory tasks. These data converge on the conclusion that the hot system is responsible for the often-seen memory enhancement of the emotional details of a scenario. If the level of stress is sufficiently high, the decreased functioning of the cool system may contribute to a loss of memory for the spatio-contextual surround that defines the event.
3. Memorial Effects of Traumatic Stress: Type C Predictions The hot-cool framework predicts that at extremely high levels of stress the hot system is hyper responsive and dominates the cool system. There are no experimental studies with humans at this level of stress. However, this prediction seems reasonable given the results of the animal studies that we described above showing that under the influence of extremely high levels of stress, glucocorticoids levels were elevated, hippocampal neurons disabled, hippocampal LTP diminished, and spatial memory impaired. If similar responses occur in humans, then certain patterns of fragmented memorial responses to stress are expected. There are many clinical reports on the effects of traumatic stress on human memory. For example, Classen et al. (1993) report an example provided by a young woman who accidentally fell 60 feet from a balcony. She described her experience as: “ . . . standing on another balcony watching a pink cloud float down to the ground.” Presumably, only fragments were remembered, and they were pieced together in a manner that showed a disruption of the spatio-temporal context. Similarly, upon recalling a panic attack, Hawkrigg’s (1975) patient reported: “I saw all the people looking at me-just faces, no bodies, all merged into one.” Again, the spatio-temporal context is misconstrued. According to van der Kolk and Fisler (1995), disturbing fragmentary and sometimes intrusive ideation is endemic to most (initial) reports of traumatic events. They report having listened to trauma narratives from hun-
Emotional Memory
211
dreds of children and adults over 20 years of experience. The reports are frequently organized without semantic representations, and are experienced as fragments of the sensory components of the event including visual images, olfactory, auditory or kinesthetic sensations, and intense waves of feeling. To investigate these pervasive reports, they administered a structured interview called the Traumatic Memory Inventory to PTSD patients, inquiring about the nature, duration, circumstances, sounds, images, smells, feelings, emotions, and narrative content of the patients’ memories, as well as about intrusive recollections, nightmares, and flashbacks experienced. They also asked the patients for documentation via witnesses and hospital records. This inventory was collected both for a traumatic event and for a nontraumatic event, such as high school graduations, births of their children, or weddings. The patients frequently considered inventory questions about the nontraumatic event to be nonsensical, and none reported olfactory, visual, auditory, or kinesthetic reliving experiences, or flashbacks or vivid dreams. In contrast, fragmentary images were common for the traumatic memories. Of the 46 subjects interviewed, none reported having a narrative for the traumatic event as their initial mode of awareness. Indeed, they reported initially being unable to tell a story about it. Rather, they remembered the event in the form of sensory-somatic-emotional fragmentary images. Fragmentary cues may also elicit uncontrollable feelings of fear, long after the traumatic event. The hot-system-based fear conditioning may provoke a heightened emotional reaction to the stimulus in question, regardless of whether the contextual details needed to locate the source of the fear in a past spatio-temporal context are present. This seemingly disembodied autonomic reaction may produce panic. The often-observed clinical phenomenon that traumatic memories are intrusive may be consistent with rigid reactions to fragments related to the original fear-provoking stimuli that are primarily governed by the hot system, and not informed by the contextual-narrative grounding of the cool system. The person who has experienced a trauma may encounter such fragments accidentally during his or her day-to-day living without knowledge of their significance, or of their causal effect in provoking the hyper-encoded fear responses. The lack of coherence of these memories, and the apparent lack of conformity to normal reality, although easily understandable in terms of the fragmentary coding enacted by the hot system acting in relative isolation, may nevertheless be dissonant with a person’s knowledge of his or her more mundane memories, which, having input from the cool system, would normally be less emotional, and more spatio-temporally grounded, narrative, and contextually integrated. The fragmentation of the memories of a traumatic event may exacerbate the person’s fear.
212
Janet Metcalfe and W.Jake Jacobs
The stereotypy of responses implicates the hot system as does the fragmentary nature of the eliciting features. Furthermore, the finding that the responses are resistant to extinction, even in light of patients’ conscious knowledge of the lack of current contingent effect, is consistent with experimental results showing the permanence of the conditioning of fear responses involving the amygdala (LeDoux, Romanski, & Xagoraris, 1989). If stress disabled the cool system at the time of encoding, and control devolved onto the hot system, a strong fear response, accompanied by little knowledge about the origin, or episodic source of the events, should result. Kaszniak, Nussbaum, Berren, and Santiago (1988) reported a case in which a person experienced strong emotional reactions in response to the cue of being approached from behind and to other fragmentary ambiguous stimuli. They suggested that these fears may have originated from what turned out to have been a traumatic experience involving a violent sexual assault. This and many other memory reports described in the modern literature on post-traumatic stress disorder as well as early reports by Janet (1925), Breuer and Freud (1895/1955), Freud (1919/1954), and James (1884), all fit this profile. There are many case studies of people, suffering from intrusive memory fragments, who, under sodium pentothal or hypnosis, seem to relive the fear-producing event (Spiegel & Scheflin, 1994). If the cool system was not functioning normally during the event, then true contextually grounded memories could not later emerge: what was not encoded cannot be retrieved. This is not to say that the fragments that do emerge are not real, or did not depend on some experience in the person’s past. The reliving experience may result primarily from the compellingly immediate emotions that are evoked under prompting by the cues given by the therapists and the sodium pentothal. Many researchers (Bartlett, 1932; Bransford & Johnson, 1973; Mandler, 1982) have demonstrated that people impose meaning according to their own schemata. People seek meaning, and will impose it when it is not apparent. They remember the meaning they imposed, even if that meaning was either not apparent in the events that occurred, or was different from the person’s initial interpretation. Faced with meaningless, intrusive, and disturbing fragmentary memories, apparently from a trauma, we suggest that people may impose a schema on these fragments too. As Janet suggested, “A situation has not been fully . . . assimilated until we have achieved an inward reaction through the words we address to ourselves, through the. . . recital of the event to others and to ourselves, and through the putting of this recital in its place as one of the chapters in our personal history” (in van der Kolk & van der Hart, 1991, p. 440). “Memories” of this sort may be inaccurate or entirely accurate. The individual may have
Emotional Memory
213
struck on a proper combination of fragments. The problem is there is no way to decide. It may be therapeutic to attempt such a reconstruction, regardless of the veracity of the scenarios so formed. Van der Hart, Brown, and van der Kolk (1989) outlined Janet’s three-stage approach to treating posttraumatic stress: neutralization, substitution, or reframing. Neutralization is similar to the modern technique of systematic desensitization. Starting with the least threatening memory and preceding through the most threatening, the client’s task is to reexperience and verbalize traumatic memories. Forming such a narrative, even after the fact, could allow cool-system control to develop over these reexperienced memories. Substitution involves changing the interpretation of a traumatic memory. This method relates to manipulations used in modern cognitive-behavioral therapies. Reframing involves creating an interpretation that is intellectually and morally acceptable. The client’s task is to understand and accept such an interpretation. Reframing is similar to procedures used in modern family therapy. None of these interventions rely on the objective accuracy of traumatic memory. Instead they depend upon the beliefs the client holds about the content of the memory, or the “memory” modified to form a comprehensible and acceptable story. The repetition of a story constructed from memory fragments will make it more coherent, believable, and increase the teller’s confidence in its veracity (Loftus, 1992).The interventions suggested by Janet, and refined and used under different labels by modern therapists, appear designed to take fixed, inflexible, intrusive, traumatic fragments and transform them to flexible, narrative “memories.” As such, these phenomena may lose their intrusive characteristics and hence such procedures may be therapeutic.
III. Adaptive Significance of the Two Systems It has been proposed that the cool system-content neutral, multimodal, cognitive-allows people to remember highly detailed and spatiotemporally organized narratized representations. The ability to reconsider, in this manner, situations from the past affords the person flexibility in judging past events, in reevaluating their own and other people’s past motives, in reviewing past events in a detached manner allowing complex reflection, in constructing plans concerning future behaviors based on past events, and, as a bonus, in resavoring remembered enjoyments and past aesthetic experiences in contextually grounded detail. This cool memory system, however, although undoubtedly central to highly intelligent behav-
214
Janet Metcalfe and W. Jake Jacobs
ior when a person is not under stress, may not always be optimal for responding to threat. In a dangerous situation, it may behoove the person or the animal to process the threatening cues-without regard for their integration into the whole-and escape the situation immediately based on such minimal cues. These kind of emotionally provocative details are selectively emphasized by the hot system. Under stressful conditions, these trigger stimuli, selectively enhanced by the hot system, are thus more dominant in the jointly formed memory of the events that transpired than are the cool, spatio-temporal aspects of the event. Under nonstressful conditions-the conditions under which the cool system functions optimally-the complex and interwoven spatio-temporal representations provide an adaptive advantage. They contribute in a significant way to human intelligence. Under stressful conditions-the conditions under which the cool system becomes dysfunctional-there is an adaptive advantage to relinquishing control to a more primitive and automatic hot system that allows rapid protective responses, blocking the cool system from introjecting its complex musings. There also may be an advantage to remembering, in sharp detail, the fragments to which the hot system responded under conditions of high stress. Should such triggers occur again, the organism will be in a position to respond automatically in efforts to quickly protect itself. Unfortunately, this automatic response to fragmentary cues also has potentially serious negative implications for humans who have been exposed to traumatic stress, in which, for example, decontextualized stimulus fragments may trigger fearful and aggressive responses in ways that are inappropriate and maladaptive. IV. Conclusion In this chapter, we proposed a framework that integrates research from two distinct traditions in psychology concerning two memory systems-a hot emotional system and a cool cognitive system. We suggested that the contents of the two systems are superimposed at the level of consciousness, yielding a cognitive-emotional representation. The cool-system components form a cognitively informed spatio-contextual web upon which the hot system highlights emotionally significant fragments. The weighting of emotional features or of the contextual-relational content depends on the extent of hot or cool system activation, respectively. This activation, in turn, depends upon the stress levels experienced by the person: as stress levels increase, the “hot” emotional-amygdala system becomes increasingly responsive whereas the “cool” episodic-hippocampal system initially be-
Emotional Memory
215
comes more responsive, but at high stress levels becomes less responsive resulting in an inverted U-shaped “Yerkes-Dodson” function. At traumatic levels of stress, the cool system is notably disregulated, leaving the hot system to dominate both immediate responding and memory encoding. We illustrated how this hot-cool framework can explain memory phenomena falling under the rubric of weapon focus and the fragmentation of memory under conditions of traumatic stress. ACKNOWLEDGMENTS We thank David Baldwin, Constance Dalenberg, David Cleaves, Steven Grant, Walter Mischel, and Wayne Wickelgren. This research was supported by National Institute of Mental Health grant MH48066 to Janet Metcalfe and by funding from the McDonnell-Pew Foundation and the Zumberge Foundation to W. J. Jacobs.
REFERENCES Abraham, S., Pickering, A., Polkey, C. E., & Morris, R. C. (1996). Spatial memory deficits in patients with unilateral damage to the right hippocampal formation. Neuropsychologia, 35, 11-24. Adolphs, R., Tranel, D., Damasio, H., & Damasio, A. R. (1995). Fear and the human amygdala. Journal of Neuroscience, 15, 5879-5891. Aggleton, J. P. (1992). The functional effects of amygdala lesions in humans: A comparison of findingsfrom monkeys. In J. P. Aggleton (Ed.), The amygdala: Neurobiological aspects of emotion, memory, and mental dysfunction (pp. 485-503). New York: Wiley-Liss. Aggleton, J. P., & Passingham, R. E. (1981). Syndrome produced by lesions of the amygdala in monkeys (Macaca Mulatta). Journal of Comparative and Physiological Psychology, 95, 961-977. Barnes, C. A. (1988). Spatial learning and memory processes:The search for their neurobiological mechanisms in the rat. Trends in Neurosciences, 11, 163-169. Bartlett, F. C. (1932). Remembering. London: Cambridge University Press. Bechara, A., Tranel, D., Damasio, H., Adolphs, R., Rockland, C., & Damasio, A. R. (1995). Double dissociation of conditioning and declarative knowledge relative to the amygdala and hippocampus in humans. Science, 269, 1115-1118. Bohannon, J. N. (1988). Flashbulb memories for the space shuttle disaster: A tale of two theories. Cognition, 29, 179-196. Bransford,J. D., & Johnson, M. K. (1973). Considerationsof some problems of comprehension. In W. G. Chase (Ed.), Visual information processing (pp. 383-438). New York: Academic Press. Breuer, J., & Freud, S. (1896/1955). The standard edition of the complete psychological works of Sigmund Freud: Volume 2. Studies on hysteria (J. Strackey, Ed. & Trans.). London: Horarth Press. Burke, A., Heuer, F., & Reisberg, D. (1992). Remembering emotional events. Memory & Cognition, 20, 277-290. Buschke, H. (1%5). Impairment of short-term memory. Neurology, 15, 913-918. Buschke, H. (1984). Cued recall in amnesia. Journal of Clinical Neuropsychology, 6,433-440.
216
Janet Metcalfe and W. Jake Jacobs
Cahill, L., & McGaugh, J. L. (1996). Modulation of memory storage. Current Opinion in Neurobiology, 6, 237-242. Cahill, L., Prins, B., Weber, M., & McGaugh, J. L. (1994). 8-Adrenergic activation and memory for emotional events. Nature, 371, 702-704. Cermak, L. S., Butters, N., & Morienes, J. (1974). Some analyses of the verbal encoding deficits of alcoholic Korsakoff patients. Brain and Language, I , 141-150. Chalfonte, B., & Johnson, M. K. (1994). Memory binding. In D. L. Schacter & E. Tulving (Eds.), Memory Systems 1994 (pp. 87-117). Cambridge: MIT Press. Chapman, W. P., Schroeder, H. R., Geyer, G., Brazier, M. A. B., Fager, C., Popper, J. L., Solomon, H. C., & Yakovlev, P. 1. (1954). Physiological evidence concerning importance of amygdaloid nuclear region in the integration of circulatory function and emotion in man. Science, 120, 949-950. Christianson, S. A. (1984). The relationship between induced emotional arousal and amnesia. Scandinavian Journal of Psychology, 25, 147-160. Christianson,S. A. (1992). Emotional stress and eyewitness memory: A critical review. Psychological Bulletin, 112, 284-309. Classen, C., Koopman, C., & Spiegel, D. (1993). Trauma and dissociation. Bulletin of the Menninger Clinic, 57, 178-194. Cohen, N. F., & Squire, L. (1980). Preserved learning and retention of pattern analyzing skill in amnesia: Dissociation of knowing how and knowing that. Science, 210, 207-210. Cohen, S., & Wills, T. A. (1985). Stress,social support, and the buffering process. Psychological Bulletin, 98, 310-357. Cutler, B. L., Penrod, S. D., & Martens, T. K. (1987). The reliability of eyewitness identification: The role of system and estimator variables. Law and Human Behavior, 11, 233-258. Damasio, A. (1994). Descartes’ error: Emotion, reason, and the human brain. New York: Grosset/Putnam. Davis, M. (1992). The role of the amygdala in fear and anxiety. Annual Review of Neuroscience, 15, 353-375. Davis, M., Rainnie, D., & Cassell, M. (1994). Neurotransmission in the rat amygdala related to fear and anxiety. Trends in Neurosciences, 17, 208-214. de Kloet, E. R., Oitzl, M. S., & Joals, M. (1993). Functional implications of brain corticosteroid receptor diversity. Cellular and Molecular Neurobiology, 13, 433-455. Diamond, D. M., Bennett, M. C., Fleshner, M., & Rose, G. M. (1992). Inverted-U relationship between the level of peripheral corticosterone and the magnitude of hippocampal primed burst potentiation. Hippocampus, 2, 421-430. Dicks, D., Myers, R. E., & Wing, A. (1969). Uncus and amygdala lesions: Effects on social behavior in the free-ranging monkey. Science, 165, 69-71. Downer, J. L. de C. (1961). Changes in visual gnostic functions and emotional behavior following unilateral temporal pole damage in the “split-brain” monkey. Nature, 191, 50-51. Easterbrook, J. A. (1959). The effect of emotion on cue utilization and the organization of behavior. Psychological Review, 66, 183-201. Egeth, H. (1994). Emotion and the eyewitness. In P. M. Niedenthal & S. Kitayama (Eds.), The heart’s eye: Emotional influences in perception and attention (pp. 245-267). New York: Academic Press. Fanselow, M. S. (1994). Neural organization of the defensive behavior system responsible for fear. Psychonomic Bulletin and Review, 1, 429-438. Feindel, W., & Rasmussen, T. (1991). Temporal lobectomy with amygdalectomy and minimal hippocampal resection: Review of 100 cases. Canadian Journal of Neurological Science, 18, 603-605.
Emotional Memory
217
Foy, M. R., Foy, J. G., Levine, S., & Thompson, R. F. (1990). Manipulation of pituitaryadrenal activity affects neural plasticity in rodent hippocampus. Psychological Science, I, 201-294. Foy, M. R., Stanton, M. E., Levine, S., & Thompson, R. F. (1987). Behavioral stress impairs long-term potentiation in rodent hippocampus. Behavioral and Neural Biology, 48, 138-1 49. Freud, S. (1919/1954). Introduction to psychoanalysis and the war neuroses. In J. Strackey (Ed. & Trans.), The standard edition of the complete psychological works of Sigmund Freud: Volume 2. Studies on hysteria. London: Horarth Press. Gaffan, D. (1992). Amygdala and the memory of reward. In J. P. Aggleton (Ed.), The amygdala: Neurobiological aspects of emotion, memory, and mental dysfunction (pp. 471-483). New York: Wiley-Liss. Gallagher, M., & Holland, P. C. (1994). The amygdala complex: Multiple roles in associative learning and attention. Proceedings of the National Academy of Science, 91, 11771-11776. Gardner, H., Boller, F., Morienes, J., & Butters, N. (1973). Retrieving information from Korsakoff patients: Effects of categorical cues and reference to the task. Cortex, 9, 165-175. Gentile, C. G., Jarrell, T. W., Teich, A. P. M., & Schneiderman, N. (1986). The role of the amygdaloid central nucleus in the retention of differential Pavlovian conditioning of bradycardia in rabbits. Behavioral Brain Research, 20, 263-273. Gloor, P., Olivier, A., & Quesney, L. F. (1981). The role of the amygdala in the expression of psychic phenomena in temporal lobe seizures. In Y. Ben-Ari (Ed.), The amygdaloid complex. (pp. 489-498). New York: ElseviedNorth Holland Biomedical Press. Graf, P., & Schacter, D. L. (1985). Implicit and explicit memory for new associationsin normal subjects and amnesic patients. Journal of Experimental Psychology: Learning, Memory, and Cognition, I I , 501-518. Graf, P., Shimamura, A. P., & Squire, L. R. (1985). Priming across modalities and priming across category levels: extending the domain of preserved function in amnesia. Journal of Experimental Psychology: Learning, Memory, and Cognition, 1I , 385-395. Graf, P., Squire, L. R., & Mandler, G. (1984). The information that amnesic patients do not forget. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 164-178. Grant, S. G. N., O’Dell, T. J., Karl, K. A., Stein, P. L., Soriano, P., & Kandel, E. R. (1992). Impaired long-term potentiation, spatial learning, and hippocampal development in fyn mutant mice. Science, 258, 1903-1909. Halgren, E. (1981). The amygdala contribution to memory and emotion: Current studies in humans. In Y. Ben-Ari (Ed.) The amygdaloid complex (pp. 395-408). Amsterdam: Elsevier. Halgren, E. (1992). Emotional neurophysiology of the amygdala within the context of human cognition. In J. P. Aggleton (Ed.), The amygdala: Neurobiological aspects of emotion memory, and mental dysfunction (pp. 191-228). New York: Wiley-Liss. Hamann, S. B., Stefanacci,L., Squire, L. R., Adolphs, R., Tranel, D., Damasio, H., & Damasio, A. (1996). Recognizing facial emotion. Nature, 379, 497. Hawkrigg, J. J. (1975). Agoraphobia. Nursing Times, 71, 1280-1282. Hayman, C. A. G., & Tulving, E. (1989). Is priming in fragment completion based on a ‘traceless’ memory system? Journal of Experimental Psychology: Learning, Memory and Cognition, 15, 941-956. Helmstetter, F. J. (1992). The amygdala is essential for the expression of conditional hypoanalgesia. Behavioral Neuroscience, 106, 518-528.
218
Janet Metcalfe and W. Jake Jacobs
Helmstetter, F. J., & Bellgowan, P. S . (1994). Effects of muscimol applied to the basolateral amygdala on acquisition and expression of contextual fear conditioningin rats. Behavioral Neuroscience, 108, 1005-1009. Heuer, F., & Reisberg, D. (1990). Vivid memories of emotional events: The accuracy of remembered minutiae. Memory & Cognition, 18, 496-506. Hitchcock, J. M., & Davis, M. (1986). Lesions of the amygdala, but not of the cerebellum or red nucleus, block conditioned fear as measured with the potentiated startle paradigm. Behavioral Neuroscience, 100, 11-22. Holmes, T. H., & Rahe, R. H. (1967). The social readjustment rating scale.Journal of Psychometric Research, 14, 121-132. Hosch, H. M., & Bothwell, R. K. (1990). Arousal, description, and identification accuracy of victims and bystanders. Journal of Social Behavior and Personality, 5, 481-488. Introini-Collison, I. B., Nagahara, A. H., & McGaugh, J. L. (1989). Memory enhancement with intra-amygdala post-training naloxone is blocked by concurrent administration of propranolol. Brain Research, 476, 94-101. Jacobs, B. L., & McGinty, D. J. (1972). Participation of the amygdala in complex stimulus recognition and behavioral inhibition: Evidence from unit studies. Brain Research, 36, 431-436. Jacobs, W. J., & LoLordo, V.M. (1977). The sensory basis of avoidance responding in the rat: Relative dominance of auditory or visual warning signals and safety signals. Learning and Motivation, 8, 448-466. Jacobs, W. J., & LoLordo, V. M. (1980). Selective associationin aversive conditioning: Implications for avoidance learning in the rat. Learning and Motivation, 11, 427-455. Jacobs, W. J., & Nadel, L. (1985). Stress induced recovery of fears and phobias. Psychological Review, 92, 512-531. Jacobson, L., & Sapolsky, R. (1991). The role of the hippocampus in feedback regulation of the hypothalamic-pituitaryadrenocortical axis. Endocrine Reviews, 12, 118-134. Jacoby, L. L., & Witherspoon, D. (1982). Remembering without awareness. Canadian Journal Of Psychology, 38, 631-668. James, W. (1884). What is emotion? Mind, 9, 188-205. Jaget, P. (1925). Psychological healing, Vols. 1 & 2. New York: MacMillan. Joals, M., & de Kloet, E. R. (1989). Effects of glucocorticoid and norepinephrine on the excitability in the hippocampus. Science, 245, 1502-1505. Joals, M., & de Kloet, E. R. (1990). Mineralocorticoidreceptor-mediated changes in membrane properties of rat CAI pyramidal neurons in vitro. Proceedings of the National Academy of Science, 87, 4495-4498. Joels, M., & de Kloet, E. R. (1992). Control of neuronal excitabilityby corticosteroid hormones. Trends in Neurosciences, 15, 25-29. Kapp, B. S., Whalen, P. J., Supple, W. F., & Pascoe, J. P. (1992). Amygdaloid contributions to conditioned arousal and sensory information processing. In J. P. Aggleton (Ed.), The amygdala: Neurobiological aspects of emotion, memory, and mental dysfunction (pp. 229-254). New York Wiley-Liss. Kassin, S . M., Ellsworth, P. C., & Smith, V. L. (1989). The “general acceptance” of psychological research on eyewitness testimony. American Psychologist, 44, 1089-1098. Kaszniak, A. W., Nussbaum, P. D., Berren, M.R., & Santiago, J. (1988). Amnesia as a consequence of male rape: A case report. Journal of Abnormal Psychology, 97, 100-104. Kerr, D. S.,Huggett, A. M., & Abraham, W. C. (1994). Modulation of hippocampal longterm potentiation and long-term depression by corticosteroid receptor activation. Psychobiology, 22, 123-133. ~
Emotional Memory
219
Kirschbaum, C., Diedrich, O., Gehrke, J., WAst, S., & Hellhammer, D. (1991). Cortisol and behavior: The “Tier Mental Challenge Test” (TMCT)-First evaluation of a new psychological stress test. In A. Ehlers, W. Fiegenbaum, I. Florin, & J. Margraf (Eds.), Perspectives and promises of clinical psychology (pp. 67-78). New York: Plenum. Kling, A. (1972). Effects of amygdalectomy on social-affective behavior in non-human primates. In B. E. Eleftherion (Ed.), Advances in behavioral biology: The neurobiology of the amygdala, 1st ed, Vol. 2 (pp. 511-536). New York London. Kluver, H., & Bucy, P. C. (1939). Preliminary analysis of functions of the temporal lobes in monkeys. Archives of Neurology and Psychiatry, 42, 979-1000. Kroll, N. E. A., Knight, R., Metcalfe, J., Wolf, E. S., & Tulving, E. (1996). Cohesion failure as a source of memory illusion. Journal of Memory and Language, 35, 176-196. Laing, K. C., Juler, R. G., & McGaugh, J. L. (1986). Modulating effects of posttraining epinephrine on memory: Involvement of the amygdala noradrenergic system. Brain Research, 368, 125-133. Larson, S. F. (1992). Potential flashbulbs: Memories of ordinary news as a baseline. In E. Winograd & U. Neisser (Eds.), Affect and accuracy in recall: The problem of “flashbulb” memories (pp. 32-64). Cambridge, England: Cambridge University Press. Lazarus, A. A., & Folkman, S. (1984). Stress, appraisal, and coping. New York Springer. LeDoux, J. E. (1989). Cognitive-emotionalinteractions in the brain. Cognition and Emotion, 3, 267-289. LeDoux, J. E. (1993). Emotional memory systems in the brain. Behavioural Brain Research, 58, 69-19. LeDoux, J. E. (1995). Emotion: Clues from the brain. Annual Review of Psychology, 46, 209-235. LeDoux, J. E., Romanski, L., & Xagoraris, A. (1989). Indelibility of subcortical emotional memories. Journal of Cognitive Neuroscience, 1, 238-243. Lee, G. P., Arena, J. G., Meador, K. J., & Smith,J. R. (1988). Changes in autonomic responsiveness following bilateral amygdalotomy in humans. Neuropsychiatry, Neuropsychology, and Behavioral Neurology, I, 119-129. Lee, G. P., Meador, K. J., Smith, J. R., Loring, D. W., & Flanigan, H. F. (1988). Preserved cross modal association following bilateral amygdalotomy in man. International Journal of Neuroscience, 40, 47-55. Loftus, E. F. (1992). When a lie becomes memory’s truth: Memory distortion after exposure to misinformation. Current Directions in Psychological Science, I , 121-123. Loftus, E. F., & Burns, T. (1982). Mental shock can produce retrograde amnesia. Memory & Cognition, 10, 318-323. Loftus, E. F., Loftus, G. R., & Messo, I. (1987). Some facts about “weapon focus.” Law and Human Behavior, 11, 1-13. Maass, A., & Krohnken, G. (1989). Eyewitness identification: Simulating the “weapon effect.” Law and Human Behavior, 13, 397-408. Mandler, G. (1982). Mind and emotion. Malabar, FL: R. E. Krieger. McEwen, B. S. (1994). Stressful experience, brain and emotions: Developmental, genetic and hormonal influences. In M. S. Gazzaniga (Ed.), The Cognitive Neurosciences (pp. 1117-1137). Cambridge, MA: MIT Press. McGaugh, J. L., & Introini-Coltison,I. B. (1987-88). Hormonal and neurotransmitter interactions in the modulation of memory storage: Involvement of the amygdala. International Journal of Neurology, 21-22, 58-72. McGaugh, J. L., Introini-Collison, I. B., Cahill, L. F., Castellano, C., Dalmaz, C., Parent, M. B., & Williams, C. L. (1993). Neuromodulatory systems and memory storage: Role of the amygdala. Behavioural Brain Research, 58, 81-90.
220
Janet Metcalfe and W. Jake Jacobs
Metcalfe Eich, J. (1982). A composite holographic associative recall model. Psychological Review, 89, 627-661. Metcalfe Eich, J. (1985). Levels of processing, encoding specificity,elaboration, and CHARM. Psychological Review, 91, 1-38. Metcalfe, J. (1990). A composite holographic associative recall model (CHARM) and blended memories in eyewitness testimony. Journal of Experimental Psychology: General, I1 9, 145- 160. Metcalfe, J. (1991). Recognition failure and the composite memory trace in CHARM. Psychological Review, 98, 529-553. Metcalfe, J. (1993a). Monitoring and gain control in an episodic memory model: Relation to P300 event-related potentials. In A. F. Collins, S. E. Gathercole, M. A. Conway, & P. E. Moms (Eds.), Theories of Memory (pp. 327-354). Hillsdale, NJ: Lawrence Erlbaum Associates. Metcalfe, J. (1993b). Novelty monitoring, metacognition, and control in a composite holographic associative recall model: Implications for Korsakoff amnesia. Psychological Review, 100, 3-22. Metcalfe, J. (1994). Novelty monitoring, metacognition,and frontal lobe dysfunction: Implications of a computational model of memory. In J. Metcalfe & A. P. Shimamura (Eds.), Metacognition: Knowing about Knowing. Cambridge, MA: MIT Press. Metcalfe, J. (1997). Predicting syndromes of amnesia from a composite holographic associative recallhecognition model (CHARM), Memory, 5, 233-253. Metcalfe, J., & Bjork, R. A. (1991). Composite models never (well, hardly ever) compromise: Comment on Schooler & Tanaka (1991). Journal of Experimental Psychology: General, 120, 203-210. Metcalfe, J., Cottrell, G. W., & Mencl, W. E. (1992). Cognitive binding-A computationalmodeling analysis of a distinction between implicit and explicit memory. Journal of Cognitive Neuroscience, 4, 289-298. Metcalfe, J., Mencl, W. E., & Cottrell, G. W. (1994). Cognitive binding. In D. L. Schacter & E. Tulving (Eds.), Memory Systems 1994 (pp. 369-394). Cambridge: MIT Press. Metcalfe, J., & Murdock, B. B., Jr. (1981). An encoding and retrieval model of single-trial free recall. Journal of Verbal Learning and Verbal Behavior, 20, 161-189. Metcalfe, J., Schwartz,B. L., & Joaquim, S. G. (1993). The cue-familiarity heuristic in metacognition. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 851-861. Mischel, W., & Shoda, Y. (1995). A cognitive-affectivesystem theory of personality: Reconceptualizing situations. dispositions, dynamics, and invariance in personality structure. Psychological Review, 102, 246-268. Mishkin, M., & Murray, E. A. (1994). Stimulus recognition. Current Opinion in Neurobiology, 4, 200-206. Mishkin, M., & Petri, H. L. (1984). Memories and habits: Some implications for the analysis of learning and retention. In L. Squire & N. Butters (Eds.), Neuropsychology of memory. New York Guilford Press. Morris, R. G. M., Garrud, J., Rawlins, N. P., & O’Keefe, J. (1982). Place navigation impaired in rats with hippocampal lesions. Nature (London), 297, 681-683. Moscovitch, M. (1982). Multiple dissociations of function in amnesia. In L. S. Cermak (Ed.), Human memory and amnesia (pp. 337-370). Hillsdale, N J Lawrence Erlbaum. Nadel, L. (1994). Multiple memory systems: What and why: An update. In D. L. Schacter & E. Tulving (Eds.), Memory systems 1994. Cambridge, MA: MIT Press. O’Keefe, J., & Bouma, H. (1969). Complex sensory properties of certain amygdala units in the freely moving cat. Experimental Neurology, 23, 384-398.
Emotional Memory
221
O’Keefe, J., & Dostrovsky, J. (1971). The hippocampus as a spatial map: Preliminary evidence from unit activity in the freely-moving rat. Brain Research, 34, 171-175. O’Keefe, J., & Nadel, L. (1978). The hippocampus as a cognitive map. London: Oxford University Press. Pacak, K., Palkovits, M., Kopin, I. J., & Goldstein, D. S. (1995). Stress-induced norepinephrine release in the hypothalamic paraventricular nucleus and pituitary-adrenocortical and sympathoadrenal activity-in-vivo microdialysis studies. Frontiers in NeuroendocrinolOgy, 16, 89-150. Packard, M. G., Cahill, L., & McGaugh, J. L. (1994). Amygdala modulation of hippocampaldependent and caudate nucleus dependent memory processes. Proceedings of the National Academy of Sciences, 91, 8477-8481. Pavlides, C., Watanabe, Y., & McEwen, B. S. (1993). Effects of glucocorticoids on hippocampal long-term potentiation source. Hippocampus, 3, 183-192. Peters, D. P. (1988). Eyewitness memory and arousal in a natural setting. In M. M. Gruneberg, P. E. Morris, & R. N. Sykes (Eds.), Proceedings of the second international conference on practical aspects of memory, Vol. 1, (pp. 89-94) New York: Wiley. Posner, M. I., & Raichle, M. E. (1994). Images of mind. New York: Freeman. Reder, L. M., & Ritter, F. E. (1992). What determines initial feeling of knowing? Familiarity with question terms, not with the answer. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 435-452. Reul, J. M. H. M., & de Kloet, E. R. (1985). Two receptor systems for corticosterone in rat brain: Microdistribution and differential occupation. Endocrinology, 117, 2505-2512. Rey, M., Carlier, E., * Soumireu-Mourat, B. (1987). Effects of corticosterone on hippocampal slice electrophysiology in normal and adrenalectomized BALB/c mice. Neuroendocrinology, 49, 120-125. Rolls, E. T. (1986). A theory of emotion, and its application to understanding the neural basis of emotions, In Y . Oomur (Ed.), Emotions: Neural and chemical control (pp. 325-433). Tokyo: Japan Scientific Societies Press. Roozendaal, B., Portillo-Marquez, G., & McGaugh, J. L. (in press). Basolateral amygdala lesions block glucocorticoid-induced modulation of memory for spatial learning. Behavioral Neuroscience. Rosvold, H. E., Mirsky, A. F., & Pribram, K. H. (1954). Influence of amygdalectomy on social interaction in a monkey group. Journal of Comparative and Physiological Psychology, 47, 1173-1178. Rudy, J. W. (1993). Contextual conditioning and auditory cue conditioning dissociate during development. Journal Behavioral Neuroscience, 107, 887-891. Rudy, J. W., & Sutherland, R. J. (1994). The memory coherence problem, configural associations, and the hippocampal system. In D. L. Schacter & E. Tulving (Eds.), Memory systems 1994 (pp. 119-146). Cambridge: MIT Press. Rudy, J. W., & Sutherland, R. J. (1995). Configural association theory and the hippocampal formation: An appraisal and reconfiguration. Hippocampus, 5, 375-389. Sakimura, K., Kutsudwada, T., Ito, I., Manage, T., Takamyama, C., Kushiya, E., Yagi, T., Aizawa, S., Inoue, Y., Sugiyama, H., & Mishina, M. (1995). Reduced hippocampal LTP and spatial learning in mice lacking NMDA 1 subunit. Nature, 373, 151-155. Schachter, S., & Singer, J. E. (1962). Cognitive, social, and physiological determinants of emotional state. Psychological Review, 69, 379-399. Schacter, D. L. (1987). Implicit memory: History and current status. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 501-518. Schwartz, B. L., & Metcalfe, J. (1992). Cue familiarity but not target retrievability enhances feeling-of- knowing judgments. Journal of Experimental Psychology: Leaning, Memory, and Cognition, 18, 1074-1083.
222
Janet Metcalfe and W. Jake Jacobs
Shimamura, A. P. (1986). Priming effects in amnesia: Evidence for a dissociable memory function. Quarterly Journal of Experimental Psychology, 38A, 619-644. Shors, T. J., Seib, T. B., Levine, S., & Thompson, R. F. (1989).Inescapable versus escapable shock modulates long-term potentiation in the rat hippocampus. Science, 244, 224-226. Shors, T. J., Weiss, C., & Thompson, R. F. (1992). Stress-induced facilitation of classical conditioning. Science, 257, 537-539. Spiegel, D., & Scheflin, A. W. (1994).Dissociated or fabricated-Psychiatric aspects of repressed memory in criminal and civil cases. International Journal of Clinical and Experimental Hypnosis, 42, 411-432. Squire, L. R. (1982). Comparisons between forms of amnesia: Some deficits are unique to Korsakoff's syndrome. Journal of Experimental Psychology: Learning, Memory, and Cognition, 8, 560-571. Squire, L.R.(1994). Declarative and non-declarative memory: Multiple brain systems supporting learning and memory. In D. L. Schacter & E. Tulving (Eds.), Memory systems 1994 (pp. 203-232). Canbridge: MIT Press. Stansbury, K., & Gunnar, M. R. (1994). Adrenocortical activity and emotion regulation. Monographs of the Society for Research in Child Development, 59, 250-283, Steblay, N. M. (1992). A meta-analytic review of the weapons focus effect. Law and Human Behavior, 16, 413-424. Sutherland, R. J., & Rudy, J. W. (1989). Configural association theory: The contribution of the hippocampus to learning, memory, and amnesia. Psychobiology, 17, 129-144. Sutton, S.,Braren, M., Zubin, J., &John, E. R. (1965).Evoked potential correlates of stimulus uncertainty. Science, 150, 1187-1188. Tulving, E.(1985).Memory and consciousness. Canadian Psychologist, 26, 1-12. van der Hart, O., Brown, P., & van der Kolk, B. A. (1989). Pierre Janet's treatment of posttraumatic stress disorder. Journal of Traumatic Stress, 2, 379-395. van der Kolk, B., & Fisler, R. (1995). Dissociation and the fragmentary nature of traumatic memories: Overview and exploratory study. Journal of Traumatic Stress, 8, 505-525. van der Kolk, B. A., & van der Hart, 0. (1991).The intrusive past: the flexibility of memory and the engraving of trauma. American Imago, 48, 425-454. Weiskrantz,L. (1956).Behavioral changes associated with ablation of the amygdaloid complex in monkeys. Journal of Comparative and Physiological Psychology, 49, 381-391. Winocur, G.,Kmsbourne, M., & Moscovitch, M. (1981).The effect of cueing on release from proactive interference in Korsakoff amnesic patients. Journal of Experimental Psychology: Human Learning and Memory, 7, 56-65. Yerkes, R. M., & Dodson, J. D. (1908). The relation of strength of stimulus to rapidity of habit-formation, Journal of Comparative Neurology of Psychology, 18, 459-482. Yuille, J. C., & Cutshall, J. L. (1989). Analysis of the statements of victims, witnesses and suspects. In J. C. Yuille (Ed.), Credibility assessment (pp. 175-191), Dordrecht the Netherlands: Kluwer.
METACOMPREHENSION OF TEXT Influence of Absolute Confidence Level on Bias and Accuracy Ruth H. Maki
I. Introduction
A growing body of literature shows that students can predict their future performance over text material with some accuracy, and they can accurately assess the level of their past performance (Maki & Serra, 1992a; Weaver, 1990). However, there is some controversy about the meaning of the term accuracy in metacognitive tasks. The measure used in studies of metacomprehension (which have usually involved predicting performance on texts) has often been a correlation between predictions and performance (Glenberg, Sanoki, Epstein, & Morris, 1987; Maki, Foley, Kajer, Thompson, & Willert, 1990; Maki, Jonas, & Kallod, 1994; Maki & Serra, 1992a, 1992b; Weaver, 1990). Correlation measures are common in other studies of metacognition conducted within the framework of cognitive psychology, including studies of feeling-of-knowing (FOK) in which recognition of nonrecalled items is predicted (e.g., Nelson, Gerler, & Narens, 1984), and judgments-of-learning (JOLs) in which predictions of recall are made following learning (Nelson & Dunlosky, 1991). Studies developed in a decision-making framework have generally had subjects make probability estimates of an answer’s correctness after they have answered test questions. The measure of accuracy has typically been calibration, based on the match THE PSYCHOLOGY OF LEARNING AND MOTIVATION, VOL. 38
223
Copyright 0 1998 by Academic Press. All rights of reproduction in any form reserved. 0079-7421~98$25.00
224
Ruth H. Maki
between the values of these posttest estimates and performance (e.g., Koriat, Lichtenstein, & Fischhoff, 1980; Liberman & Tversky, 1993; Lichtenstein, Fischhoff, & Phillips, 1982). Other studies in the educational psychology literature have used other measures. Pressley, Snyder, Levin, Murray, and Ghatala (1987) used the absolute difference between global predicted scores and actual overall scores. Glover (1989b) subtracted subjects’ actual overall performance from their overall estimates. A. A COMPARISON OF MEASURES OF METACOGNITIVE ACCURACY Accuracy has been defined differently in studies developed in different research areas. The definition of accuracy in metacognition should be guided by the theoretical questions that are of interest to the investigator and not necessarily the specific research area in which the investigator works. The main purpose of this chapter is to compare measures of metacognitive accuracy under conditions designed to manipulate the absolute level of confidence expressed in predictions and posttest estimates and under conditions designed to manipulate the level of test performance. Theoretically, metacognitive accuracy should be independent of these manipulations. This assumption was tested empirically with measures of metacomprehension based on predictions and posttest estimates of recognition test performance. 1. Measures Used in Studies of Metacognition
In metacognitive studies conducted by cognitive psychologists, values from a rating scale are correlated with performance. Some early studies (e.g., Glenberg et al., 1987) used Pearson r correlations, but later studies have used the nonparametric gamma correlation (Goodman & Kruskal, 1954), as recommended by Nelson (1984). Nelson listed six properties that a satisfactory measure of feeling-of-knowing (FOK; predicting recognition following recall failure) should possess: (1) an increase in FOK ability should be accompanied by an increase in the FOK accuracy score; (2) if two individuals have equal recognition, the one with the better FOK ability should have a higher FOK accuracy score; (3) FOK accuracy should be independent of overall performance; (4) FOK accuracy should be unrelated to recognition ability; (5) FOK accuracy should be unrelated to test difficulty; (6) FOK accuracy should be unrelated to the type of memory test (e.g., recall, relearning). Nelson (1984) selected the gamma correlation as the best measure both because it satisfies these criteria and because it makes appropriate statistical assumptions. Gamma requires only ordinal data, it can obtain its maximum value no matter how many ties there are in the data, and it is not sensitive to the number of items in each rating or performance category.
Absolute Confidence and Metacomprehension
225
Although Nelson selected gamma as the best measure for FOK judgments, it has been used as the primary measure in many studies of other types of metacognition. Recently, however, Liberman and Tversky (1993) have criticized gamma because it handles ties by dropping them. When subjects are predicting only one of two outcomes (correct or incorrect), ties are highly likely and much of the data would be dropped. This is much less likely when both the ratings and performance outcomes have multiple levels. Weaver and Keleman (1997) recently demonstrated that gamma is increased if ratings are shifted from the middle to the ends of the scale. As with any correlation, increasing the variability in a variable increased gamma. However, Weaver and Keleman also showed that this doesn’t make gamma a poor indicator of relative accuracy because their modeling of the delayed judgment of learning effect (higher gammas when predictions are delayed as compared to immediate) indicated that these scale changes could not account for the entire improvement in measured accuracy. Various measures may actually tap different aspects of metacomprehension. Schraw (1995) has recently argued that agreement accuracy, the correspondence between a specificjudgment and its actual value based on performance, should be distinguished from association, the covariance in judgments and actual performance. Schraw argued that gamma is a flawed measure both because of its statistical properties (empty cells distort its value, a zero value doesn’t necessarily imply no relationship, and an oversensitivity to the distribution of scores) and because it does not tap absolute accuracy. That is, subjects may be consistently overconfident or underconfident and still produce a perfect gamma if predictions increase monotonically with performance. Schraw recommended use of the Hamann coefficient (described by Romesburg, 1984), but it is limited in its usefulness because it cannot be generalized to arrays having different numbers of rows and columns. Both Nelson (1996) and Wright (1996) criticized Schraw’s conclusions. Nelson argued that Schraw had ignored the important distinction between calibration (absolute accuracy, defined as whether predicted performance matches actual performance) and resolution (relative accuracy, defined as whether the predicted performance varies in a correlated manner with actual peformance). Nelson argued that relative accuracy is the most important measure of metacognitive accuracy. Wright (1996) argued that Hamann’s coefficient is less appropriate than gamma because it depends upon the row and marginal totals and gamma does not. Thus, it depends both on the level of performance and the level of confidence expressed by the judgments, as well as on the association between these. Both Wright (1996) and Nelson (1996) recommend the continued use of gamma as a measure of metacognitive accuracy.
226
Ruth H.Maki
In spite of these arguments against Schraw’s conclusions, the point that different measures tap different aspects of metacognition is well taken. Focusing purely on gamma as a measure of association may mask interesting effects that would be observed with measures of absolute accuracy. For example, gamma does not indicate whether a subject is showing overconfidence or underconfidence. Alternative measures developed in the context of decision-making studies do measure levels of confidence. 2. Judgment and Decision Making Tradition
Measures that have been used in judgment and decision-making studies require that judgments be in the same units as performance so that subtraction of the two is possible. Subjects are often asked to assign a probability value to possible events (Lichtenstein et al., 1982). Bias is the mean signed difference between judgments and performance (Yates, 1990). With bias, overconfident and underconfident judgments cancel. Other measures such as the Brier score and calibration avoid this canceling problem. The Brier score uses the mean of squared differences between judgments and performance across items (Brier, 1950, cited in Lichtenstein et al., 1982). Murphy (1973) showed that the Brier score can be broken into three components: knowledge (which is unrelated to the judgments), calibration, and resolution. Calibration is the mean squared difference between the mean performance on items in a judgment category (e.g., mean for items judged as having a .6 probability of being correct) and the value assigned to the category (e.g., .6), weighted by the number of judgments in that category. Calibration differs from the Brier score in that a performance mean for items within a judgment category is subtracted from judgment value for calibration, but the performance for each item in the category is used with the Brier score. Resolution measures the variance in performance among judgment categories, or the subject’s ability to discriminate among categories. It is the squared deviation of the proportion correct within each judgment category minus the overall mean proportion correct, weighted by the number of items in the category. Like correlations, resolution is independent of the actual value of the judgments, but, unlike correlations, it does not test for a monotonic relationship between judgments and performance. A summary of these measures is shown in Table 1. Cutler and Penrod (1989) compared these various measures in a face recognition study in which either the same faces or changed faces were presented for recognition. All of the measures, except for resolution, showed more confidence judgment accuracy for same than for changed faces. 3. Measures Used in Educational Psychology Literature
Educational psychologists often ask for global predictions in which subjects indicate how many items they will get correct on an entire test. These
TABLE I A COMPARISON OF MEASURES OF JUDGMENT-PERFORMANCE RELATIONSHIPS ~
Measure Bias
Brier Score
Calibration
Type of data required Interval, mean overall judgments and mean overall performance Interval, performance and judgment for each item
Definition Signed difference between mean judgments and mean performance Mean squared difference between judgments and performance
Interval, mean performance Mean squared difference and judgment for each between mean judgments judgment category and mean performance for each category Resolution Interval, overall mean Mean squared difference between overall performance and mean performance and mean performance for each performance for each judgment category category Square root of proportion Pearson r Interval, judgment and of variance in performance for each item performance accounted for by judgments Gamma Correlation Ordinal, rank of judgment Proportional reduction in error between and rank of performance performance rankings for each item based on judgments and random rankings
Requires Shows judgments and overconfidence performance in or same units underconfidence
Tests for monotonic relationships
Yes
Yes
No
Yes
No
No, but smallest values
Yes
No
occur with perfect judgmentperformance match No, but smallest values occur with perfect judgmentperformance match
Yes
No
No
No
No
Yes, and linear
No
No
Yes
228
Ruth H.Maki
predictions are then usually subtracted from the actual score and either absolute difference (e.g., Pressley et al., 1987) or the signed difference (e.g., Glover, 1989). This signed difference is bias according to Yates (1990). It shows whether a subject is overconfident or underconfident overall, but it does not show whether performance increases as predictions increase. That is, it measures absolute accuracy, but not relative accuracy. Schraw and Roedel(l994) recently showed that bias increases as test difficulty increases; subjects showed more overconfidence on difficult tests than on easy tests. 4. Summary of the Measures of Metacognitive Accuracy
One purpose of the present experiments was to examine the various types of measures of judgment-performance relationships in the context of metacomprehension. Judgments can be understood on two dimensions. One is the overall level of confidence expressed by the judgments. Subjects can either be overconfident and give judgments that are higher than actual performance, or they can be underconfident and give judgments that are lower than actual performance. Bias appears to be the best measure of this; neither calibration, resolution, nor correlations tap overconfidence and underconfidence. The other aspect of metacomprehension is the match between judgments and performance. This can be measured in terms of absolute accuracy with calibration and the Brier score (mean squared deviations between judgments and performance). However, subjects could have fairly low calibration error if all judgments were at the performance mean because the mean minimizes squared deviations. In contrast, calibration error could be large if judgments increased monotonically with performance but were always several units too high or too low. Other measures that tap relative accuracy would be high in the case of monotonicity and low in the mean judgments case. Resolution measures the amount of variance in performance across judgment categories with the idea that different judgments should result in different levels of performance. However, resolution does not specify the function relating judgments and performance. Correlations are necessary to investigate the existence of a monotonic relationship between judgments and performance.
11. Effect of Expectancy and Processing on
Metacornprehension Measures Two experiments were designed to investigate two variables that may affect metacomprehension differently, either in terms of the overall level of subjects’ confidence or in terms of accuracy in an absolute or relative sense.
Absolute Confidence and Metacomprehension
229
Subjects’ confidence was manipulated by giving them either a high or low expectation for their performance. This should affect bias (the signed difference between judgments and performance), but it would not necessarily affect accuracy as measured by calibration or correlations. Manipulating accuracy is more difficult because very few variables influence the accuracy of metacomprehension judgments. Maki and Serra (1992a) showed that gaining more knowledge about the content of texts by reading the texts themselves rather than descriptor sentences about the texts increased the correlation between predictions and performance. Glenberg et al. (1987) and Maki and Serra (1992b) showed that having a pretest that was identical to the actual test improved prediction accuracy. Maki et al. (1994) found that students with higher comprehension ability assessed prior test performance more accurately than students with poorer ability, although the same effect was not seen for predictions of future test performance. Glover (1989) also found that high ability students made more accurate posttest assessments, but Pressley and colleagues (Pressley et al., 1987; Pressley, Ghatala, Woloshyn, & Pirie, 1990) have generally not found a relationship between the ability to assess test performance and verbal ability. A variable that influences memory and that increases metacomprehension accuracy is reading text with deleted letters as compared to reading intact text (Maki et al., 1990). Maki et al. interpreted this finding in terms of an increased level of processing increasing the accuracy of judgments. If this interpretation is correct, then other methods of increasing subjects’ processing of the texts should also have a positive effect on the ability to predict performance. In the studies presented here, processing (and presumably prediction accuracy) was manipulated by having some subjects write reasons about why they would do well or poorly after reading paragraphs of text and by having other subjects not write reasons. This manipulation was patterned after a study by Koriat et al. (1980), who showed that writing reasons for why a given answer to a general knowledge question might be correct and why it might be incorrect improved subjects’ abilities to assess their performance. Their second experiment showed that writing contradictory reasons particularly improved subjects’ abilities to assess the correctness of their answers. Writing reasons for why one will do well over text material or why one might do poorly could have an effect on predictions that is similar to that observed by Koriat et al. (1980). Writing reasons might force subjects to consider the basis of their predictions making them assess the strength of the text in memory more completely than if they simply make predictions without writing reasons. In addition, writing reasons for why one might do well or poorly when tested over a paragraph should increase
230
Ruth H. Maki
the processing of the material in that paragraph as subjects search for reasons. In order to investigate these questions, some subjects wrote reasons for why they would do poorly over a section of text they had just read, some wrote reasons for why they would do well, and other subjects did not write reasons. Subjects who wrote reasons should predict their performance more accurately than those who did not because of more processing of the text. If the Koriat et al. (1980) finding with posttest estimates extends to text predictions, then writing reasons for doing poorly should produce more accurate predictions than writing reasons for doing well. In addition, subjects’ expectancies were varied by informing half of them that other subjects had gotten about 40% correct on test questions and informing the other half that other subjects had gotten about 80% correct. To anticipate the results of the first experiment, students were generally underconfident, but the level of performance on the tests was fairly high. In addition, subjects who were given the higher expectation (4 correct) showed greater relative accuracy than subjects given the lower expectation. Performance was closer to the higher expectation, so it is unclear whether higher accuracy was related to a closer match between expectations and performance or to higher expectations. To test this, more difficult texts were used in a second experiment. If an expectation that is closer to actual performance produces higher relative accuracy, then students who expect to do more poorly might produce greater relative accuracy. A condition in which no specific expectation was given was also used to determine if the lower expectation reduced confidence, the higher expectation increased confidence, or both. Experiment 2 was similar to Experiment 1 so the two will be reported together.
A. METHOD 1. Design and Subjects
The design in Experiment 1 was a 2 X 3 X 2 mixed design. Subjects were given an expectation of two or four correct (out of five). One-third of the subjects wrote reasons for why they would do well, one-third wrote reasons for why they would do poorly, and one-third did not write reasons. The within-subjects variable was the timing of judgments, either before or after the test. Subjects included 96 undergraduate students enrolled in introductory psychology courses at North Dakota State University. All subjects agreed to participate in exchange for extra credit points. Sixteen subjects were randomly assigned to each of the six groups that were defined by reasons and expectations. Six subjects were dropped and replaced because their prediction ratings did not vary across texts. Two of these replaced
Absolute Confidence and Metacomprehension
231
subjects were in the Control, Expect 2 group; two were in the Control, Expect 4 group; and one each was in the Poorly, Expect 4 and the Well, Expect 4 groups. The design of Experiment 2 was a 3 X 2 X 2 mixed factorial. Subjects were randomly assigned to one of three expectancy groups: Expect 2, No Expectancy, and Expect 4. Half of the subjects in each of these groups wrote reasons for why they would do poorly and half wrote reasons for why they would do well. Again, the within-subjects factor was time of rating; each subject made both predictions and posttest estimates of their scores on each text. A total of 141 subjects participated in the second experiment for extra credit in their psychology classes. They were randomly assigned so that 24 subjects participated in the three well reasons conditions (Expect 2, No Expectation, Expect 4), and 23 subjects participated in each of the corresponding three poorly reasons conditions. 2. Materials and Apparatus
Six expository tests having about 400 words each were taken from a timed reading series (Spargo, 1989). The reading series has a number of difficulty levels and two texts were selected from each of three difficulty levels, so there should have been some heterogeneity in the text difficulties. The topics of the texts were: range plants preferred by grazing animals; controlling insect pests; functions and types of food preservatives; the Confederate prison, Andersonville; employee evaluation; and the design of kitchens. Each text was divided in half with about 200 words per half. Five multiplechoice questions that tested the material in the first half and five that tapped material from the second text half were taken from the reading series. These questions had only three alternatives, so a fourth incorrect alternative was added to each to reduce accuracy due to guessing. In addition, two practice texts were selected: one on recycling and one on the origins of drugs. Previous research with these texts had shown that mean performance on the recycling text used in the Expect 2 condition was about 2.0 correct, and the mean for the drugs text used in the Expect 4 condition was about 3.5 correct. Texts, test questions, and instructions were presented on IBM-compatible computer monitors. Response recording was also done by the computers that were programmed with the Micro Experimental Laboratory (MEL) software package (Schneider, 1988). In Experiment 2, six tests were selected from the same source as those in Experiment 1 (Spargo, 1990). Twelve texts (including the six texts from Experiment 1) had been used in a previous experiment (Maki, in press).
232
Ruth H. Maki
The six most difficult texts as judged by multiple-choice test performance were selected for Experiment 2. The mean performance in the earlier study for these texts was .58, or 2.88 correct out of 5. The topics of these texts were origins of drugs, Virgin Islands, home wine making, Andersonville prison, food preservatives, and bird migration. In the Expect 2 condition, the practice text was the same recycling text as in Experiment 1; it had a mean correct performance of 2.0. In the Expect 4 condition, the practice text was on household pests that had a mean of 4.5 correct in earlier research. In the No Expectancy condition, the practice text was the one on range plants which had a mean correct of 3.65. 3. Procedure
In Experiment 1,all subjects read the same six experimental texts, and one of two practice texts. They were informed that a multiple-choice test over the texts would be given after all texts had been read, and that they would be asked to make a prediction as to how many questions they would answer correctly before taking the test, and how many they answered correctly upon test completion. Subjects in the Expect 2 condition were told that the majority of students answer two out of five questions correctly, and were given a difficult practice text to read. Subjects in the Expect 4 condition were told to expect to answer four questions correctly, and were given an easy practice passage. Control subjects were instructed to read the texts, make their predictions, take the tests, and finally estimate the number of questions they answered correctly on the test over each section. Subjects in the reasons conditions were given an additional assignment. Half were instructed to write reasons describing why they would do well on a test over the text passage they had just read, and the remaining half were asked to write reasons describing why they would do poorly on the multiplechoice test over the passage. The text passages were presented in idea units, varying from two to ten words, in the center of the computer screen. Subjects could advance to the next idea unit, at their own pace, by pressing a key on the computer keyboard. This erased one idea unit and replaced it with another. After reading approximately half of the text, subjects were instructed to write the reasons for doing well or poorly on the text section depending upon the experimental condition. Subjects were asked to write two reasons for each of the half texts. Control subjects pressed a key to continue instead of writing reasons. Next, all subjects were instructed to type in the number of questions they expected to answer correctly on the multiple-choice test for each half text. Subjects were allowed to type any number from zero to five. Subjects then read the second half of the passage. At the end of the
Absolute Confidence and Metacornprehension
233
text, experimental subjects were again asked to write their reasons, and all subjects were asked to enter the number of test questions they would answer correctly on the second half of the text. The order of presentation of the six passages was randomized by the computer. The practice passage was followed immediately by test questions, and subjects received feedback on the correctness of their answers for the practice questions. The test questions for the remaining six passages were given after all six texts had been read. Subjects in the reasons conditions were instructed to turn over their reasons sheets while they were answering the test questions on the computer. Five four-alternative, multiple-choice test questions were presented one question at a time. The title of the appropriate passage was presented at the top of the screen above each question. The subjects responded by selecting and highlighting their answers, and pressing the return key. No feedback was given on the correctness of answers. When subjects had completed five test items from a half of a text, they were instructed to estimate how many of the five questions they had answered correctly. Subjects responded to test items from all the texts in this fashion. Sessions lasted approximately one hour. In Experiment 2, all procedures were identical to those of Experiment 1 expect that all subjects wrote reasons (half for why they would do poorly and half for why they would do well) and the No Expectancy group was not given any normative information about how students perform on the texts.
B. RESULTS 1. Ratings and Performance
In the first experiment, the overall mean number of questions correctly answered out of five was 3.70 (SE = .05). A 2 (expectation) X 3 (reasons) analysis of variance (ANOVA) conducted on the actual numbers correct showed no differences among the conditions (Fs[l, 901 5 1.87, M S E = .22). In the second experiment with the more difficult texts, the mean number correct was 2.95 (SE = .06). This did not differ across reason (F < 1) or expectancy groups (F[2, 1351 = 1.10, M S E = .443). Figure 1 shows the mean predicted and the posttest estimated scores in each condition. The results of Experiment 1 are on the left and the results of Experiment 2 are on the right. The Experiment 1estimates were analyzed in a 2 (expectation) X 3 (reasons) X 2 (time of estimate) mixed-design ANOVA. Subjects in the Expect 4 condition gave higher estimates than subjects in the Expect 2 condition ( M = 3.59 versus M = 2.82; F[1, 901 = 70.50, M S E = .41). There was also a main effect of reasons (F[l, 901 = 5.37, M S E = .41). Pairwise comparisons with Tukey corrections showed that the poorly reasons group gave lower estimates than did the well reasons
Ruth H.Maki
234
4.00 3.50 - -
d 3.00 .2.50 ..
1.00
..
0.50 . 0.00
-
Expect2
Expect4
Plsdlctlonr
Expect2
-4
Pomttmt Ertlmatm
Expea None Expeol Exped Nono Expacl 2
4
2
4
Pdlctlonr Posttrrt Estimates
Fig. 1. Mean prediction and posttest estimates in the expectancy and reasons groups for Experiment 1 (left) and Experiment 2 (right) with error bars showing standard errors of the mean.
group ( M = 3.00 versus M = 3.36). The control group gave estimates in the middle (M = 3.24) and did not differ from either the poorly or the well groups. Estimates were higher after the test than before (3.30 versus 3.10; F[1,90] = 25.12, M S E = .15). As can be seen in Figure 1, the difference between the Expect 2 and Expect 4 conditions was larger before the test than after (F[1, 901 = 14.46, M S E = .15, for the interaction). However, the estimates in the Expect 4 condition were significantly higher than in the Expect 2 condition both for prediction estimates (F[1, 901 = 70.50, MSE = .34) and for posttest estimates (F[l, 901 = 34.34, M S E = .22). The ratings from Experiment 2 were analyzed in a 3 (expectation) X 2 (reasons) X 2 (time of rating) ANOVA. Overall, subjects who wrote reasons why they would do well tended to give higher ratings than subjects who wrote why they would do poorly (M = 3.42 versus M = 3.22), although the difference only approached significance (F[1,135] = 3.42, M S E = 1.02, p < .lo). Expecting four correct produced higher ratings ( M = 3.78) than no expectation (M = 3.38), which, in turn, was higher than expecting two correct (M = 2.80; F[2, 1351 = 22.99, MSE = 1.02). As in the first experiment, expectancy interacted with type of ratings because the differences among conditions was greater in the prediction than in the posttest estimates ( F [ 2 , 1351 = 12.46, M S E = .14), but expectancy was significant for both types of ratings. In contrast to Experiment 1, ratings did not change from before to after the test ( M = 3.36 versus M = 3.28; F[1, 1351 = 2.74, MSE = .14.) In both experiments, the expectation manipulation influenced the absolute level of prediction and posttest confidence ratings; the reasons
Absolute Confidence and Metaeomprehensioa
235
manipulation had a significant effect in Experiment 1, but it only approached significance in Experiment 2. 2. Prediction and Performance Relationships
Bias was calculated to examine the overall confidence levels of subjects in the various conditions. The mean signed difference between each subject’s prediction and posttest estimates and their actual scores for each half text was determined. Mean bias for both experiments is shown in Figure 2. Negative bias indicates that subjects were underconfident, giving lower estimates than their actual performance; positive bias indicates overconfidence, giving higher estimates than actual performance. For the first experiment, bias was analyzed in a 3 (reasons) X 2 (expectancy) X 2 (time of rating) mixed-design ANOVA. There was a main effect of reasons (F[2, 901 = 3.20, MSE = 0.63) and of expectancy (F[1, 90) = 62.59, MSE = 0.63). Subjects showed more underconfidence in the poorly condition than in the well condition ( M = -.68 versus M = -.33). The control group fell in the middle ( M = -.47) and was not significantly different from either experimental group. In addition, subjects showed more underconfidence in the Expect 2 than in the Expect 4 condition ( M = -.95 versus M = -.04). There was more bias before the test than after ( M = -.59 versus M = -.40) (F[1, 901 = 12.62, M S E = .15) but time of the estimate interacted with expectancy (F[l, 901 = 15.09, M S E = .15). Although the expectancy effect was significant both for predictions (F[l, 901 = 69.37, MSE = .34) and posttest estimates (F[l, 901 = 34.34, M S E = .22), Figure 2 shows that the difference in bias between the Expect 2 and Expect 4 conditions was larger in the predictions than in the posttest estimates. For Experiment 2, bias was analyzed in a 2 (reasons) X 3 (expectancy) X 2 (time of rating) ANOVA. Expectancy produced a main effect (F[2, 1351 = 21.87, M S E = .66), and expectancy interacted with time of rating (F[2, 1351 = 12.54, MSE = .13). Again, the expectancy effect was significant both for prediction and posttest confidence ratings, but it was larger for predictions. For predictions, pairwise comparisons with Tukey corrections showed that bias was more negative in the Expect 2 than in the No Expectancy condition and more positive in the Expect 4 than in the No Expectancy condition. For posttest estimates, bias was more negative in the Expect 2 condition than in the No Expectancy condition, but bias in the Expect 4 condition was not significantly different from bias in the No Expectancy condition. In contrast to Experiment 1,writing positive and negative reasons did not affect bias significantly (F[l, 1351 = 1.37, M S E = .66), and bias did not change significantly from before to after the test (F[1, 1351 = 2.77, MSE = .13).
Ruth H. Maki
236
1 - 1
Blas wlth Easier Texts
Blas wlth More DifficultTexts
0.70
0.70
a -0.30 -1.30
0 Poorly .Well
.Well -1.80 Expect 2
Expect 4
Expect 2
Expect 4
Predictiona Poattest Estimates
Expect 2
None
Predictions
Expect Expect 4 2
None
Expect 4
Posttest Estlmates
Fig. 2. Mean bias (estimates minus performance) in the expectancy and reasons groups for Experiment 1 (left) and Experiment 2 (right) with error bars showing standard errors of the mean.
Next, I present the mean squared deviations (the Brier Score) and calibration. The Brier score is the mean of the squared deviations between judgments and performance for each item. Calibration is similar except that mean performance for each subcategory of judgments is used rather than performance on each individual item. In the first experiment, the pattern for these two deviation measures was the same, and their correlation was .956 for the predictions and 272 for the posttest estimates. Thus, only the calibration error is shown in Figure 3. High calibration scores mean more error and less accurate judgments. In Experiment 1,there was less calibration error in the posttest estimates than in the predictions ( M = 0.95 versus M = 1.62; F[1,90] = 25.12, M S E = 3 5 ) . and there was less calibration error in the Expect 4 than in the Expect 2 condition ( M = 0.71 versus M = 1.85; F[1, 90) = 45.29, M S E = 1.38). Expectancy condition interacted with time of estimate (F[l, 901 = 14.86, M S E = .85) . Although the difference between the Expect 2 and Expect 4 condition was greater in the predictions than in the posttest estimates, it was significant for both (F[l, 901 = 36.83, M S E = 1.80 for predictions, and (F[1,90] = 20.87, M S E = 0.46 for the posttest estimates). Overall writing reasons had no significant effect on calibration (F[2, 901 = 1.66, M S E = 1.38). Calibration for Experiment 2 is shown on the right side of Figure 3. Posttest estimates were better calibrated (lower squared deviations) than were predictions (.88 versus 1.32), (F[l, 1351 = 23.96, M S E = S6). Calibration did not vary with the type of reasons written by subjects (F < l), but it did vary with expectancy ( F [ 2 , 1351 = 3.53, M S E = .128). Pairwise
Absolute Confidence and Metacomprehension Calibration with Easier Texts
Calibration with More Difficult Texts
3.50
.$
;OPoor(y OQntml .Well
I opoor~y .Well
3.00
W 2.50
~
u 2.50
es 200
C
j 2.00
0
:150
1.50
0
5 100
I
231
0
0 50
-
0 00 Expect2
Expect4
Predlctlons
Expect2
Expect4
Porttest Estimates
'5 l.OO
0.50
0.00 Expect None Expect Expect None Expect 2 4 2 4
Predictions
Posttest Estimates
Fig. 3. Mean calibration error (squared deviations of mean performance from estimates) for the expectancy and reasons groups for Experiment 1 (left) and Experiment 2 (right) with error bars showing standard errors of the mean.
comparisons with Tukey corrections showed that calibration error was greater in the Expect 2 than in the No Expectancy condition, but that calibration error did not differ in the No Expectancy and Expect 4 conditions. Resolution was analyzed next. Resolution scores give an estimate of the variance due to judgments in different categories so that higher scores mean better discrimination among categories. In Experiment 1, the overall mean resolution for the predictions was .249 and the overall mean resolution for the posttest estimates was .386. Resolution scores were significantly higher for the posttest estimates than for the predictions (F[1,90] = 20.03, M S E = .04). There were no differences as a function of expectancy or reasons (Fs[l,90] 5 1.00). In the second experiment, the mean resolution in predictions was .300 and the mean resolution in posttest estimates was .450. Again, resolution was higher in the posttest estimates than in the predictions (F[1, 1351 = 9.81, MSE = .15). The effects of writing reasons and expectancy were not significant (Fs < 1).Thus, in both experiments, subjects produced more variance in performance among their judgment categories after taking the test than before taking it, but manipulations of the absolute levels of judgments did not affect resolution. The final measures of accuracy involved the correlations between predictions and posttest judgments with performance. Both Pearson r correlations and gamma nonparametric correlations were calculated. Although Liberman and Tversky (1993) criticized gamma because it ignores ties and a large number of ties are expected with a dichotomous outcome, ties should
Ruth H.Maki
238
be less of an issue in the present case in which there were as many outcome categories (six) as there were judgment categories. Pearson r correlations were generally lower than the gamma correlations, but the two measures were highly correlated with r = .960 for predictions and r = ,924 for posttest estimates. Because of these high correlations, only the gammas are presented in Figure 4. Single-sample t-tests were conducted to determine whether each gamma was significantly greater than zero. In Experiment 1, in the Expect 2 prediction condition, only the gamma in the poorly condition was significantly greater than zero. In the Expect 4 condition, the prediction gammas in both reasons group were significantly greater than zero, but the gamma in the control group was not. All of the posttest gammas were significantly greater than zero. A 2 x 3 X 2 ANOVA on the gammas for Experiment 1, shown on the left side of Figure 4, yielded a main effect of expectancy ( F [ l , 901 = 4.42, MSE = .17), and a main effect of time of estimate (F[l, 901 = 15.58, MSE = .14). Correlations were higher in the Expect 4 than in the Expect 2 condition (M = .39 versus M = .27) and for posttest estimates rather than for predictions (M = .44 versus M = .23). There was also an interaction between expectancy and reason condition ( F [ l , 901 = 4.03, MSE = .17). This interaction was not replicated in the second experiment, so it won’t be described further. Mean gamma correlations for Experiment 2 are shown in the right half of Figure 4. Single-sample f-tests indicated that the mean prediction gammas were significantly greater than zero only in the No Expectancy, poorly condition, but they approached significance (p < .lo) in both Expect 2 conditions and the No Expectancy, well condition. Gammas were not sig-
l,::: 1 -
Gamms with Eulsr Texts OPoorly ONone .Well 0.60
3
,
B ::::
[ 050 8
Gamma wlth Mom DlfflcultTexts
0.80
0.20
040
030 0 20
0.10
0 10
0.00
0 00 Expect2
Expect4
Pmdlctlons
Expect2
Expect4
POSttMt ESt1nut.r
E x W None Expect ExpeCt None Expect 2 4 2 4
Pmdlctlons
Poattsst Eatlmatea
Fig. 4. Mean gamma correlations for the expectancy and reasons groups for Experiment 1 (left) and Experiment 2 (right) with error bars showing standard errors of the mean.
Absolute Confidence and Metacomprehension
239
nificantly greater than zero in the Expect 4 conditions. All mean posttest gammas were significantly greater than zero. Overall, gammas were higher for posttest estimates than for predictions (M = .38 versus M = .16; F[1, 1321 = 22.35, MSE = .15). However, gammas did not vary as a function of expectancy (F[2, 1321 = 1.78, MSE = .19) or type of reasons (F < l), and these two variables did not interact (F < 1). 3. Evaluation of the Measures as Valid Measures of Metamemory, Accuracy
Most of Nelson’s (1984) six criteria for a good measure of feeling-of-knowing, and, presumably, more generally of metamemory, require that the measure not relate to the level of performance. Nelson pointed out that this criterion may not hold psychologically, but indicated that the measure should not force a violation of independence. There is some evidence that this criterion does not hold psychologically.First, there are data that suggest that individuals who have higher ability (and, hence, who have higher performance) are more accurate on several metamemory tasks (e.g., Baker, 1985; Glover, 1989; Maki et al., 1994; Shaughnessy, 1979; Sinkavich, 1995; Zechmeister, Rusch, & Markell, 1986) although there are studies that do not show this ability-metamemory accuracy relationship (Cull & Zechmeister, 1994; Pressley et al., 1987; Pressley & Ghatala, 1988). To the extent that there is a relationship between memory ability and metamemory ability, then performance and metamemory accuracy will not be independent. Second, task difficultymay also be related to metamemory accuracy. Schraw and Roedel (1994) found that absolute accuracy of confidence judgments varied with task difficulty. Weaver and Bryant (1995) suggested that medium difficulty texts produced the strongest relationship between performance and ratings. I conducted two types of analyses to determine whether the various measures were related to levels of test performance. The left side of Table 2 shows Pearson r correlations between overall performance on the multiplechoice tests and the metacomprehension measures across subjects from both experiments. For predictions, bias and resolution were related to peformance. The higher the number correct, the more negative bias and the lower the resolution. For posttest estimates, gammas tended to be higher with better performance, but resolution tended to be lower. The finding with gamma is consistent with earlier studies (Maki et al., 1994) and suggests that there is a weak positive psychological relationship between posttest accuracy and test performance. This relationship must not be forced by the measure because it was not found for predictions. Resolution, a measure of variance in performance among judgment categories, is different from gamma in that it produced a negative relationship with performance.
Ruth H. Maki
240
TABLE I1 CORRELATIONS OF THE MEASURES OF METACOMPREHENSION WITH ESTIMATES AND PERFORMANCE Number correct
Bias Calibration Resolution Gamma
Number estimated
Predictions
Posttest estimates
Predictions
Posttest estimates
- .269** .070 -.181** - .076
- ,089 .001 -.135* .161*
.727** -.326**
.658** -.226** - .098 .212**
-.130*
,046
* p ‘z .05. * * p < .01.
The second way of examining the relationship between metamemory measures and test performance is to compare the measures when they are applied to easy and difficult tests. Nelson (1984) argued that metamemory measures should not be influenced by test difficulty. The data for the Expect 2 and the Expect 4 and for the good and poorly reasons conditions were analyzed with experiment as a variable because Experiment 1 used easier texts than Experiment 2. A summary of the results across experiments, along with a summary of the correlation results, can be seen in Table 3. Overall, performance was higher in Experiment 1 with easier texts than in Experiment 2 with more difficult texts ( M = 3.7 versus M = 2.95; F[1, 1501 = 51.37, M S E = .384. Bias was less negative (showing less underconfidence) with the more difficult texts in Experiment 2 than with the easier texts in Experiment 1 ( M = .03 versus M = -3; F[1, 1501 = 23.63, M S E = .69). Calibration did not differ across experiments ( M = .128 versus M = .120, for Experiments 1 and 2; F < 1).Resolution also did not differ across the experiments ( M = .442 for Experiment 1 versus M = .375 for Experiment 2; F[1,150] = 1.01,MSE = .19). Overall, the gamma correlation was higher with the easier texts of Experiment 1 ( M = .33) than with the more difficult texts of Experiment 2 ( M = .27; F[1, 1481 = 6.60, M S E = .18). This difference in gamma was not due to there being less variability in the text difficulty in Experiment 2 than in Experiment 1 because the standard deviation in scores was actually greater in Experiment 2 (SD = .66) than in Experiment 1 (SD= .49). It may, however, have been due to a different sample of subjects because they were tested in different semesters although they were volunteers from the same courses. Interestingly, bias and the gamma correlations showed opposite effects across experiments with texts varying in difficulty. Bias was less in Experi-
TABLE I11 EFFECTS OF CHANGING
LEVELS, RATING LEVELS, AND ASSUMED METACOGNKIVE ABILITY ON MEASURES OF ABSOLUTE AND RELATIVE ACCURACY
PERFORMANCE
Higher test performance related to: Measure Bias Calibration Resolution Gamma
I
Overall Correlation
Exp. 1 vs. Exp. 2
More negative No effect Less variance Greater accuracy2
More negative No effect No effect Greater accuracy
For predictions only;not for posttest estimates.
* For posttest estimates only; not for predictions.
Higher assumed metacomprehension ability related to: Exp. 1
Exp. 2
Higher absolute level of ratings related to: Overall Correlation
Less negative No effect More positive Less error Less error Less error More variance More variance Less variance’ More accuracy More accuracy Greater accuracyZ
Expect 2 vs. Expect 4, Exp. 1
Expect 2 vs. Expect 4, Exp. 2
More positive
More positive Less error No effect No effect
Less error No effect Greater accuracy
242
Ruth H. Maki
ment 2 with the more difficult texts than in Experiment 1 with the easier texts. Indeed, bias was very close to zero in the second experiment showing that overall ratings were very close to overall performance. This may have been because subjects tended to use the middle of the scale and performance was closer to the middle with the more difficult texts. However, relative accuracy was greater with the easier texts of Experiment 1.This is consistent with the correlational analysis in that bias correlated negatively with performance, but posttest gammas correlated positively with performance (although prediction gammas did not correlate with performance overall). The only measure that did not relate to overall level of performance was calibration. Each of the other measures, bias, resolution, and gamma, related to level of test performance in at least one analysis. Higher performance was associated with more bias in the form of underconfidence, less resolution, but higher gammas. The higher gammas suggest that subjects who performed better on the test were more accurate in their relative posttest estimates than those who performed more poorly. The alternative, that the measures force the relationship, seems unlikely because gamma did not relate consistently to performance in all conditions. Another characteristic of a valid measure of metamemory accuracy according to Nelson (1984) is that it should increase as metamemory ability increases. Many studies have shown that posttest accuracy is higher than prediction accuracy (e.g., Glenberg & Epstein, 1985; Maki & Serra, 1992). Thus, it is a reasonable assumption that metacomprehension ability is greater after a test than before a test. The middle columns of Table 3 show which measures increased in accuracy from before to after the test. For bias, this difference varied with experiment. In Experiment 1 with easier texts, bias became more positive from pretest to posttest. In Experiment 2, with more difficult texts, it did not change significantly from pretest to posttest. Thus, this measure of absolute accuracy depends more upon the difficulty of the test than the amount of information available to the subject in making metacognitive judgments. Each of the other measures of accuracy improved from pretest to posttest, as they should if the ability to make judgments increased with more information about the test. A difference between absolute and relative measures of metacognitive accuracy is that absolute measures depend upon the absolute level of estimates, but relative measures do not necessarily vary with absolute judgment levels. The right side of Table 3 shows the relationship between absolute judgment levels and metacomprehension measures. When the overall levels of judgments were correlated with bias (using a Pearson r), higher judgments were related to more positive bias. In addition, higher judgments related to less calibration error. For predictions, higher judgments related to lower resolution, but there was no relation in the posttest estimates.
Absolute Confidence and Metacomprehension
243
Finally, higher ratings related to more accuracy as measured by gamma, but only for posttest estimates, and not for predictions. Again, resolution and gamma showed very different patterns. The other way of investigating absolute judgment levels and the measures is to compare measures in the Expect 2 and Expect 4 conditions because judgments were always higher in the Expect 4 conditions. There was more positive bias and less calibration error in the Expect 4 than in the Expect 2 conditions in both experiments. Resolution did not vary with expectancy condition in either experiment. In Experiment 1, gammas were higher in the Expect 4 than in the Expect 2 condition, but this effect was not seen in Experiment 2. C. CONCLUSIONS FROM THE EXPERIMENTS The expectancy manipulation had a strong effect on the absolute level of ratings. Subjects who were given the low expectancy produced lower predictions and posttest estimates than subjects given the higher expectancy. In Experiment 2, the lower expectancy reduced the absolute level of ratings and the higher expectancy increased the absolute level of ratings relative to no expectancy. Such differences in the absolute levels of ratings affected absolute measures of metacomprehension accuracy. In Experiment 1, bias showed that subjects were underconfident in the Expect 2 condition and that they were very accurate in the Expect 4 condition. In Experiment 2 with more difficult texts, subjects were still underconfident in the Expect 2 condition, they were very accurate in the No Expectancy condition, but they were overconfident in the Expect 4 condition. Although overconfidence is the norm in most judgment studies (Fischhoff, Slovic, & Lichtenstein, 1977), the present study shows that overconfidence and underconfidence can be manipulated rather easily by changing test performance expectations. Underconfidence is also more likely when subjects make estimates of number correct aggregated over multiple items rather than estimates concerning the prpbability that each single item is correct (Gigerenzer, Hoffrage, & Kleinbolting, 1991). In the present case, subjects made estimates on their performance on five test items for each half text. The need to aggregate over individual test items, along with the expectancy manipulation, may have been responsible for underconfidence in most of the conditions. The expectancy manipulation also produced effects on the various measures of accuracy. The pattern suggests that having accurate expectations results in the most accurate metacomprehension. In Experiment 1,calibration was poorer in the Expect 2 than in the Expect 4 condition, and estimates matched performance most closely in the Expect 4 condition. In Experiment
244
Ruth H. Maki
2, calibration was poorer in the Expect 2 than in the No Expectancy condition, and, again, estimates matched performance most closely in the No Expectancy condition. Although these calibration results may have been forced by the nature of the measure, gamma showed a similar pattern. In Experiment 1, higher gammas were found in the Expect 4 than in the Expect 2 condition. Although this result was not replicated in Experiment 2, prediction gammas were significantly greater than zero only in the No Expectancy, “poorly” condition, and the No Expectancy condition produced absolute estimates that most closely matched performance. Resolution was not influenced by the absolute level of estimates. Writing reasons produced much weaker effects than did differing expectancies. In Experiment 1,subjects who wrote reasons about why they would do well gave higher estimates than subjects who wrote reasons for why they would do poorly. This translated into more bias in the form of underconfidence in the poorly condition relative to the well condition. In Experiment 2, the difference in the absolute level of estimates for the “well” and “poorly” groups was only marginally significant, and bias did not differ for the two reasons groups. Gamma was not affected in any consistent way by writing reasons. However, in both experiments, the condition that produced estimates that most closely matched actual performance produced significant gammas when subjects also wrote reasons for why they might do poorly. Many of the conditions in the present experiments produced very low correlations between predictions and performance. The present procedures involved making predictions immediately after reading, but the test was delayed. This procedure requiring immediate predictions with delayed testing is similar to what Nelson and Dunlosky (1991) called immediate judgments of learning. They found low prediction accuracy in this condition with word pairs, and we have found a similar pattern with predictions over text material (Maki, in press).
III. Recommendations for Metacornprehension Measures The comparison of measures suggests which measures are most useful in studies of metacomprehension. In the present experiments, bias correlated both with number correct and with absolute levels of estimates. It did not increase consistently with assumed increases in metacomprehension ability, that is, from pretest to posttest. Thus, using Nelson’s (1984) criteria, bias is a poor measure of metacognitive accuracy. However, bias is the only measure that shows underconfidence or overconfidence. Researchers using a decision-making framework have been interested in overconfidence (e.g., Lichtenstein et al., 1982), but researchers in the cognitive literature have
Absolute Confidence and Metacornprehension
245
generally not used bias. Although the present experiments have shown that it is quite easy to manipulate overconfidence and underconfidence by changing expectations, knowing about the absolute level of confidence as a function of the experimental conditions may often be theoretically interesting. Because bias is the only measure that allows this, researchers should consider using it; however, to do so, they must collect estimates on an absolute scale (actual number or percent correct) rather than on a rating scale. Calibration is also a measure of absolute rather than relative accuracy. However, because the deviations of estimates from actual performance are squared, overestimates and underestimates do not cancel as they do in bias. In the two experiments reported here, calibration did not vary with the overall level of performance, both when performance was correlated with calibration error and across Experiments 1 and 2 with texts of different difficulty. Calibration improved from pretest to posttest in both experiments. One problem with calibration was that it varied with the absolute level of estimates in that there was less error when higher estimates were given. Overall, however, calibration was a good measure of metacognitive accuracy according the Nelson’s (1984) criteria. Nelson did not discuss calibration because he was dealing with 2 X 2 contingency tables (with “Yes” and “No” predictions and correct and incorrect responses). Researchers in the metacognitive literature should consider using calibration, but for calibration, as with bias, the units of estimate must be the same as the units of performance, that is rating scales cannot be used. Resolution was a reasonably good measure of relative accuracy. It did not differ with test difficulty across experiments, although it did relate negatively to performance in an overall correlation. Resolution in the predictions correlated overall with absolute level of estimates, but this was not true for posttest estimates. In addition, resolution did not differ in the Expect 2 and Expect 4 conditions of either experiment. Resolution increased from pretest to posttest estimates. This pattern suggests that resolution should be a recommended measure of relative accuracy. However, higher resolution does not indicate higher accuracy. It is a measure of the variance among performance categories but the nature of the relationship between performance and estimates is not specified by resolution. That is, the performance categories are treated as nominal variables and resolution does not detect whether estimates actually increase with increasing performance (i.e., there is no requirements of an ordinal relationship). Thus, although resolution showed the characteristics of a good measure of relative accuracy, it may not address the interesting theoretical question in metacognition experiments. That is, resolution does not tell the researcher if changes in estimates are accompanied by corresponding changes in performance.
246
Ruth H. Maki
The results with gamma were mixed. On the positive side, gamma increased from pretest to posttest in both experiments, but all measures, except bias, showed this effect. Gamma also showed some less desirable properties. Posttest gammas increased with better test performance, although this may have been because better test scorers are better predictors. Gammas were also higher with the easier texts of Experiment 1 than with the more difficult texts of Experiment 2. Posttest estimate accuracy correlated with the absolute level of the estimates given, and this effect also showed up when the Expect 2 and Expect 4 conditions were compared in Experiment 1 (although it was not significant in Experiment 2). Thus, gamma showed a pattern that indicated more potential problems than did calibration. Alternatively, it may be that calibration was a less sensitive measure, and there really is a psychological relationship between metacomprehension ability and comprehension ability that was detected with gamma and not with calibration. In addition to these problems with gamma, Liberman and Tversky (1993) criticized its treatments of ties, and Weaver and Keleman (1997) showed that it is highly sensitive to the nature of the distribution of estimates. Given the present results, I recommend that researchers use bias in studies of metamemory and metacomprehension to show the overall level of underconfidence or overconfidence. Thus, if possible, researchers should use a scale for estimates that is in the same units as performance. The choice of a second measure of accuracy is more difficult. In some ways, calibration appeared to be a better measure of metacomprehension accuracy than the relative measures (resolution and gamma) because calibration was not sensitive to the level of test performance, and the relative measures were in some cases. But, calibration varied with the absolute level of estimates more consistently than gamma and resolution. Theoretically, the difficulty with resolution is that it does not test specifically for a monotonic relationship between subjective estimates and performance. Gamma (and the Pearson r if a researcher assumes interval data) test for a monotonic relationship between estimates and performance, but researchers should be aware that it may not be independent either of level of test performance or absolute level of estimates. Given that each of these measures is flawed, the use multiple measures of metacomprehension is recommended in order to adequately understand how variables affect students’ abilities to assess their knowledge.
REFERENCES Baker, L. (1985). Differences in the standards used by college students to evaluate their comprehension of expository prose. Regarding Research Quarterly, 20, 297-313.
Absolute Confidence and Metacomprebension
247
Cull, W. L., & Zechmeister, E. B. (1994). The learning ability paradox in adult metamemory research: Where are the metamemory differences between good and poor learners? Memory and Cognition, 22, 249-257. Cutler, B. L., & Penrod, S. D. (1989). Moderators of the confidence-accuracy correlation in face recognition: The role of information processing and base-rates. Applied Cognitive Psychology, 3, 95-107. Fischhoff,B., Slovic, P., & Lichtenstein,S. (1977). Knowing with certainty:The appropriateness of extreme confidence. Journal of Experimental Psychology: Human Perception and Performance, 3, 552-564. Gigerenzer, G., Hoffrage, U., & Kleinbolting, H. (1991). Probabilistic mental models: A Brunswikian theory of confidence. Psychological Review, 98, 506-528. Glenberg, A. M., & Epstein, W. (1985). Calibration of comprehension.Journal of Experimental Psychology: Learning, Memory, and Cognition, 11, 702-718. Glenberg, A. M., Sanocki, T., Epstein, W., & Morris, C. (1987). Enhancing calibration of comprehension. Journal of Experimental Psychology General, 116, 119- 136. Goodman, L. A., & Kruskal, W. H. (1954). Measures of association for cross classifications. Journal of the American Statistical Association, 49, 732-764. Glover, J. A. (1989). Reading ability and the calibrator of comprehension. Educational Research Quarterly, 13, 7-11. Koriat, A., Lichtenstein, S., & Fischhoff,B. (1980). Reasons for confidence. Journal of Experimental Psychology: Human Learning and Memory, 6, 107-118. Liberman, V., & Tversky, A. (1993). On the evaluation of probability judgments: Calibration, resolution, and monotonicity. Psychological Bulletin, 114, 162-173. Lichtenstein, S., Fischhoff, B., & Phillips, L. D. (1982). Calibration of probabilities: The state of the art to 1980. In D. Kahneman, P. Slovic & A. Tversky, Judgment under uncertainty: heuristics and biases (pp. 306-334). Cambridge: Cambridge University Press. Maki, R. H. (in press). Predicting performance on text: Delayed versus immediate predictions and tests. Memory and Cognition. Maki, R. H., Foley, J. M., Kajer, W. K., Thompson, R. C., & Willert, M. G. (1990). Increased processing enhances calibration of comprehension. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 609-616. Maki, R. H., Jonas, D., & Kallod, M. (1994). The relationship between Comprehension and metacomprehension ability. Psychonomic Bulletin and Review, 1, 126-129. Maki, R. H., & Serra, M. (1992a). The basis of test predictions for text material. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 116-126. Maki, R. H., & Serra, M. (1992b). Role of practice tests in the accuracy of test predictions on text material. Journal of Educational Psychology, 84, 200-210. Murphy, A. H. (1973). A new vector partition of the probability score. Journal of Applied Meteorology, 12, 595-600. Nelson, T. 0. (1984). A comparison of current measures of accuracy of feeling-of-knowing predictions. Psychological Bulletin, 95, 109-133. Nelson, T. 0. (1996). Gamma is a measure of the accuracy of predicting performance on one item relative to another item, not of the absolute performance on an individual item. Applied Cognitive Psychology, 10, 257-260. Nelson, T. O., & Dunlosky, J. (1991). When people’s judgments of learning (JOLs) are extremely accurate at predicting subsequent recall. Psychological Science, 2, 267-270. Nelson, T. O., Gerler, D., & Narens, L. (1984). Accuracy of feeling of knowing judgments for predicting perceptual identification and relearning. Journal of Experimental Psychology: General, 113, 282-300.
248
Ruth H.Maki
Pressley, M., & Ghatala, E. S. (1988). Delusions about performance on multiple-choice comprehension tests. Reading Research Quarterly, 23, 454-464. Pressley, M., Ghatala, E. S., Woloshyn, V.. & Pirie, J. (1990). Sometimes adults miss the main ideas and do not realize it: Confidence in responses to short-answer and multiple-choice comprehension questions. Reading Research Quarterly, 25, 232-249. Pressley, M., Snyder, B. L., Levin, J. R., Murray, H. G., & Ghatala, E. S. (1987). Perceived readiness for examination performance (PREP) produced by initial reading of text and text containing adjunct questions. Reading Research Quarterly, 22, 219-236. Romesburg, H. C. (1984). Cluster analysis for researchers. London: Wadsworth, Inc. Schneider, W. (1988). Micro Experimental laboratory (MEL). [Computer program]. Pittsburgh, PA: Psychology Software Tools. Schraw, G . (1995). Measures of feeling-of-knowing accuracy: A new look at an old problem. Applied Cognitive Psychology, 9, 321-332. Schraw, G., & Roedel, T. D. (1994). Test difficulty and judgment bias. Memory and Cognition, 22, 63-69. Shaughnessy, J. J. (1979). Confidence-judgment accuracy as a predictor of test performance. Journal of Research in Personality, 13, 505-514. Sinkavich, F. J. (1995). Performance and metamemory: Do students know what they don’t know? Journal of Instructional Psychology, 22, 77-87. Spargo, E. (1989). Timed readings (3rd ed., Books 5 , 7, and 10). Providence, RI: Jamestown Publishers. Weaver, C. A., 111 (1990). Constraining factors in calibration of comprehension. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 214-222. Weaver, C. A., 111, & Bryant, D. S. (1995). Monitoring of comprehension: The role of text difficulty in metamemory for narrative and expository text. Memory and Cognition, 23, 12-22. Weaver, C. A., 111, & Keleman, W. L. (1997). Judgments of learning at delays: Shifts in response patterns or increased metamemory accuracy? Psychological Science, 8,318-321. Wright, D. B. (1996). Measuring feeling of knowing: Comment on Schraw (1995). Applied Cognitive Psychology, 10, 261-268. Yates, J. F. (1990). Judgment and decision making. Englewood Cliffs, NJ: Prentice-Hall. Zechmeister, E. B., Rusch, K. M., & Markell, K. A. (1986). Training college students to assess accurately what they know and don’t know. Human Learning, 5, 3-19.
LINKING OBJECT CATEGORIZATION AND NAMING: Early Expectations and the Shaping Role of Language Sandra R. Waxman
I. Introduction This chapter addresses fundamental issues of early conceptual development, language development, and the relation between them. Infants across the world’s communities are raised in enormously rich environments populated with novel objects and events. Because this diversity would be overwhelming if each object were treated as unique, an essential developmental task is to form categories that capture the commonalities among objects and to learn words to express them. I will argue that these developmental tasks are not independent. On the contrary, from the earliest stages of word learning, there are powerful, implicit expectations linking object categorization and naming. For infants on the brink of word learning, novel words (both count nouns and adjectives) highlight commonalities among objects and, in this way, foster the formation of object categories. This initial expectation serves three fundamental functions. First, it guides infants in their earliest efforts to establish object reference. Second, it promotes the establishment of a stable conceptual system that promotes the organization of object categories and the acquisition of additional information about category members. Third, this initial expectation sets the stage for the acquisition of more specific expectations linking particular types of words (count nouns versus adjectives) to particular types of relations among obTHE PSYCHOLOGY OF LEARNING AND MOTIVATION. VOL. 38
249
Copyright 8 1998 by Academic Press. All rights of reproduction in any form reserved. 0079-7421/98 $25.00
250
Sandra R Waxman
jects (object categories versus object properties) (Waxman & Markow, in press). These latter expectations are shaped by the structure of the native language under acquisition, and become more entrained with age (Waxman, Senghas, & Benveniste, 1997). The overarching goal of this chapter is to underscore the vital interaction between the child’s expectations and the shaping role of the environment in this process (Waxman, in press). I gather together recent developmental and crosslinguistic evidence to illustrate the origin and scope of infants’ early expectations, and to discover how these are shaped over the course of development.
IT. The Developmental Process: A Dynamic Interplay between the Constraints in the Learner and the Amount and Information Present in the Environment A portrait can capture beautifully an individual at any one moment in time, but no matter how finely it is drawn, it cannot reveal the dynamic processes and influences that have shaped the individual. Similarly, detailed analyses of ecosystems can catalog exhaustively the elements, life-forms, and relations among them at one point in time, but cannot capture the original state of the system or the pressures motivating its changes over time. To discover the origins of any system and the mechanisms responsible for its unfolding, a developmental approach is required. Evidence regarding the acquisition of any complex system sheds light on the mature state and on the processes underlying its evolution. The significance of adopting a developmental approach to questions of acquisition and change has now been widely recognized across a range of disciplines. See, for example, Marler (1991) on the acquisition of birdsong in the white-crowned sparrow, Held and Hein (1963) on the acquisition of depth perception in kittens, Baillargeon (1993) and Spelke (1993) on infants’ acquisition of physical knowledge about objects, and R. Gelman (1991) on the acquisition of number concepts in humans. Although these research programs focus on vastly different topics, they share a commitment to characterizing the rapid acquisition of complex systems by considering the interplay between the structural constraints imposed by the learner and the amount and type of information present in the environment. (See Gallistel, Brown, Carey, Gelman, & Keil(l991) for an extended discussion.) I have adopted a similar strategy to examine the relation between object categorization and naming. Together with my colleagues, I have argued for a dynamic account that embraces both the expectations held by the child and the shaping role of the environment. I have also argued for a
Linking Object Categorization and Naming
251
continuous view of development, in which infants’ and young children’s considerable perceptual and conceptual abilities are recruited early in the fundamental task of forming object categories and mapping words to their meanings. In our view, early development is guided by general, broad initial constraints or expectations that direct the infant’s attention toward precisely the sort of regularities in the environment that will facilitate the rapid acquisition of complex systems of knowledge, including the acquisition of word meaning and object categories (Gelman & Williams, 1998). These broad initial expectations are themselves fine-tuned over the course of development. This interplay between factors within the child and factors within the environment (including the objects the child encounters and the structure of the native language under acquisition) is essential. Children raised in different cultures will encounter different objects, and will acquire different languages. Acquisition must be sufficiently constrained to permit the child to form fundamental categories of objects and to acquire their native language, yet sufficiently jlexible to accommodate the systematic variations that occur across cultures and languages.
111. A Vignette Infants living in remote villages in Tibet and infants living in Manhattan grow up in worlds filled with objects and events that the other has not experienced, and with words that the other cannot understand. Yet despite these vast differences, the conceptual and language development of infants in these diverse communities will be strikingly similar. Within the first year of life, infants in each of these communities will form categories that will capture both the similarities and differences among the objects they encounter. Most of these early object categories will be at the basic level (i.e., dog, rather than the more inclusive animal or the more specific terrier). Infants will use these early categories as an implicit inductive base to make inferences about the behaviors and properties of new objects. In addition to these conceptual advances, the infants in each community will acquire their native language naturally, at a remarkably swift pace. By their first birthdays, they will begin to produce words, most of which will refer to salient objects and categories of objects. By their second birthdays, they will have mastered hundreds of words and will begin to combine these words to produce short, well-formed phrases. Thus, although infants are exposed to widely varying types of experiences, they follow similar paths in their conceptual and language development.
252
Sandra R. Waxman
IV. The Relation between Object Categorization and Naming: Background Issues
In the past decade, there has been a decisive renewal of scientific interest in questions regarding object categorization, object naming, and the relation between these. A central focus has been to discover whether and how the categorization of objects-a conceptual task-is influenced by the introduction of novel words. More recently, attention has been focused on how the categorization of objects is influenced by semantic aspects of the particular language under acquisition (Bowerman, 1996; Imai & Gentner, 1997; Naigles & Eisenberg, in press; Lucy, 1996; Waxman, Senghas et al., 1997). Object categorization has figured centrally because it is a fundamental building block of human cognition, “ . . . a contrivance for the best possible ordering of the ideas of objects in our minds . . . in such a way as shall give us the greatest command over our knowledge already acquired, and lead most directly to the acquisition of more” (Mill, 1843). Because object categorization provides an exceptionally efficient and powerful foundation for organizing existing knowledge and for discovery (Medin & Heit, in press; Shipley, 1993), it should be especially valuable early in development, as infants encounter new objects and witness new events. Indeed, an essential task of early human development is to form categories that capture the commonalities among objects and to learn words to express them. Infants accomplish each of these tasks naturally. There is now ample evidence that infants appreciate many different kinds of relations among objects, including object categories based on taxonomic relations (e.g., flamingos and dogs are both animals). They also appreciate thematic groupings (e.g., flamingos run on sand), associative or ad hoc groupings (e.g., flamingos call to mind a tropical beach) and causal relations (e.g., flamingos sit on nests to hatch their eggs). These conceptual advances are remarkable in and of themselves. But they also set the stage for what has been described as the “induction problem” (Gleitman, 1990; Quine, 1960). To understand this problem, consider a typical word-learning scenario, in which an adult introduces a child to a novel object (say, aflumingo) and offers a novel label (“That is a flamingo”). How do infants discover which relation the new word is intended to convey? How do they so rapidly learn that a given word (e.g.,flumingo) applies to a particular whole object, that it may be extended to other members of that object category (e.g., other flamingos), but not to salient parts or properties of the object (e.g., its long neck or unusual color), to salient actions in which it is engaged (e.g., feeding its young), or to salient thematic relations (e.g., a flamingo and palm trees)?
Linking Object Categorization and Naming
253
A. LINKING WORDLEARNING AND CONCEP~UAL DEVELOPMENT To account for infants’ seemingly effortless solution to this logically difficult task, several scholars have proposed that there may be constraints on acquisition, and that these may lead infants to favor some types of conceptual relations over others when ascribing meaning to a new word (Chomsky, 1986; Landau & Gleitman, 1985; Pinker, 1984; Markman, 1989; Waxman, 1990,1991). There is now substantial evidence for this position. One of the most robust findings is that infants and toddlers expect that a novel word, applied ostensively to an object, refers to the whole named object and can be extended to other members of the same object category (Golinkoff, Mervis, & Hirsh-Pasek, 1994;Markman & Wachtel, 1988;Taylor & Gelman, 1988;Hall, Waxman, & Hunvitz, 1993;Markman, 1989; Mervis, Johnson, & Scott, 1993; Soja, Carey, & Spelke, 1991, 1992; Waxman & Hall, 1993; Waxman & Markow, 1995). More specifically, their tendency is to extend the word to an object category at (roughly) the basic level (Golinkoff, Shuff-Bailey, Olguin, & Ruan, 1995; Gopnik & Meltzoff, 1992; Mervis & Mervis, 1988 Saah et al., 1996; Waxman & Senghas, 1992). This tendency reflects both the infants’ appreciation of the perceptual and conceptual similarity among objects at this level, and the naming practices of the adult community. Perhaps even more remarkable is the fact that children as young as 2 years of age consistently use the syntactic form of a novel word as a clue to its meaning. For example, English-speaking children expect that a count noun, applied ostensively to an object (“That is a blicket”), will refer to the whole object and other members of its kind (e.g., flamingo, animal), but their expectations for novel words presented as proper nouns or adjectives is quite different. They readily map a proper noun (“That is Blicket”) to the named individual, but they do not extend this to other members of its kind (Bloom, 1994; Hall, 1991,1994; Katz, Baker, & Macnamara, 1974; Gelman and Taylor, 1984). They expect that adjectives (and other modifiers) (“That is a blickish one”) will mark object properties (e.g., color, textures, size) and will mark distinctions within a basic level kind (Brown, 1957; Prasada, 1996; Smith, Jones, & Landau, 1992; Taylor & Gelman, 1988; Waxman, 1990 Klibanoff & Waxman, 1997; Waxman & Markow, in press). Interestingly, children’s ability to recruit these syntactic cues in word learning is influenced deeply by their existing knowledge. Children are most likely to use syntactic cues to word learning when they are introduced to a novel word for a familiar object-an object for which the child knows a basic level name, (e.g., a horse) (Au, 1990 Hall et al., 1993; Markman & Wachtel, 1988;Taylor & Gelman, 1988;Waxman, 1995). Under this circum-
254
Sandra R. Waxman
stance, children rely systematically on syntactic form, as described above. But when a novel word is applied to an unfamiliar object (an object for which the child knows no basic level name, e.g., an armadillo), children tend to extend that word, independent of its syntactic form, to the basic level object kind (Hall et al., 1993). Thus, children’s interpretation of a novel word is mediated by their familiarity with a basic level name for that object. This suggests that there is a strong conceptual priority for establishing names for basic level categories (a.k.a., kinds or sortals). This priority is consistent with arguments pertaining to the logic of count nouns (Macnamara, 1994). Basic level count nouns provide principles of object individuation and object identity (Hall, 1993; Hall & Waxman, 1993;Macnamara, 1982);these terms also are the gateway for the acquisition of additional words, referring to other aspects of the named object. To summarize, early object categorization and naming are not independent. On the contrary, children harbor strong, implicit, and precise expectations linking particular types of words (e.g., count nouns, adjectives) with particular types of conceptual relations (e.g., categories of objects, properties of objects, respectively). These expectations reveal children’s linguistic capacity to distinguish among the relevant syntactic forms, their conceptual and perceptual ability to appreciate various relations among objects, and finally, their tacit expectation that these systems are linked. These linkages have been invoked to help explain how children so rapidly map novel words to meanings, and so successfully construct object categories and taxonomic systems of organization (Markman, 1989; Waxman, 1990, 1994). B. SOMECONTROVERSIES REGARDING THE ORIGIN AND EMERGENCE OF THE LINKAGES BETWEEN WORDLEARNING AND CONCEPTUAL DEVELOPMENT These discoveries have received considerable attention, in part because they offered empirical evidence for abstract linkages between syntax and semantics (e.g., Chomsky, 1965; Gleitman, 1990; Grimshaw, 1994). The evidence from children was also notable because it offered a developmental parallel to Berlin’s (1992) discovery of a universal, systematic relation between object naming and categorization (Waxman, in press). Yet this claim has also generated considerable controversy. One locus of controversy concerns the origin of these expectations. When do they become available to the learner? Another concerns their developmental course. Do these expectations exert a uniform influence across development, or are they modified as a consequence of interaction with the objects in the environment and the language under acquisition? A third concern
Linking Object Categorization and Naming
255
has focused on the characterization of infants’ early object categories and the role of perceptual versus conceptual bases in categorization and naming. These concerns are addressed throughout the chapter.
V. Developmental and Crosslinguistic Considerations To address these important issues, we have developed two distinct, but complementary, lines of evidence. First, because most of the research in this area was derived almost exclusively from preschool-aged children (who have already made significant linguistic and conceptual advances), we have begun to chart the emergence of these linkages in prelinguistic infants and in toddlers by examining the influence of language on their categorization abilities. This line of work permits us to ascertain which linkages (if any) guide acquisition from the outset and how these are shaped by experience. Second, because so much of the evidence was originally derived from children acquiring English, we have begun to examine the expectations of children acquiring languages other than English. This crosslinguistic line of work permits us to ascertain which (if any) of these linkages are universal features of human development and which (if any) are acquired via language-specific learning. This wedding of crosslinguistic and developmental evidence remedies several of the shortcomings of earlier work in this area. (Also see Bowerman, 1996 and Lucy, 1996 for fuller discussion.)
A. UNIFYING FEATURES OF THE EXPERIMENTS The experiments that I will describe utilize a wide array of methods and subject populations, but share several important features. Each is essentially an object categorization task, tailored to suit the very different behavioral repertoires of infants versus young children, and of children raised in the United States versus those raised elsewhere. In each series, the goal is to observe the relation between object naming and categorization. To do so, we compare subjects’ categorization of objects in “neutral” conditions (involving no novel words), with their performance when they are introduced to novel words, applied ostensively to the objects under consideration. Because our goal is to examine an abstract linkage between particular syntactic forms and particular types of relations among objects, we introduce novel words (e.g., fauna),as opposed to familiar words (e.g., animal). This insures that the individual words themselves carry no a priori meaning for the child. To examine the influence of syntactic form, we vary the syntactic frame in which the novel word is embedded. We use short, simple syntactic constructions that (1) are typical in infant-directed speech, and that
256
Sandra R Waxman
(2) provide unambiguous contextual evidence that the novel word is either a count noun or an adjective. In the Novel Noun conditions, we introduce objects saying, for example, “This is a blicket” In the Novel Adjective conditions, I use a syntactic frame that is indicative of an adjective: “This is a blick-ish one” (See Gerken & McIntosh, 1993; Waxman & Markow, in press; Waxman, 1995 for evidence regarding infants’ sensitivity to these distinct syntactic frames.) In the No Word control conditions, we introduce no novel words, but point out the objects, saying: “Do you like this?” Performance in this No Word control condition assesses how readily subjects form the various categories presented in our tasks (e.g., dog, animal, pink things). Performance in the Novel Noun condition assesses the role of naming in this important endeavor. Performance in the Novel Adjective condition permits a strong test of the specificity of the relation between object naming and categorization. Because both count nouns (“That is a flamingo”) and adjectives (“That is pink”) can be applied ostensively to objects, this is an important control.
BETWEEN COUNT NOUNSAND ADJECTIVES B. DISTINCTIONS
Despite the fact that both count nouns and adjectives can be applied sensibly to objects, there are several crucial distinctions between these grammatical forms. These distinctions are evident in crosslinguistic,developmental, and semantic analyses. Crosslinguistic analyses have conferred a special status upon the grammatical category noun (Dixon, 1982; Gentner, 1981, 1982; Greenberg, 1963; Macnamara, 1982; Maratsos, 1991; Wierzbicka, 1986). Across languages, this grammatical category includes terms for referring to object categories. Indeed, the universality of this linkage between nouns and object categories has been noted by anthropologists (Berlin, 1973, 1992) and linguists alike (Gleitman, 1990; Grimshaw, 1981; Jackendoff, 1990). In contrast to this crosslinguistic stability of nouns, members of the predicate system (including, e.g., adjectives, prepositions, verbs) have a more fluid status across languages. In comparison to nouns, there is considerably more variation across languages as to what information is conveyed as part of one predicate class as compared to another (Bowerman, 1985,1996; Choi & Bowerman, 1991; Gentner, 1981, 1982; Maratsos, 1991; Maratsos & Chalkley, 1980; Talmy, 1985; Wierzbicka, 1986). In addition, developmental evidence suggests that infants reveal a special talent for mapping words to objects and categories. Longitudinal, crosssectional and crosslinguistic studies have revealed that infants’ early lexicons consist predominantly of nouns-that is, words that are consid-
Linlring Object Categorization and Naming
257
ered nouns in the adult grammar (Au, Dapretto, & Song, 1994; Gentner, 1982; Gentner & Boroditsky, in press; Gleitman, 1990; Goldin-Meadow, Seligman, & Gelman, 1976; Huttenlocher & Smiley, 1987; Macnamara, 1982; Nelson, 1973; Saah et al., 1996). (See Bloom, Tinker & Margulis (1993) and Nelson, Hampson & Shaw (1993) for a different interpretation of the English data; see Choi & Gopnik (1995) and Tardif (1996) for a different interpretation of crosslinguistic data.) In contrast to the developmental and crosslinguistic stability for the grammatical form count noun, there is substantially more variation associated with the form adjective. Although many languages (like English) have a richly developed adjective system, in others (like the Bantu languages) the adjective system is sparse, including as few as 8 to 10 property terms. In such languages, the types of meanings typically conveyed with adjectives in one language are expressed with a different grammatical form in another (Choi & Bowerman, 1991; Dixon, 1982; Talmy, 1985; Wierzbicka, 1986). Developmentally, adjectives tend to be acquired later than nouns (Bloom et al., 1993; Fenson et al., 1994; Gentner, 1981,1982; Maratsos, 1991; Saah et al., 1996). Finally, adjectives are semantically, morphologically, and syntactically dependent upon the nouns they modify (Gleitman, 1990; Hall et al., 1993; Prasada, 1992; Waxman & Markow, in press). These observations accord well with most current theories of language acquisition, which, despite marked differences in theoretical orientation, seem to agree that the grammatical category noun may be acquired earlier than other grammatical categories (including adjective), and that the mappings between nouns and their meanings may be established via different mechanisms than the mappings for other grammatical forms (Gentner, 1982; Gleitman, 1990; Grimshaw, 1994; Maratsos, 1991; Pinker, 1994; Tomasello & Olguin, 1993). Thus, in contrast to the developmental and crosslinguisticstability for the grammatical form noun, there is substantially more variation associated with the form adjective. In addition to these developmental and crosslinguisticdistinctions, a core semantic distinction between count nouns and adjectives has figured largely in psycholinguistictheories: Count nouns (but not adjectives) supply principles of identity and individuation; count nouns also support stronger and more enduring inferences about objects than do adjectives (Gelman & Markman, 1986;Macnamara, 1986; Hall & Waxman, 1993;Markman, 1989; Wierzbicka, 1986; Waxman, Senghas et al., 1997). For example, within any given language, there are words that can be classified as both count nouns and adjectives (e.g., liberal, intellectual, drunk). Although the phrases “She is a liberal” (noun) and “She is liberal” (adjective) can refer to the same individuals, they are not synonymous (Clark, 1997; Putnam, 1975). According to several syntactic- and semantic-
258
Sandra R. Waxman
based accounts, the nouns carry distinctly different and richer meaning than the corresponding adjectives. (See Jesperson, 1968; Lyons, 1977; Wierzbicka, 1986; for theoretical arguments pertaining to this claim.) A count noun (e.g., “She is a liberal,” “This is aflurningo”) (1) identifies an individual within a kind, and (2) provides principles of individuation and identity for that individual, tracing that individual across time and place. In contrast, an adjective (e.g., “She is liberal,” “This is pink”) does not name an individual within a kind; instead, its canonical semantic function is to identify a (possibly transient) property of the named individual (Hall, 1994; Hall & Waxman, 1993; Macnamara, 1986; Wierzbicka, 1986). Adult speakers of English are sensitive to this semantic distinction, drawing deeper inferences about an object when it is labelled with a noun as opposed to an adjective, even when the nouns and adjectives are matched (liberal) (Markman, 1989). By four years of age, English-speaking children are sensitive to this core semantic distinction between count nouns and adjectives (Hall & Moore, 1997). C. A DEVELOPMENTAL, CROSSLINGUISTIC PROPOSAL
Based on this comprehensive review, I proposed that early acquisition is guided by an initial, general linkage between words and object categories. This initial, rudimentary linkage serves three essential functions: (1) it guides infants in their earliest efforts to establish object reference; (2) it promotes the acquisition of a developmentally stable conceptual system for organizing object categories and gaining information about category members; and (3) it sets the stage for the acquisition of more specific expectations linking particular types of words (count nouns versus adjectives) to particular types of relations among objects (object categories versus object properties). I proposed that these more finely tuned linkages between specific syntactic forms (e.g., count nouns, adjectives) and specific types of meaning (e.g., object categories, object properties) would be shaped by language-specificexperience (Waxman & Markow, 1995; Waxman, Senghas et al., 1997). The crosslinguistic stability of count nouns insures that children across the world’s languages will find support in their environment for the specific expectation that count nouns refer to object categories. The crosslinguistic variability of adjectives suggests that children’s expectations for this grammatical category will vary, as children hone in on the particular functions associated with adjectives in their native language. By examining the influence of naming on (a) prelinguistic infants and (b) children acquiring various languages, this research strategy permits us to identify the contributions of language-specific experience.
Linking Objeet Categorization and Naming
259
VI. Empirical Evidence A. DEVELOPMENTAL STUDIES (ENGLISH) 1. Capturing Infants’ Initial Expectations: Do Infants Harbor an Initial, General Expectation that Novel Words Refer to Object Categories?
To address this question, we examined the influence of novel words on object categorization in 12- to 14-month old infants on the brink of language production (see Waxman & Markow, 1995, for a complete description). We developed a novelty-preference task, analogous to those used widely in infancy research. In the familiarization phase, an experimenter offered an infant four different toys from a given category (e.g., four animals) one at a time, in random order. This was immediately followed by a test phase in which the experimenter simultaneously presented both (a) a new member of the now-familiar category (e.g., another animal) and (b) an object from a novel category (e.g., a fruit). Each infant completed this task with four different sets of objects. Two sets involved basic level categories (e.g., cars versus airplanes; horses versus cats); two involved superordinate level categories (e.g., animals versus fruit; tools versus vehicles). Infants manipulated the toys freely throughout this procedure; their manipulation served as the dependent measure. We argued that if infants detect the commonalities among the objects presented during familiarization, then (1) their attention to the objects presented during familiarization should wane, and (2) at test, infants should reveal a preference for the novel object. To test the influence of novel words, we randomly assigned infants to one of three conditions. These differed only in the experimenter’s comments during familiarization. In the No Word condition (control), she said, “See here?” as she introduced each object; in the Novel Noun condition, she said, “See the blicket?” In the Novel Adjective condition, she said, “See the blick-ish one?” In the test phase, infants in all conditions heard precisely the same phrase (“See what I have?”). We reasoned as follows: If novel words direct intants’ attention to object categories, then infants who hear novel words in conjunction with the objects presented during familiarization should be more likely than those in the No Word condition to form object categories. Including both a Novel Noun and Novel Adjective condition permitted us to test the specificity of this initial expectation. If the expectation is general, then infants hearing either novel nouns or adjectives should be more likely than those hearing no novel words to form object categories. The data were entirely consistent with this prediction. Infants hearing either novel nouns or adjectives were more likely to form object categories
Sandra R. Wlurman
260
30
30
m W
25
P z
30
UNOVELNOUN -0- NOVEL ADJECTIVE
+NOVEL NOUN +NOVEL ADJECTIVE -A- NO WORD
15
4'
1
i
NOWORO 15
0
FAMILIARIZATIONTRIAL BASIC LEVEL
1
3
4
FAMILIARIZATIONTRIAL SUPERORDINATELEVEL
Fig. 1. Familiarization phase (Waxman and Markow. 1995).
than were infants in the No Word condition. Interestingly, this influence of novel words was most evident on superordinate level trials. See Figs. 1 and 2. On basic level trials, infants in all conditions successfully formed object categories. However, on superordinate level trials, the facilitative effect of introducing novel words became apparent. Infants in the No Word condition revealed no evidence of object categorization. In contrast, infants who heard either novel nouns or adjectives during familiarization successfully formed superordinate level object categories. This clear pattern indicates that novel words focused infants' attention on commonalities among objects. I have interpreted this striking finding as evidence that words serve as invitations to form categories. Novel words (both adjectives and nouns) direct infants' attention to commonalities among objects, and in this way, 0.70
8
NOVEL NOUN
0'70
NOVEL NOUN
E NOVEL ADJECTIVE
W
0 NO WORD
0.65
E NOVEL ADJECTIVE 0 NO WORD
0.65
w
w z
E
0.60
OAO
a
$
0.55
0.55
0.50
0.55 1
BASIC LEVEL
1 SUPERORDINATELEVEL
Fig. 2. Test phase (Waxman and Markow, 1995).
Linking Object Categorization and Naming
261
facilitate the acquisition of object categories. We have argued that the invitation to form object categories is especially powerful where the commonalities among the objects are not as readily apparent as those at the basic level (Gentner & Waxman, 1994; Waxman & Markow, 1995). In subsequent work, we have considered the specificity of this phenomenon, exploring both (a) the range of signals that can serve as invitations to object categorization and (b) the range of commonalities that can be facilitated by naming. a. Specifying the Range of Signals: Words, Tones, and Gestures It is important to bear in mind that in all of our experiments, we employ the “motherese” speech register to capture all infants’ attention, including those in the No Word control condition. We therefore conclude that infants’ expectations are related to the introduction of the novel words, rather than to the more general arousing, attention-getting effects of infant-directed speech. This well-controlled result fits nicely with the observation that the function of infant-directed speech undergoes a developmental progression “ . . . from the more general affective attentional and affective functions in the early months to linguistic functions toward the end of the first year” (Fernald, 1992, p. 279). But can signals other than words serve as invitations to form object categories? We have considered the influence of various types of signals to answer this question. In one series, we compared the influence of words, content-filtered words, and (pure sine-wave) tones on object categorization at 9 months (Balaban & Waxman, 1996; 1997). We selected 9-month-olds because it is at this developmental moment that infants first reveal evidence of rudimentary word comprehension (Fenson et al., 1994; Jusczyk & A s h , 1995). We found that although infants’ attention was captured equally by both tones and words, only words facilitated object categorization. Tones did not have this facilitative effect. This suggests that the invitation to form object categories cannot be explained as a consequence of a general alerting or attention-engaging function of auditory stimulation (cf., Roberts & Jacob, 1991). Instead, by 9 months of age, infants’ expectations are specifically related to words and to other symbolic forms. Namy (Namy & Waxman, in press, 1997) has also examined infants’ expectations concerning manual gestures as symbols. In her experiments, objects are introduced either with novel gestures or with novel words. Her results reveal that early in acquisition, gestures (like words) facilitate object categorization (Namy & Waxman, in press, 1997). This intriguing result suggests that infants may initially accept a broad range of symbolic forms (including words and gestures, but excluding tones) as names for object categories, but that the range of signals that can serve this important func-
262
Sandra R. Waxman
tion becomes more restricted over development (Waxman & Markow, 1995, in press; Waxman, Stote, & Philippe, 1997). These data also reveal another fundamental ability on the part of very young word learners. Infants notice when novel symbols (words, gestures) are introduced, and they distinguish between phrases in which novel symbols are presented and those containing no novel symbols. b. Specifying the Range of Commonalities among Objects Do words direct attention to commonalities other than those underlying object categories? At issue here is whether infants embark upon the process of acquisition with an expectation linking words (including both count nouns and adjectives) specifically to object categories (e.g., jlamingos), or whether this linkage itself emerges out of a more sweeping expectation linking words to a wider range of commonalities, including perceptual commonalities like color (e.g., pink things) and texture (e.g., bumpy things). If naming initially highlights a wide range of commonalities, then novel words (both count nouns and adjectives) should highlight not only object categories (flamingo, animal), both other commonalities as well, including those that cut across object categories. To address this issue, we adapted the novelty-preference paradigm (Waxman & Markow, 1995), this time examining the influence of novel words on infants’ attention to shared perceptual properties (as opposed to shared category membership). In these studies, infants are familiarized to objects sharing a common salient perceptual property (either color or texture). For example, during familiarization, we present infants with four different purple objects (e.g., a car, a dog, a spoon, and a key) and then examine their novelty preferences at test (e.g., a new purple object versus a blue object). By 12 to 14 months, the invitation to form object categories is not extended to include salient property-based commonalities among objects (Waxman, Philippe, & Branning, 1997). When the objects presented during familiarization shared a common property, rather than category membership, a very different pattern emerged. Infants in the No Word condition revealed a consistent preference for the novel objects at test. This suggests that infants are sensitive to the color- and texture-based commonalities among objects, and do not depend upon a novel word to focus attention on them. Infants in the Novel Adjective condition were also successful in noticing the property-based commonalities. However, performance in the Novel Noun condition was somewhat surprising: Unlike infants in the Novel Adjective and No Word conditions, those hearing novel nouns revealed no novelty preference at test. This outcome is consistent with the possibility that by 14 months, infants in an English-speaking environment have begun to develop a specific expec-
Linking Object Categorization and Naming
263
tation that count nouns can refer to object categories, but not to individual object properties (color, texture). (Note that ours is a receptive measure; using a productive measure, Tomasello & Akhtar (1995) report evidence of a grammatical category noun by 23 months.) There is no doubt that additional work will be necessary before any strong conclusions can be drawn. However, this intriguing outcome suggests that infants’ expectations for novel adjectives may enjoy greater latitude at this point in development than their expectations for novel nouns; adjectives focus attention on object categories (Waxman & Markow, 1995) as well as individual object properties (Klibanoff & Waxman, in press); nouns may be reserved for the types of commonalities underlying object categories. In sum, infants on the threshold of lexical acquisition harbor a broad, initial expectation that words (presented either as count nouns or as adjectives) applied to objects will refer to those individual objects and will be extended to refer to other members of the same object category. This rudimentary linkage is important because it is available early enough to guide infants in their earliest efforts to map words to categories of objects. It also sets the stage for the emergence of a more specific set of expectations regarding particular types of words (e.g., nouns, adjectives, verbs) and particular types of conceptual relations (e.g., object categories, object properties, events). For although infants expect that both nouns or adjectives can refer to object categories, a more specific set of expectations emerges with development. By 2.5 years of age, there is clear evidence that children are sensitive to linguistic form cues, can distinguish novel nouns from adjectives, and assign each distinct types of meaning (e.g., Golinkoff, HirshPasek, Cauley, & Gordon, 1987; Hall et al., 1993; Taylor & Gelman, 1988; Smith et al., 1992; Waxman, 1990; Waxman & Kosowski, 1990). But how do infants accomplish this task?
2. Modification of Initial Expectations: How Do Learners Develop the More Specific Expectations? This is an especially intriguing problem, particularly because count nouns (“This is aflarningo”), adjectives (“This is pink”), and even proper nouns (“This is Alice”) can all be applied ostensively to objects. How, then, do learners gain a toe-hold that permits them to tease apart these syntactic forms and to develop the more specific expectations that (1) count nouns, but not adjectives, refer to object categories, and (2) adjectives, but not count nouns, refer to object properties? To address this question, we designed two independent series of crosssectional experiments, targeting 4 strategic points in language acquisition: 16-month-olds (each had fewer than 50 words in their productive lexicons);
264
Sandra R. Waxman
21-month-olds (each had over 50 words in their productive lexicons); 30month-olds (each had begun to combine words); and 48-month-olds (each had full productive command of various syntactic constructions). We examined how the lexical and syntactic advances that occur at these developmental points were related to the ability to distinguish between novel words presented as nouns versus adjectives. We predicted that throughout this period, (1) count nouns would continue to promote object categorization (e.g., Jlarningos),but would not highlight object properties, and that (2) novel adjectives would no longer promote object categorization, but instead would begin to map specifically to object properties (e.g., pink things, striped things). We tested these predictions with forced choice wordextension tasks. a. Count Nouns and Object Categories In this series, children were introduced to a target (e.g., a dog) and asked to choose between two alternatives: a matching test object (from the same category as the target, e.g., another dog) versus a contrasting test object (from a different category, e.g., a banana). The target and matching test object were members of the same basic level category (e.g., dog) on half of the trials; these were members of the same superordinate category (e.g., animal) on remaining trials. We reasoned as follows: If count nouns draw attention to object categories, then children in the Novel Noun condition should be more likely than those in a No Word control condition to select the matching (same-category) test object. When this effect becomes specific to novel nouns, then children hearing novel adjectives should reveal no such preference. As predicted, (1) count nouns exerted a uniform influence across this period. At all ages, children consistently extended nouns to the same-category alternative, (2) a clear distinction between count nouns and adjectives emerged during this period. By 21 months, the tendency to extend adjectives to object categories began to diminish markedly. This distinction between novel nouns versus adjectives in object categorization became more entrained with age (Waxman, 1995; Waxman, Stote et at., 1997). b. Adjectives and Object Properties This series of experiments represented a subtle, but important, change in focus. Rather than using adjectives as a control (to test the specificity of the noun-category linkage), we sought a more positive characterization of the influence of adjectives. Comparatively little research has been devoted to the acquisition of words other than count nouns. Although research on verb acquisition has begun to pick up pace (Behrend, 1990; Naigles, 1996; Fisher, Hall, Rakowitz, & Gleitman, 1994; Hirsh-Pasek, Golinkoff, & Naigles, 1996; Gleitman, 1990; Tardif, Shatz, & Naigles, in press; Tomasello, 1992), research on adjectives has been sparse. However, because both nouns and adjectives (but not verbs)
Linking Object Categorization and Naming
265
can be applied ostensively to objects, it is important to discover how learners begin to distinguish these forms. Moreover, the crosslinguistic and developmental variability associated with adjectives (like verbs) provides a fascinating opportunity to observe the rich interplay between expectations within the child and the shaping role of language input. We examined the emerging expectation that an adjective, applied ostensively to an object (“That one is yellow”), will refer to an object property (e.g., its color, texture) rather than to the object category itself (e.g., apple). Note that this question hinges on a distinction between object categories (e.g., flamingos, animals) and object properties (e.g., pink things, striped things). Most current accounts distinguish object categories (also known as kinds or sortals) from other groupings on at least three grounds: Object categories (1)are richly structured, (2) capture many commonalities, including deep, nonobvious relations among properties (as opposed to isolated features), and (3) serve as the basis for induction (Barsalou, 1983; Gelman, 1996; Gelman & Medin, 1993; Gelman, Coley, & Gottfried, 1994; Kalish & Gelman, 1992; Keill994; Landau, Smith, & Jones, 1988; Macnamara, 1994; Markman, 1989; Medin & Heit, in press; Murphy & Medin, 1985; Rips, 1989; Waxman, Lynch, Casey, & Baer, 1997). See Gelman (1996) and Keil (1994) for evidence that although children may lack detailed knowledge about particular object categories, they nonetheless expect object categories to serve these functions. See Bhatt and Rovee-Collier (1996) and Younger and Cohen (1986) for evidence of a psychological distinction between individual properties and relations among properties in infants. In a forced-choice task, infants were introduced to a familiar target (e.g., a yellow object) and asked to choose between two test objects. However, in this series, the matching test object shared a salient property with the target (e.g., another yellow object); the contrasting test object contrasted with the target along that property dimension (e.g., a green object). We examined 21-month-olds, using a Novel Adjective, Novel Noun, or No Word condition. We reasoned: If adjectives draw attention to object properties, then infants hearing novel adjectives should be more likely than those in a No Word control to select the matching-property test object. If this effect is specific to adjectives, infants hearing novel nouns should reveal no preference. We also examined another factor: the range of application for novel adjectives. For adults, adjectives can identify a salient property of an individual (e.g., a spotted chair); they can be extended across individuals within a given category (e.g., the spotted dogs); and they can be extended across different categories (e.g., the spotted things, including dogs and cups). But the precise meaning for most adjectives is influenced by the noun they modify: Soft slippers and soft ice cream do not have the same texture; an
Sandra R. Waxman
266
expensive tie and an expensive house do not have the same cost. This is because most adjectives do not indicate absolute measures, but instead indicate a relative point along a continuum. The range of this continuum is delimited by the category itself. This semantic dependency of adjectives on nouns has been observed across languages. Syntactic and morphologic dependencies have been documented as well. For example, in languages that mark grammatical gender or number, adjectives must accord with the basic level nouns they modify. These observations suggest that there may be a linguistic or conceptual priority for establishing an object's basic level kind before marking its properties. If children are sensitive to this dependency, then they should succeed in extending property terms (e.g., spotted, applied to a spotted dog) to other objects from the same basic level category (e.g., other spotted dogs), but should fail to extend property terms to objects from different basic level categories (e.g., other spotted objects, including fish and cups). To test this hypothesis, children at each age were randomly assigned to either a Within Basic level or to a Across Basic level condition (see Fig. 3). The results were straightforward. First, by 21 months of age, only children in the Novel Adjective condition consistently selected the matching-property test objects; those in the No Word and Novel Noun conditions did not select the property match. By 21 months, then, infants expect that adjectives, but not nouns, refer to object properties; they expect that nouns, but not adjectives, refer to object categories (cf., Waxman & Hall, 1993; Waxman, 1995; Waxman & Markow, in press). Second, infants were sensitive to the range of application. They successfully mapped novel adjectives to object properties only when the target and test objects were all drawn from the same familiar basic level kind (e.g., all dogs). In sharp contrast, when the target (e.g., a fish) and test objects (e.g., dogs) were drawn from different basic levels, infants failed to extend adjectives systematically; they performed at chance.
Dxsi WilhinBasic
&
slairT-
Malching Objecl Contrasting Object
k3,',
Across Basic i
Fig. 3. One representative stimulus set (from Waxman & Markow, in press; Klibanoff and Waxman, in press).
Linking Object Categorization and Naming
267
Somewhat surprisingly, we have found that this pattern persists into the preschool years (Klibanoff & Waxman, in press). See Fig. 4.Three-yearolds successfully extended novel adjectives to object properties only if the objects were all drawn from the same familiar basic level category; they failed to extend novel adjectives systematicallywhen the objects were drawn from different basic level categories. This difficulty was evident even when the objects represented different basic level categories within the same ontological kind (e.g., animate objects: dogs and snakes). By 4 years, children map novel adjectives to object properties even when objects are drawn from different basic level categories. These studies are important for two reasons. First, they offer the earliest evidence of an ability to (a) distinguish the grammatical form noun from adjective and (b) assign each distinct types of meaning. Previous reports revealed distinct patterns for these two grammatical forms at 2.5 years (Hall et al., 1993; Taylor & Gelman, 1988; Smith et al., 1992; Waxman, 1990,1995). Second, infants’ early expectations for novel adjectives appears to unfold within the support of a familiar basic level kind. This is consistent with arguments for the priority of establishing and naming basic level object kinds (Hall et al., 1993; Macnamara, 1986). The results from this series suggest that basic level distinctions, rather than global or ontological distinctions (e.g., animate versus inanimate object) may serve as the entry point
..
1.01
ADJECTIVE ADJECTIVE No4 years 3 years 4 years
**
E
1
NoVKFiD 3 years
differs from chance, p c .01
Fig. 4. Three- and four-year-olds’consistent property choices as a function of word and level (Klibanoff and Waxman, in press).
268
Sandra
R. Waxman
for working out the further semantic and syntactic distinctions between count nouns and adjectives. This is consistent with documentation of the dependency of adjectives on basic level count nouns-a dependency that has been observed across languages in morphological, semantic, syntactic, and lexical analyses. Interestingly, it turns out that most of the developmental research on the acquisition of adjectives has capitalized on conditions in which the adjective picks out a property within, but not across, basic level kinds (cf., Au & Markman, 1987;Gelman & Markman, 1985;Hall et al., 1993;Prasada, 1992; Taylor & Gelman, 1988; Waxman, 1990). In our current work, we have begun to pursue this notion of dependency vigorously, examining the factors underlying the foundational role of the basic level, and the types of experience that permit children to advance beyond this initial entry point in interpreting novel adjectives. In sum, these developmental studies have documented an initially general linkage between words and object categories that gives way to a more specific set of expectations regarding the particular types of referring associated with particular syntactic categories in English. The results suggest that count nouns exert a continuous influence from infancy through the preschool years, directing attention to the commonalities underlying object categories, particularly those at the so-called basic level. The expectations for adjectives vary considerably across this period. This variability in the adjective system provides us with an exciting opportunity to observe the developmental change that occurs as children hone in on the specific meanings associated with adjectives in their native language.
B. CROSSLINGUISTIC EVIDENCE Crosslinguistic evidence is essential in piecing together the origin of these linkages, determining which links might be universal, and discovering how these are shaped by the language environment. We therefore buttress the developmental work with English-speaking children by examining the influence of novel words (nouns or adjectives) on object categorization monolingual children acquiring languages other than English, French, or Spanish as their native language (Waxman, Senghas et al., 1997). 1. Differences in Status of Adjectives across the Target Languages
We began our crosslinguistic work with a detailed examination of the expectations of children acquiring either French or Spanish. These IndoEuropean languages are closely related, but they provide an interesting crosslinguisticcontrast, primarily because of differences in the grammatical
Linking Object Categorization and Naming
269
use and referential status associated with the grammatical form adjective. To get a sense of this difference, consider, for example, a cupboard filled with several cups. In English, we distinguish these linguistically by modifying a noun with an adjective (e.g., “the blue cup”). In Spanish, however, the head noun is omitted obligatorily from the surface of the sentence, leaving the determiner and adjective alone (e.g., “el azul” or “the blue”). Constructions like this are known as determiner-adjective (det+A) phrases. Two features of det-A phrases are especially relevant to the issues at hand. Notice, first, that in det+A phrases, adjectives appear in syntactic contexts that are identical to those for count nouns. In Spanish, det+A constructions are ubiquitous; they are spontaneously productive in children as young as 2 years. In French and English, det+As are much less common; they are restricted to a much narrower set of properties (e.g., color, age). (See Waxman, Senghas, et al., 1997 for a more detailed discussion). As a consequence, in Spanish adjectives often appear in syntactic constructions that are identical to those for count nouns. Second, adjectives in det+A constructions are extended to include other members of the kinds denoted by a salient property. In Spanish, then, there is considerable overlap in both the syntactic contexts and the extensions for count nouns and adjectives. (See Gathercole, 1997 for other semantic/syntactic differences between Spanish and English, and their relation to acquisition.) We suspected that experience with these structurally different native languages would lead to different outcomes in the expectations concerning the grammatical form adjective. Children acquiring French (like those acquiring English) should learn that adjectives do not, as a rule, refer to categories of objects. However, children acquiring Spanish may learn that novel adjectives, like nouns, can indeed be extended to refer to objects and categories of objects. If this is the case, then we would expect crosslinguistic consistency in children’s expectations for novel nouns; in contrast, their extensions of novel adjectives should vary as a function of the structure of the language under acquisition.
2. Consequences on Children’s Expectations
To test these predictions, we adapted a very robust forced-choice procedure to extend our examinations of the influence of novel nouns and adjectives on the formation of object categories to include monolingual children acquiring either French or Spanish (Waxman & Kosowski, 1990; Waxman, Senghas et al., 1997). Each child sat individually with an experimenter to “read” through a picture book that we had created. Each page of the book depicted five different objects, including a target (e.g., a cow), two objects from the same superordinate category as the target (e.g., a fox and a zebra), and
270
Sandra R. Waxman
two objects that were thematically related to the target (e.g., a barn and milk). Children were assigned randomly to conditions that differed only in the way the experimenter introduced the target object. In the No Word (control) condition, the experimenter pointed to the target and said, “See this? Can you find another one?” In the Novel Noun condition, she said, for example, “See this ‘blicket’? Can you find another ‘blicket’?” In the Novel Adjective condition, she said, for example, “See this ‘blick-ish’one? Can you show me another one that is ‘blick-ish’?” (We examined Spanishspeaking children’s performance with novel adjectives (a) when these were presented in conjunction with a head noun, as above, and (b) when they were presented within det-A phrases.) The child and experimenter went through the book two times. On the second reading, the experimenter reminded the children of their first choices and asked them to select another from the remaining (3) alternatives. The results of these experiments were entirely consistent with our proposal. Performance in the Novel Noun condition was uniform across all three languages. Children extended novel nouns taxonomically, selecting objects from the same superordinate category as the target. Performance in the No Word control condition was also uniform across languages, suggesting that the materials were not biased in any way in any language. Children in this control condition revealed no preference for either the taxonomic or thematic alternatives. However, as predicted, performance in the Novel Adjective condition varied systematically as a function of the language under acquisition. Children acquiring French (like those acquiring English) performed at chance in the Novel Adjective condition, revealing no preference for either the taxonomic or thematic alternatives. In contrast, Spanish-speaking children displayed a strong inclination to extend novel adjectives (like a novel nouns) to other members of the same superordinate level object category. This tendency was apparent whether the novel adjectives were presented in conjunction with an overt noun or within det+A phrases. This pattern, which has now been replicated in four different studies, suggests that Spanish-speaking children have learned that both count nouns and adjectives can be extended taxonomically, to include the named object and other members of an object category. We currently are extending this crosslinguistic work in two directions. First, we are seeking replication in two additional languages (Swedish and Italian) that share important structural features with Spanish: both are prodrop languages, in which det+A phrases appear fluidly across property types. These (predicted) replications will constitute strong converging evidence that (1) the noun-category linkage is robust across languages, and that (2) experience with the language system shapes the more specific
Linking Object Categorization and Naming
27 1
linkages for adjectives. This outcome would also serve as a firm foundation for considering a wider range of languages and language-types. A second goal is to deepen the analysis of crosslinguistic differences in the adjective system, going beyond a word’s extension alone to focus more specificallyon the semantic distinctions between count nouns and adjectives. We predict that despite crosslinguistic differences in the extension of novel adjectives in Spanish, as compared to French and English, the core semantic distinction between the grammatical categories noun and adjective are as distinct in one language (e.g., Spanish) as another (e.g., French, English). Across languages, then, (1) count nouns (but not adjectives) should supply principles of object identity and individuation; (2) count nouns should also support stronger and more enduring inferences about objects than do adjectives (Gelman & Markman, 1986; Macnamara, 1986; Hall & Waxman, 1993; Markman, 1989; Wierzbicka, 1986; Waxman, Senghas et al., 1997). We have now demonstrated, following Hall and Moore (1997), that 4year-old French-speakers, like their English-speaking counterparts, expect that count nouns support more enduring inferences about objects than do adjectives (Waxman, Bredart, Nicolay, & Hall, 1998). Evidence from speakers of Spanish will provide a fascinating test case for this semantic distinction. Despite the extensional overlap between count nouns and adjectives (Waxman, Senghas et al., 1997), Spanish-speakers may nonetheless be sensitive to the (possibly universal) underlying semantic distinction between these forms, with count nouns providing richer and more enduring information about individuals than adjectives. VII. Integrating the Developmental and Crosslinguistic Evidence Together, the developmental and crosslinguistic findings offer important insights into the relation between object categorization and naming across languages and across development. From the earliest stages of word learning, infants interpret words (from various grammatical categories, including both count nouns and adjectives), applied to solid objects, as referring to those objects and to other members of the same kind. (See Echols, 1992, for evidence that verbs, too, may initially direct infants’ attention to object categories.) This initial expectation is itself fine-tuned as a result of the child’s experience with the native language under acquisition. For infants on the brink of word learning, novel words (both count nouns and adjectives) highlight commonalities among objects and, in this way, foster the formation of object categories. This initial expectation is important in several respects. First, because it is evident as early as 9 to 12 months
212
Sandra R. Waxman
of age, it is available to guide infants in their earliest efforts to link words to their meaning, In particular, it supports the early establishment of object reference and promotes the early establishment of a repertoire of object categories. These object categories will be stable over development and will support the acquisition of additional information about category members. This strong empirical finding challenges directly the claim that a linkage between object categorization and naming is unavailable at the onset of lexical acquisition, that it is learned or constructed as a consequence of word learning (Bloom et al., 1993; Nelson, 1988; Smith, 1995; Xu & Carey, 1996). Instead, the evidence indicates that this general expectation fuels the acquisition of object naming and categorization from the start (also see Waxman & Hall, 1993). This finding also suggests that although infants’ concepts are certainly less elaborate, and less richly structured than those of older children and adults, there is nonetheless considerable conceptual continuity. In contrast, Xu and Carey (1996) have argued recently for radical conceptual change in infancy. They have suggested that before the onset of lexical acquisition, infants are unable to represent any concepts (sortals) that are more specific that the sortal object (but see Wilcox & Baillargeon, in press). Our results suggest a more continuous developmental trajectory: For prelinguistic infants (at 9 months of age), words focus attention on basic level object concepts, the very types of object concepts that figure prominently in the conceptual systems of older infants and adults. Interestingly, Xu (1997) has recently presented evidence that appears to concur with this conclusion. One advantage of infants’ initially general expectation is that although it offers a guide in the early establishment of reference, particularly for object kinds, it does not require infants on the brink of learning language to be able to identify the relevant surface cues that distinguish among the particular grammatical categories (e.g., nouns, adjectives) in their native language (Pinker, 1984; Gleitman, 1990; Gerken & McIntosh, 1993). This is important, because grammatical categories are not marked transparently or universally in the input children receive. There appear to be no universally constant stress levels, affixes, or positions in a sentence that are associated with any of the grammatical categories (Pinker, 1984). Indeed, in most current theories of language acquisition, the very ability to identify instances of the various grammatical categories is itself a major achievement. In fact, our proposal is consistent with the speculation that the ability to recognize the grammatical categories and establish their meaning may be dependent upon the prior establishment of object reference (Gentner, 1982; Gleitman, 1990; Hall et al., 1993), particularly reference to basic level kinds (Hall & Waxman, 1993; Waxman & Markow, 1995, in press).
Linking Object Categorization and Naming
273
Another advantage is that this perspective is flexible enough to account for the fact that infants readily acquire languages that differ in the ways in which they recruit particular grammatical categories to convey particular types of meaning (sometimes known as the conflation problem: Bowerman, 1996; Naigles & Eisenberg, in press). We suspect that it is to infants’ advantage to begin with an initially general expectation-an expectation that will guide them in establishing their first word-meaning mappings and that can then be tailored to fit the particular variations (or conflations) encountered in their native language. We have also demonstrated that infants’ broad initial expectations are subsequently shaped by the structure of the native language under acquisition, and become more entrained with age and experience. Once infants have acquired a stable lexical repertoire, they begin to show consistent evidence of a more specific set of expectations linking particular types of words with particular types of relations among objects. Based upon a crosslinguistic analysis of the nouns and adjectives, we predicted that an expectation that count nouns refer to object categories would persist across development and be uniform across languages. As predicted, we discovered that the nouncategory linkage was evident in French- and Spanish-speaking children, just as it has been evident in English-speaking children, and in infants immersed in an English-speaking environment (Markman & Hutchinson, 1984; Balaban & Waxman, 1996; Waxman & Hall, 1993; Waxman & Markow, 1995; Waxman, Senghas et al., 1997). We also predicted that the specific mappings for adjectives would vary according to the structure of the native language under acquisition. As predicted, children’s extensions of novel adjectives did change across development (Waxman & Markow, 1995, in press) and across languages (Waxman, Senghas et al., 1997). We also discovered that the mappings between adjectives and object properties emerge within the support of a familiar basic level kind. The results of this program of research raise several intriguing issues. In the following sections, I touch upon these and discuss their consequences for a comprehensive and integrative theory of development. A. MECHANISMS UNDERLYING THE TRAJECTORY FROM A GENERAL TO A MORESPECIFIC SETOF EXPECTATIONS One central issue concerns the mechanisms underlying this proposed developmental trajectory. That is, how does an initially general expectation, in which infants treat nouns and adjectives identically with respect to object categorization, give way to the more specific expectations that characterize older children and adults? This issue touches upon both linguistic and conceptual factors. We consider several possible mechanisms by which this
274
Sandra R. Waxman
evolution may come about. One possibility is that infants begin the process of lexical acquisition with a truly general expectation that words refer to commonalities among objects, including those underlying object categories, as we have proposed. Once they establish object reference and a stable set of object categories, they may be able (a) to identify more subtle regularities in the linguistic input that permit them to tease apart the different grammatical forms (e.g., count nouns versus adjectives) and (b) to discover that these tend to signal (or correlate with) particular kinds of relations among objects (e.g., object categories versus object properties). On this view, infants’ early referential abilities serve as the essential bedrock for the subsequent acquisition of more specific expectations. A second possibility grants infants a somewhat more “advanced” outlook, in which they share with older children and adults the expectations that (a) their language will include distinct grammatical categories and (b) these are linked to distinct types of conceptual relations (c.f., Gleitman, 1990; Grimshaw, 1981; Pinker, 1984). However, either because infants have not yet learned the relevant surface cues to distinguish among these grammatical forms, or because processing limitations prevent them from perceiving the differences between the syntactic frames surrounding novel words, infants initially fail to distinguish novel words presented as adjectives versus nouns (c.f., Grimshaw, 1981; Pinker, 1984). A third possibility is that infants’ initially general expectation is primarily a consequence of their very limited lexical knowledge. Perhaps infants (a) expect to find distinct grammatical categories, (b) are able to distinguish among these grammatical forms, and (c) expect that these are linked to distinct types of conceptual relations. On this view, infants’ failure to reveal this specific expectation is related to their limited repertoire of basic level names. Recall that the use of grammatical form as a cue to meaning is dependent upon mastery of a basic level name for the objects under consideration (Hall, 1993; Hall et al., 1993; Macnamara, 1982). When a novel word is applied to an object for which children have no basic level name, they tend to extend any word (either a count noun or an adjective) to other members of the same object category. This pattern is evident in children as old as 4 years of age, children who clearly hold specific expectations linking particular types of words to particular types of conceptual relations. This is relevant because, by definition, infants on the brink of lexical acquisition are unfamiliar with most words, including the names for most basic level categories. Therefore, infants’ initially general tendency to extend novel adjectives, like nouns, to object categories may be a consequence of their limited repertoire of basic level names. Each of the possibilities outlined above reflects different theoretical commitment and a different account of the mechanism underlying the trajectory
L d n g Object Categorization and Naming
275
from an initially general to a more specific set of expectations between words and object categories. However, these possibilities are difficult to disentangle empirically, because each is consistent with our demonstration that infants’ first expectations for mapping words is general and that more specific linkages emerge later. Each possibility also predicts that object categories are prime candidates for word meaning early in acquisition. B. WHATIs THE RELATION BETWEEN OBJECT CATEGORIES FORMED EARLY IN DEVELOPMENT AND THOSE CHARACTERISTICS OF A MOREMATURE SYSTEM? I have just discussed some of the mechanisms by which infants make advances on the linguistic side of the relation between words and object categories. Here, I discuss some of the advances that I expect come about on the conceptual side. Recent research has revealed that, under certain circumstances, infants and young children will group together the same sets of objects as do older children and adults. However, this does not constitute evidence that infants’ conceptualization of various object categories is on a par with that of older children and adults. On the contrary, infants’ conceptualizations of object categories (e.g., bear, animal) lacks the elaborate information, the richly interconnected theories, and the inductive depth that are available to older children and adults (Carey, 1975; Gelman & Coley, 1990;Gentner & Waxman, 1994;Keil, 1987;Murphy & Medin, 1985). Although it is beyond the scope of this chapter to describe how children come to acquire these deeper, more elaborate, more richly structured conceptual systems (for discussions, see Gelman, 1996;Keil, 1994), it is important to consider here the role of naming in this process (also see Gentner & Waxman, 1994). I have proposed that words focus infants’ attention on commonalities among objects and in so doing, facilitate the establishment of object categories. Words, particularly count nouns, exert a similarly dramatic effect in preschool-aged children (Markman, 1989;Ward, Becker, Hass, & Vela, 1991; Waxman, 1990,1991,1994).Providing a common label for a set of objects initiates a search for coherence among them. This search may be unsuccessful. For example, novel nouns do not influence infants’ performance when sets of objects share no discernable perceptual or conceptual commonalities (Waxman & Markow, 1995). However, when a commonality can be discerned, words serve as invitations to form categories. This invitation has several dramatic consequences. First, novel words invite infants to assemble together objects that might otherwise be perceived as disparate entities. In this way, words promote comparison among them. This process of comparison may lead infants to discover other commonalities that might otherwise have gone unnoticed (Gentner & Waxman, 1994; Gentner & Namy, 1998).
276
Sandra R. Waxman
Naming may also have dramatic consequences in situations in which infants have already formed groupings and noticed (some of) the commonalities among objects. For example, although 12-month-old infants successfully formed basic level object categories (whether or not they were introduced to novel words), their knowledge about these categories is not on a par with the knowledge of an older child or adult. Even preschool-aged children lack detailed knowledge about most categories (Gelman, 1996; Keil, 1994). Nonetheless, despite their relative lack of information, children seem to expect that members of object categories share deep, nonobvious commonalities. Indeed, children depend upon these to support induction. I suspect that novel words are instrumental in motivating infants and young children to discover the deeper commonalities that underlie our richly structured object categories (Barsalou, 1983; Gelman, 1996; Gelman & Medin, 1993; Gelman et al., 1994; Kalish & Gelman, 1992; Keil, 1994; Landau, Smith, & Jones, 1988; Lassaline & Murphy, 1996; Macnamara, 1994; Markman, 1989; Medin & Heit, in press; Murphy & Medin, 1985; Landau, 1994; Rips, 1989; Ross, 1996; Waxman & Markow, 1995). Finally, I speculate that expectation that count nouns refer to object categories plays a central role in the acquisition of count nouns referring to a wide range of concepts, including abstract (e.g., idea, dream) and relational (e.g., kinship terms) concepts. Although infants may initially depend upon some perceptual support to form categories, I suspect that they can then extend their expectation more broadly. In this way, an expectation that count nouns refer to object categories can itself support the acquisition of abstract and relational concepts, particularly when these concepts are named with count nouns. C. CONTINUITY IN DEVELOPMENT: PERCEPTUAL AND CONCEPTUAL TO CATEGOR~ZATION AND NAMING CONTR~BUTIONS
I have argued for developmental continuity in (a) the types of concepts entertained by infants, young children, and adults, and (b) the contribution of both perceptual and conceptual factors in categorization and naming. I have proposed that words are powerful engines, advancing us beyond our initial conceptual and perceptual groupings, and fueling the acquisition of the essential, rich commonalities and relations that characterize our most powerful concepts.
D. THESCOPE OF INFANTS’EARLY OBJECT CATEGORIES AND THE ROLEOF NAMING IN THEIR ESTABLISHMENT
Two other developmental proposals have been offered recently. Each is predicated on a very different set of assumptions about the scope of infants’ early object categories and the role of naming in their establishment.
Linkiog Object Categorization and Naming
277
1. On the Acquisition of Global versus Basic Level Object Categories
Although there is strong evidence that basic-level object categories emerge very early in development (Brown, 1958; Fenson, Cameron, & Kennedy, 1988; Mervis & Crisafi, 1982; Quinn, Eimas, & Rosenkrantz, 1993; Roberts, 1988 Roberts & Cuff, 1989;Roberts & Horowitz, 1986;Rosch, 1978;Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976; Waxman, 1990), this view has recently been challenged. Mandler and her colleagues (Mandler, 1988; 1992; Mandler & Bauer, 1988;Mandler, Bauer, & McDonough, 1991; Mandler & McDonough, 1993) have argued (1) that conceptual development begins at a more abstract, global (e.g., animate versus inanimate objects) level, (2) that the acquisition of these global conceptsprecedes the organization of concepts at basic levels, and (3) that the basic-level groupings of infants and toddlers are entirely perceptually based, lacking in any conceptual grounding or inductive strength (Mandler & McDonough, 1996). In contrast, we have argued that basic-level object categories serve as an important entry point in object categorization, naming, and inductive inference. Although we acknowledge that basic-level object categories certainly enjoy considerable perceptual support, this does not, in itself, mean that they lack conceptual force. On the contrary, like most cognitive scientists (coming from a wide range of disciplines and perspectives), we are committed to the idea that perceptual and conceptual information are not mutually exclusive. Indeed, to the best of our knowledge, there is no strong basis to the claim that in object categorization, perceptual and conceptual information or processes are frankly distinct, either in infancy or in adulthood. We take no issue with the possibility that infants acquire both global(e.g., animate versus inanimate objects; land versus sea animals) and basiclevel conceptual categories. The suggestion that infants can, under certain circumstances, form object categories at various levels of abstraction is attractive. It incorporates the view that infants’ attend to several different kinds of perceptual and conceptual features in categorization, including an object’s form, function, and type of movement (Baldwin, Markman, & Melartin, 1993; Kemler-Nelson, 1995; Smith & Heise, 1992). The early acquisition of global, as well as basic, categories also ensures considerable continuity in conceptual development. However, on both logical and empirical grounds, we strongly doubt that the acquisition of global concepts precedes the acquisition of basic-level concepts, with the latter being strictly derivative of the former. We also doubt that there is a frank dissociation between perceptual and conceptual factors, in which infants’ basic-level object categories are entirely perceptu-
218
Sandra R. Waxman
ally based, whereas their global categories are conceptual. These doubts come from several different sources. First, there are perceptual, as well as conceptual distinctions underlying global categories. For example, infants are sensitive to the perceptual distinction between the types of motion displayed by animate versus inanimate objects early in development (Bertenthal, Proffitt, Kramer, & Spetner, 1987). Second, the argument for the precedence of global- over basic-level categories in infancy is not supported strongly by the data. To take one example, Mandler and McDonough (1993) used a novelty-preference task that was similar in many respects to the one we have discussed here (Waxman & Markow, 1995). They report that 11-month-old infants successfully formed one global-level distinction (animals versus vehicles), but failed to reveal a consistent pattern of performance at the basic level. However, a careful analysis indicates that these infants reliably formed one set of basiclevel categories (car versus plane), but failed to form another (dog versus fish). This reveals an effect of stimulus set at the basic level, an effect that may be attributed to several different factors. The contradictory pattern observed at the basic level also raises questions about the generalizability of the phenomenon. Unfortunately, this set effect was not pursued, leaving only a single test of categorization at the global level and contradictory results from the two tests of categorization at the basic level. This pattern provides insufficient empirical support for the precedence of global concepts. Infants’ performance in the novelty-preference tasks presented in our lab also raise doubts (Waxman & Markow, 1995). If Mandler is correct, then infants in the No Word conditions should have formed object categories at the more abstract superordinate level quite successfully, for the conceptual commonalities among these objects (e.g., animate versus inanimate objects) should have been readily available. However, this was not the case: although infants in the No Word conditions formed basic-levelcategories successfully, they evidenced no appreciation of these more abstract categories. The contrast between performance with and without novel words suggests that infants in this task depended upon the ihfluence of a novel word to successfully form these more abstract object categories. This pattern underscores the facility with which infants form (or recognize) commonalities at the basic level, as opposed to those at a more abstract global level. Leaving methodological issues aside (see Waxman & Hall, 1993; Waxman & Markow, 1995), consider the evidence from analyses of early lexical acquisition. Early in lexical development, words for basic-level categories are very common; indeed words falling into this category are the predominant form (Nelson, 1973; Nelson, Hampson, & Shaw, 1993; Gentner, 1982; Gentner & Boroditsky, 1997; Saah et al., 1996). In contrast, words for global categories are extremely rare (Fenson et al., 1994). As we have discussed,
Linking Object Categorization and Naming
279
infants’ and young children’s tendency to extend novel words, applied to novel objects, to other members of the same basic level object category is so strong that it overrides children’s use of syntactic form as a cue to word meaning (Hall et al., 1993; Hall & Waxman, 1993; Au & Markman, 1987). This empirical result is also consistent with anecdotal evidence that infants and toddlers often interpret a novel word (e.g., hot) applied to a novel object (e.g., a stove or pot) as referring to a whole object, and not to an (intended) property. Recall also that infants’ and young children’s extension of novel adjectives appeared to depend upon the support of a familiar basic-level category. When objects were drawn from the same basic-level category, they mapped novel adjectives successfully to object properties such as color and texture. Strikingly, global superordinate level categories did not support such extensions (Klibanoff & Waxman, 1997; Waxman & Markow, in press). We therefore maintain that the acquisition of basic-level object categories and their names serve as a foundation for the acquisition of other types of words referring to other types of relations among objects. How can Mandler’s position be reconciled with these arguments supporting the central conceptual role of basic-levelcategories and the early acquisition of their names? One possible reconciliation is to argue that infants map their first words to perceptual, rather than to conceptual, groupings. However, this reconciliation is incomplete. First, it is unclear why abstract symbols (words) would be tethered so tightly to perceptual groupings (basiclevel categories, in Mandler’s view), but not extended to groupings with an abstract conceptual base (global categories, in Mandler’s view). Second, even if one accepted the assumption that basic-level categories are purely perceptually based, it is unclear why infants would reveal such a strong talent for mapping words to some perceptual groupings (namely, those at the basic level), but not to other perceptual groupings, including words referring to color and texture. Names for perceptual properties like these enter the lexicon much later than names for basic-level object categories. Indeed, the acquisition of these terms seems to depend upon the prior acquisition of basic-level categories and their names (Waxman & Markow, in press; Klibanoff & Waxman, in press; Hall et al., 1993). These observations, coupled with the arguments presented throughout this chapter, provide clear evidence for the importance of basic-level categories, their names, and their conceptual status, early in development. Our account also embraces a continuous view of development, in which perceptual and conceptual factors are recruited to support object categorization and naming throughout the course of development.
2. Object Categories vs. Object Shape There has also been recent debate concerning the types of meaning that children (and adults) associate with novel count nouns. Simply stated, this
280
Sandra R. Waxman
debate centers around whether count nouns direct word learners’ attention to categories of objects (henceforth, the noun-category position: e.g., Gelman & Medin, 1993; Markman, 1994;Soja et al., 1991,1992;Waxman, 1990, 1994) or to shapes of objects (henceforth, the noun-shape position: e.g., Landau, Jones, & Smith, 1992). Because it is not possible to do justice here to the complexity of this issue, I focus principally on three fundamental points: (1) the role ascribed to development, (2) the issue of generalizability, and (3) the underlying model of word extension. I begin by noting several key points of convergence between the nouncategory and noun-shape positions. Both illustrate the powerful role of count nouns in object categorization. Both assume that perceptual properties can strongly influence our judgments of category membership and noun extension. Both assume that an object’s shape and category membership often covary, particularly when considering object categories at the basic level. However, these positions differ in the role they ascribe to development. According to the noun-shape position, children (sometime after age 2, or perhaps even earlier) develop an expectation that count nouns refer to objects with the same shape (Landau, Smith, & Jones, 1988; Landau, Jones, & Smith, 1992; Smith, Jones, & Landau, 1992). This could be a powerful expectation, particularly early in development, when children may rely rather heavily on perceptual information (Imai, Gentner, & Uchida, 1994). Yet shape-based similarities do not represent the full extent of adults’ or children’s judgments regarding either category membership or noun extension. On the contrary, children and adults expect that two objects that look alike (say, by virtue of shape) will also share other, perhaps deeper, nonperceptual commonalities as well (Baldwin et al., 1993;Gelman, 1996; Gentner & Namy, 1998). In addition, if two objects share a common label, then children expect that these objects also share other nonperceptual commonalities, even when the objects are perceptually dissimilar (Gelman, 1988). Thus, there is strong evidence that for both adults and children, neither the extension nor the intension of novel nouns rests upon shape-based similarities alone (Gelman & Medin, 1993; Gentner & Waxman, 1994; Soja et al., 1991, 1992; Waxman & Braig, 1996; Waxman & Markow, 1995). However, the noun-shape account cannot specify how the learner moves from primarily shape-based extensions to those that capture additional, perhaps deeper, properties. Because an object’s shape is readily available from even a cursory inspection of an object, the noun-shape account includes no mechanism to look further for additional properties that bind a particular set together and serve as the basis for noun’s extension. This mechanism appears to be beyond the scope of the noun-shape position.
Linking Objed Categorization and Naming
281
In contrast, the noun-category position ascribes a more central role to development, with words serving as a catalyst for object categorization and for change. On this account, grouping objects together on the basis of shape can serve as one entry point (and perhaps an especially important entry point for simple artifacts in particular, as discussed below). Naming promotes the discovery of additional, perhaps deeper commonalities than those that might have formed the initial basis of the word’s extension of the word. Therefore, the noun-category position incorporates a developmental mechanism that motivates learners to discover these powerful, additional commonalities among objects (Gelman & Medin, 1993; Gentner & Waxman, 1994; Waxman & Braig, 1996; Waxman & Markow, 1995). Another advantage is that this mechanism provides for continuity over development, with adults and children attending to both perceptual and conceptual relations among objects in the context of word learning (also see Gentner & Markman, 1994). Second, the noun-shape is more restricted in its generalizability than is the noun-category position. The positive evidence for the noun-shape position rests primarily on observations of the influence of count nouns on the categorization of unfamiliar inanimate objects or artifacts. However, the reliance on object shape in word extension does not generalize beyond inanimates to include animate objects (Jones, Smith, & Landau, 1991). On the contrary, when a count noun is offered as a name for an animate-like object, the reliance on shape drops off dramatically (see Landau, 1994). This is consistent with the evidence that for adults and children, the signal value of shape (as well as other features) varies as function of object kind or ontology (Keil, 1994; Medin & Shoben, 1988). For animate objects, shape is not a sufficient predictor of an object’s category or its name. (See Ward et al., 1991, for additional evidence for this interpretation.) In contrast to the noun-shape position, the evidence for the noun-category position is robust for a broad range of ontological kinds, including both animate and inanimate objects. Crosslinguistic comparisons reveal another limitation in the generalizability of the noun-shape position. In classifier languages, an object’s shape is not incorporated within the head noun itself; instead, object shape is conveyed within a closed class system of noun classifier terms and the head noun refers to object substance. (See Craig, 1986; Lucy, 1996; Gentner & Boroditsky, in press for more extended discussions.) The noun-category position has sufficient breadth to account for this crosslinguistic variation. On this view, all infants begin with an early expectation that words refer to commonalities among objects. This general expectation is refined as infants come to notice (a) the distinct types of words in their language, and (b) the distinct types of relations among objects to which these words refer.
282
Sandra R. Waxman
This prediction is consistent with Lucy’s (1996) results on the acquisition of Maya (a classifier language). Lucy reports that early in acquisition, Maya children, like their English-speaking counterparts, prefer to extend nouns to object categories. As we have predicted, with development, this expectation is refined in accordance with their native language: infants acquiring classifier languages learn that nouns refer primarily to objects sharing a common substance, and that shape is marked within the classifier system. Imai and Gentner (1997) have offered similar evidence for children acquiring Japanese. What is missing from the noun-shape account is how children and adults come to discover commonalities among category members beyond those based on shape or substance. Third, the noun-shape and noun-category positions instantiate different models of word extension. The noun-shape position is essentially a univariate model, with shape as its primary index. At its most complex, a univariate model permits researchers to pit one property (e.g., shape) against another (e.g., texture, size, function, animacy cues). And indeed, most work on the shape bias involvespitting object shape, texture and size against one another in a binary fashion. But it is clear that the interactions among these predictors are crucial. Because a univariate model can only isolate and compare each of these predictors, it cannot capture the complexity underlying word meaning and object categorization. The noun-category position offers a more comprehensive account because it takes into account multiple indices, including (but not limited to) shape. It can therefore accommodate the fact that the relative weights associated with each predictor will vary as a function of ontological kind. For example, although shape may be weighted heavily in some categorization judgments (e.g., judgments regarding simple artifacts), it will carry less weight in others (e.g., judgments regarding animate objects). This position can also accommodate the fact that shape may receive greater emphasis early in development, but its relative importance may wane as infants and young children discover the additional commonalities underlying categorization and word extension (Imai, Gentner, & Uchida, 1994).This multivariate approach articulates well with current work documenting the contribution of many perceptual and conceptual factors in object categorization, naming and induction (Medin & Heit, in press). The noun-category position also permits an explanation for the acquisition of words for object categories that are not bound together by shape similarities (idea, museum, dream). On the noun-category account, extending a novel word to objects of a similar shape is only part of the process: an initial grouping, perhaps on the basis of a perceptual feature (e.g., shape), is then subjected to further scrutiny. This leads learners (children and
Linking Object Categorization and Naming
283
adults) to discover additional commonalities that may extend beyond perceptual features like shape. In sum, the noun-category position offers a more comprehensive account. It ensures that categorization in humans is a flexible and ongoing process. We continually incorporate new, perhaps deeper, information about commonalities within our existing categories; we also admit new instances into existing categories. These evolutions are, at least in part, an effect of naming. The complexities of objects (artifacts and natural kinds) and languages in the real world require that we attend to more than shape alone if we are to develop words and categories for the objects with which we daily do commerce.
Conclusion Humans are uniquely endowed with a natural capacity for building complex, flexible, and creative conceptual and semantic systems. The research program described in this chapter articulates a precise link between these two systems early in developmgnt, and provides insights into the origins and unfolding of the relation between them-across development, across languages, and across ontological kinds. We have proposed a continuous view of development, in which infants’ considerable perceptual and conceptual abilities are recruited early in the fundamental task of forming object categories and mapping words to their meanings. By integrating crosslinguistic and developmental investigations, we amplify the rich interplay between the expectations of the learner and the shaping role of the environment.
REFERENCES Au, T. K. (1990). Children’s use of information in word learning. Journal of Child Language, 17, 393-416. Au, T. K., Dapretto, M., & Song, Y.K. (1994). Input versus constraints: Early word acquisition in Korean and English. Journal of Memory and Language, 33, 567-582. Au,T. K., & Markman, E. M. (1987). Acquiringword meanings via linguisticcontrast. Cognitive Development, 2(3), 217-236. Baillargeon, R. (1993). The object concept revisited New direction in the investigation of infants’ physical knowledge. In C. Granrud (Ed.), Visual perception and cognition in infancy. Carnegie Mellon symposia on cognition (pp. 265-315). Hillsdale, NJ: Lawrence Erlbaurn Associates. Balaban, M. T., & Waxman, S. R. (1996). An examination of the factors underlying the facilitative effect of word phrases on object categorization in 9-month-old infants. In A. Stringfellow, D. Cahana-Amitay, E. Hughes, & A. Zukowski (Eds.), Proceedings of the
284
Sandra R Waxman
20th Boston University conference on Language Development, Vol. I (pp. 483-493). Somerville, MA: Cascadilla Press. Balaban, M. T., & Waxman, S. R. (1997). Do word labels facilitate categorization in 9-monthold infants? Journal of Experimental Child Psychology, 64, 3-26. Baldwin, D. A., Markman, E. M., & Melartin, R. L. (1993). Infants’ ability to draw inferences about nonobvious object properties: Evidence from exploratory play. Child Development, 64, 711-728. Barsalou, L. (1983). Ad hoc categories. Memory and Cognition, 11, 211-227. Behrend, D. A. (1990). Constraints and development: A reply to Nelson (1988). Cognitive Development, 5, 313-330. Berlin, B. (1973). The relation of folk systematicsto biological classificationand nomenclature. Annual Review of Systematics and Ecology, 4, 259-271. Berlin, B. (1992). Ethnobiological classification: Principles of categorization of plants and animals in traditional societies. Princeton, NJ: Princeton University Press. Bertenthal, B. I., Proffitt, D. R., Kramer, S. J., & Spetner, N. B. (1987). Infants’ encoding of kinetic displays varying in relative coherence. Developmental Psychology, 23, 171-178. Bhatt, R. S., & Rovee-Collier, C. (1996). Infants’ forgetting of correlated attributes and object recognition. Child Development, 67, 172-187. Bloom, L., Tinker, E., & Margulis, C. (1993). The words children learn: Evidence against a noun bias in early vocabularies. Cognitive Development, 8, 431-450. Bloom, P. (1994). Recent controversies in the study of language acquisition. In M. A. Gernsbacher (Ed.), Handbook of psycholinguistics (pp. 741-779). San Diego: Academic Press. Bloom, P. (1996). Intention, history, and artifact concepts. Cognition, 60, 1-29. Bowerman, M. (1985). What shapes children’s grammars? In D. I. Slobin (Ed.), The crosslinguistic study of language acquisition, Vol. I: The data; Vol. 2: Theoretical issues (pp. 1257-1319). Hillsdale, NJ: Lawrence Erlbaum Associates. Bowerman, M. (1996). The origins of children’s spatial semantic categories: Cognitive versus linguistic determinants. In J. J. Gumperz & S.C. Levinson (Eds.), Rethinking linguistic relativity. Studies in the social and cultural foundations of language, No. 17 (pp. 145-176). Cambridge: Cambridge University Press. Brown, R. (1957). Linguistic determinism and the part of speech, Journal of Abnormal and Social Psychology, 55, 1-5. Brown, R. (1958). Words and things: A n introduction to language. New York MacMillan. Carey, S. (1975). Conceptual change in childhood. Cambridge, MA: MIT Press. Choi, S., & Bowerman, M. (1991). Learning to express motion events in English and Korean: The influence of language-specificlexicalization patterns. Cognition, 41, 83-121. Choi, S., & Gopnik, A. (1995). Early acquisition of verbs in Korean: A cross-linguisticstudy. Journal of Child Language, 23, 497-529. Chomsky, N. (1965). Aspecrs of a theory of syntax. Cambridge: MA: MIT Press. Chomsky, N. (1986). Knowledge of language:Its nature, origin, and use. Westport, CT:Praeger. Clark, E. V. (1997). Conceptual perspective and lexical choice in acquisition. Cognition, 64, 1-37. Craig, C. G.(1986). Jacaltec noun classifiers: A study in language and culture. In C. G. Craig (Ed.), Noun classes and categorization (pp. 263-294). Philadelphia: Benjamins Publishing Company. Dixon, R. M. W. (1982). Where have all the adjectives gone? New York: Mouton. Echols, C. (1992, May). Developmental changes in attention to labeled events during the transition to language. Paper presented at the International Conference on Infancy Studies, Miami, FL.
Linking Object Categorization and Naming
285
Fenson, L., Cameron, M. S., & Kennedy, M. (1988). Role of perceptual and conceptual similarity in category matching at age two years. Child Development, 59, 897-907. Fenson, L., Dale, P. S., Reznick, J. S., Bates, E., Thal, D. J., & Pethick, S. J. (1994). Variability in early communicative development. Monographs of the Society for Research in Child Development, 5, 1-185. Fernald, A. (1992). Meaningful melodies in mothers’ speech to infants. In H. Papousek, U. Jurgens, & M. Papousek (Eds.), Nonverbal vocal communication: Comparative and developmental approaches. Studies in emotion and social interaction (pp. 262-282). New York: Cambridge University Press. Fisher, C., Hall, G., Rakowitz, S., & Gleitman, L. (1994). When it is better to receive than to give: Syntactic and conceptual constraints on vocabulary growth. Lingua, 92,333-376. Gallistel, C. R., Brown, A. L., Carey, S., Gelman, R., & Keil, F. C. (1991). Lessons from animal learning for the study of cognitive development. In S. Carey & R. Gelman (Eds.), The Epigenesis of mind: Essays on biology and cognition (pp. 3-35). Hillsdale, NJ: Lawrence Erlbaum Associates. Gathercole, V. C. M. (1997). The linguistic masskount distinction as an indicator of referent categorization in monolingual and bilingual children. Child Development, 68, 832-842. Gelman, R. (1991). Epigenetic foundations of knowledge structures: Initial and transcendent constructions.In S. Carey & R. Gelman (Eds.), The Epigenesis of mind: Essays on biology and cognition (pp. 293-322). Hillsdale, N J Lawrence Erlbaum Associates. Gelman, R., & Williams, E. M. (1998). Enabling constraints for cognitive development and learning: domain specificity and epigenesis. In W. Damon (Series Ed.), D. Kuhn & R. S. Siegler (Vol. Eds.), Handbook of child psychology: Vol. 2. Cognition, perception, and language (pp. 575-630). New York Wiley. Gelman, S. A. (1988). The development of inductionwithin natural kind and artifact categories. Cognitive Psychology, 20, 65-95. Gelman, S . A. (1996). Concepts and theories. In R. Gelman & T. K. Au (Eds.), Handbook of perception and cognition, Vol. 13: Perceptual and cognitive development (pp. 117-150). San Diego: Academic Press. Gelman, S. A., & Coley, J. D. (1990). The importance of knowing a dodo is a bird Categories and inferences in 2-year-old children. Developmental Psychology, 26, 796-804. Gelman, S. A., Coley, J. D., & Gottfried, G. M. (1994). Essentialist beliefs in children: The acquisition of concepts and theories. In L. A. Hirschfeld & S.A. Gelman (Eds.), Mapping the mind: Domain specificity in cognition and culture (pp. 341-365). New York: Cambridge University Press. Gelman, S. A., & Markman, E. M. (1985). Implicit contrast in adjectives versus nouns: Implications for word-learning in preschoolers. Journal of Child Language, 12, 125-143. Gelman, S . A,, & Markman, E. M. (1986). Categories and induction in young children. Cognition, 23, 183-209. Gelman, S. A., & Medin, D. L. (1993). What’s so essential about essentialism? A different perspective on the interaction of perception, language, and conceptual knowledge. Cognitive Development, 8, 157-167. Gelman, S. A., & Taylor, M. (1984). How two-year-old children interpret proper and common nouns for unfamiliar objects. Child Development, 55, 1535-1540. Gentner, D. (1981). Some interesting differences between verbs and nouns. Cognition and Brain Theory, 4, 161-178. Gentner, D. (1982). Why nouns are learned before verbs: Linguistic relativity versus natural partitioning. In S. Kuczaj (Ed.), Language development, Vol. 2: Language, thought, and culture (pp. 301-334). Hillsdale, NJ: Lawrence Erlbaum Associates.
286
Sandra R. Waxmao
Gentner, D., & Boroditsky, L.(in press). Individuation, relativity and early work learning. In M. Bowerman & S. Levinson (Eds.), Language acquisition and conceptual development. England Cambridge University Press. Gentner, D., & Markman, A. B. (1994). Structural alignment in comparison: No difference without similarity. Psychological Science, 5(3), 152-158. Gentner, D., & Namy, L. L. (1998). Comparison in the development ofcategories. Manuscript submitted for publication. Gentner, D., & Waxman, S. R. (1994, June). Perceptual and conceptual bootstrapping in early word meaning. Paper presented at the meeting of the International Conference on Infancy Studies, Paris, France. Gerken, L., & McIntosh, B. J. (1993). Interplay of function morphemes and prosody in early language. Developmental Psychology, 29, 448-457. Gleitman, L. R. (1990). The structural sources of word meaning. Language Acquisition, I, 3-55. Goldin-Meadow, S., Seligman, M. E., & Gelman, R. (1976). Language in the two-year-old. Cognition, 4, 189-202. Golinkoff, R. M., Hirsh-Pasek, K., Cauley, K. M., & Gordon, L. (1987). The eyes have it: Lexical and syntactic comprehension in a new paradigm. Journal of Child Language, 14, 23-45. Golinkoff, R. M., Mervis, C. B., & Hirsh-Pasek, K.(1994). Early object labels: The case for a developmental lexical principles framework. Journal of Child Language, 21, 125-155. Golinkoff, R. M., Shuff-Bailey, M., Olguin, R., & Ruan, W. (1995). Young children extend novel words at the basic level Evidence for the principle of categorical scope. Developmental Psychology, 31, 494-507. Gopnik, A., & Meltzoff, A. N. (1992). Categorization and naming: Basic-level sorting in eighteen-month-olds and its relation to language. Child Development, 63, 1091-1 103. Greenberg, J. H. (1963). Some universals of grammar with particular reference to the order of meaningful elements. In J. H. Greenberg (Ed.), Universals of language. Cambridge, MA: MIT Press. Grimshaw, J. (1981). Form, function, and the language acquisition device. In C. L. Baker & J. McCarthy (Eds.), The logical problem of language acquisition. Cambridge, MA: MIT Press. Grimshaw, J. (1994). Lexical reconciliation. Lingua, 92, 411-430. Hall, D. G. (1991). Acquiring proper nouns for familiar and unfamiliar animate objects: Twoyear-olds’ word learning biases. Child Development, 62, 1142-1 154. Hall, D. G. (1993). Basic-level individuals. Cognition, 48, 199-221. Hall, D. G. (1994). How children learn common nouns and proper names. In J. Macnamara & G. E. Reyes (Eds.), The logical foundations of cognition. Vancouver studies in cognitive science, Vol. 4 (pp. 212-240). New York: Oxford University Press. Hall, D. G., & Moore, C. E. (1997). Red bluebirds and black greenflies: Preschoolers’understanding of the semanticsof adjectives and count nouns.Journal of Experimental Psychology, 67, 236-267. Hall, D. G . , & Waxman, S. R. (1993). On the acquisition of restricted and unrestricted nouns in lexical acquisition. Child Development, 64, 1550-1570. Hall, D. G., Waxman, S. R., & Hurwitz, W. M. (1993). How 2- and 4-year-old children interpret adjectives and count nouns.Child Development, 64, 1651-1664. Held, R., & Hein, A. (1963). Movement-produced stimulation in the development of visually guided behavior. Journal of Comparative and Physiological Psychology, 56, 872-876. Hirsh-Pasek, K.,Golinkoff, R., & Naigles, L. (1996). Young children’s use of syntactic frames to derive meaning. In K. Hirsh-Pasek & R. Golinkoff (Eds.), The origins of grammar (pp. 123-158). Cambridge: MIT Press.
Linking Object Categorization and Naming
287
Huttenlocher, J., & Smiley, P. (1987). Early word meanings: The case of object names. Cognitive Psychology, 19, 63-89. Imai, M., & Gentner, D. (1997). A cross-linguistic study of early word meaning: Universal ontology and linguistic influence. Cognition, 62, 169-200. Imai, M., Gentner, D., & Uchida, N. (1994). Children’s theories of word meaning: The role of similarity in early acquisition. Cognitive Development, 9, 45-75. Jackendoff, R. (1990). Semantic structures. Cambridge, MA: MIT Press. Jesperson, 0. (1968). The phirosophy of grammar. London: George Allen & Unwin. Jones, S., Smith, L., & Landau, B. (1991). Object properties and knowledge in early lexical learning. Child Development, 62, 499-516. Juscyzk, P., & A s h , R. N. (1995). Infants’ detection of the sound patterns of words in fluent speech. Cognitive Psychology, 29, 1-23. Kalish, C. W., & Gelman, S. A. (1992). On wooden pillows: Multipleclassificationand children’s category-based inductions. Child Development, 63, 1536-1557. Katz, N., Baker, E., & Macnamara, J. (1974). What’s in a name? A study of how children learn common and proper names. Child Development, 45, 469-473. Keil, F. C. (1987). Conceptual development and category structure. In U. Neisser (Ed.), Concepts and conceptual development: Ecological and intellectualfactors in categorization (pp. 175-200). New York: Cambridge University Press. Keil, F. C. (1994). The birth and nurturance of concepts by domains: The origins of concepts of living things. In L. A. Hirschfeld & S. A. Gelman (Eds.), Mapping the mind: Domain spec@city in cognition and culture (pp. 234-254). New York Cambridge University Press. Kemler-Nelson, D. G. (1995). Principle-based inferences in young children’s categorization: Revisiting the impact of function of the naming of artifacts. Cognitive Development, 10, 347-380. Klibanoff, R. S., & Waxman, S. R. (in press). Preschoolers’ acquisition of novel adjectives and the role of basic-level kind. To appear in Proceedings of the 22th Boston University Conference on Language Development. Somerville, M A Cascadilla Press. Landau, B. (1994). Object shape, object name, and object kind: Representation and development. In D. L. Medin (Ed.), The psychology of learning and motivation: Advances in research and theory, Vol. 31 (pp. 253-304). San Diego: Academic Press. Landau, B., & Gleitman, L. R. (1985). Language and experience: Evidence from the blind child. Cambridge, M A Harvard University Press. Landau, B., Jones, S., & Smith, L. (1992). Perception, ontology, and naming in young children: Commentary on Soja, Carey, and Spelke. Cognition, 43, 85-91. Landau, B., Smith, L., & Jones, S. (1988). The importance of shape in early lexical learning. Cognitive Development, 3, 299-321. Lassaline, M. E., & Murphy, G. L. (1996). Induction and category coherence. Psychological Bulletin and Review, 3, 95-99. Lucy, J. A. (1996). The scope of linguistic relativity: An analysis and review of empirical research. In J. J. Gumpen & S. C. Levinson (Eds.), Rethinking linguistic relativity. Studies in the social and cultural foundations of language, No. 17 (pp. 37-69). Cambridge: Cambridge University Press. Lyons, J. (1977). Semantics: Vol I. New York Cambridge University Press. Macnamara, J. (1982). Names for things. Cambridge, MA: MIT Press. Macnamara, J. (1986). A border dispute: The place of logic in psychology. Cambridge, MA: MIT Press. Macnamara, J. (1994). Logic and cognition. In J. Macnamara & G. E. Reyes (Eds.), The logical foundations of cognition. Vancouver studies in cognitive science, Vol. 4 (pp. 11-34). New York: Oxford University Press.
288
Sandra R Warnran
Mandler, J. M. (1988). How to build a baby: On the development of an accessible representational system. Cognitive Development, 3, 113-136. Mandler, J. M. (1992). How to build a baby: 11. Conceptual primitives. Psychological Review, 99, 587-604. Mandler, J. M., & Bauer, P. J. (1988). The cradle of categorization: Is the basic level basic? Cognitive Development, 3, 247-264. Mandler, J. M., Bauer, P. J., & McDonough, L. (1991). Separating the sheep from the goats: Differentiating global categories. Cognitive Psychology, 23, 263-298. Mandler, J. M., & McDonough, L. (1993). Concept formation in infancy. Cognitive Development, 8, 291-318. Mandler, J. M., & McDonough,L. (1996). Drinking and driving don’t mix: inductive generalization in infancy. Cognition, 59, 307-335. Maratsos, M. P. (1991). How the acquisition of nouns may be different from that of verbs. In N. A. Krasnegor, D. M. Runbaugh, R. L. Schiefelbusch, & M. Studdart-Kennedy (Eds.), Biological and Behavioral Determinants of Language Development (pp. 67-88). Hillsdale, NJ: Lawrence Erlbaum Associates. Maratsos, M., & Chalkley, M. A. (1980). The internal language of children’s syntax: The ontogenesis and representation of syntactic categories. In K. Nelson (Ed.), Children’s language, Vol. 2. New York Gardner Press. Markman, E. M . (1989). Categorization and naming in children: Problems of induction. Cambridge, MA: MIT Press. Markman, E. M. (1994). Constraints on word meaning in early language acquisition. Lingua, 92, 199-227. Markman, E. M., & Wachtel, G. F. (1988). Children’s use of mutual exclusivity to constrain the meanings of words. Cognitive Psychology, 20, 121-157. Markman, E. M., & Hutchinson, J. E. (1984). Children’s sensitivity to constraints on word meaning: Taxonomic vs. thematic relations. Cognitive Psychology, 16, 1-27. Marler, P. (1991). The instinct to learn. In S. Carey & R. Gelman (Eds.), The epigenesis of mind: Essays on biology and cognition (pp. 37-66). Hillsdale,N J Lawrence Erlbaum Associates. Medin, D. L., & Heit, E. J. (in press). Categorization. In D. Rumelhart & B. Martin (Eds.), Handbook of cognition and perception. Hillsdale, NJ: Lawrence Erlbaum Associates. Medin, D. L., & Shoben, E. J. (1988). Context and structure in conceptual combination. Cognitive Psychology, 20, 158-190. Mervis, C. B., & Crisafi, M. A. (1982). Order of acquisition of subordinate-, basic-, and superordinate-level categories. Child Development, 53, 258-266. Mervis, C. B., & Mervis, C. A. (1988). Role of adult input in young children’s category evolution: I. An observational study. Journal of Child Language, 15(2), 257-272. Mervis, C. B., Johnson, K. E., & Scott, P. (1993). Perceptual knowledge, conceptual knowledge, and expertise: Comment on Jones and Smith. Cognitive Development, 8, 149-155. Mill, J. S . (1843). A system of logic, ratiocinative and inductive. London: Longman Group. Murphy, G. L., & Medin, D. L. (1985). The role of theories in conceptual coherence. Psychological Review, 92, 289-316. Naigles, L. (1996). The use of multiple frames in verb learning via syntactic bootstrapping. Cognition, 58, 221-251. Naigles, L., & Eisenberg, A. (in press). Language development: A key to the influence of language on thought. Social Cognition. Namy, L. L., & Waxman, S . R. (1997). Naming and exclaiming: Infants’ sensitivity to naming contexts. Manuscript submitted for publication. Namy, L. L., & Waxman, S. R. (in press). Words and gestures: Infants’ interpretations of different forms of symbolic reference. Child Development.
Linking Object Categorization and Naming
289
Nelson, K. (1973). Some evidence for the cognitive primacy of categorization and its functional basis. Merrill-Palmer Quarterly of Behavior and Development, 19, 21-39. Nelson, K. (1988). Constraints on word learning? Cognitive Development, 3, 221-246. Nelson, K., Hampson, J., & Shaw, L. K. (1993). Nouns in early lexicons: Evidence,explanations and implications. Journal of Child Language, 20(1),61-84. Pinker, S. (1984). Language learnability and language development. Cambridge, MA: Harvard University Press. Pinker, S. (1994). How could a child use verb syntax to learn verb semantics? Lingua, Y2, 377-410. Prasada, S. (1992). Acquisition of adjective meanings:A lexical semantic approach. Unpublished doctoral dissertation, Massachusetts Institute of Technology, Cambridge, MA. Prasada, S. (1996). Quantification, arbitrariness of structure, and the count-mass noun distinction. In A. Stringfellow,D. Cahana-Amitay, E. Hughes, & A. Zukowski (Eds.), Proceedings of the 20th Boston University Conference on Language Development, Vol. 2 (pp. 600-609). Somerville: M A Cascadilla Press. Putnam, H. (1975). Mind, language and reality: Philosophical papers, Vol. 2. Cambridge: Cambridge University Press. Quine, W. V. (1960). Word and object. Cambridge, MA: MIT Press. Quinn, P. C., Eimas, P. D., & Rosenkrantz, S. L. (1993). Evidence for representations of perceptually similar natural categories by 3-month-old and 4-month-old infants. Perception, 22, 463-475. Rips, L. (1989). Similarity, typicality, and categorization. In S. Vosniadou & A. Ortony (Eds.), Similarity and analogical reasoning (pp. 21-59). New York Cambridge University Press. Roberts, K. (1988). Retrieval of a basic-level category in pre-linguistic infants. Developmental Psychology, 24, 21-27. Roberts, K., & Cuff, M. D. (1989). Categorization studies of 9- to 15-month-old infants: Evidence for superordinate categorization? Infant Behavior and Development, 12, 265-288. Roberts, K., & Horowitz, F. D. (1986). Basic level categorizationin 7- and 9-month-old infants. Journal of Child Language, 13, 191-208. Roberts, K.,& Jacob, M. (1991). Linguistic versus attentional influences on non-linguistic categorization in 15-month-old infants. Cognitive Development, 6, 355-375. Rosch, E. (1978). Principles of categorization. In E. Rosch & B. B. Lloyd (Eds.), Cognition and categorization (pp. 27-48). Hillsdale, NJ: Lawrence Erlbaum Associates. Rosch, E., Mervis, C. B., Gray, W.D., Johnson, D. M., & Boyes-Braem, P. (1976). Basic objects in natural categories. Cognitive Psychology, 8, 382-439. Ross, B. H. (1996). Category representations and the effects of interacting with instances. Journal of Experimental Psychology: Learning, Memory and Cognition, 22, 1249-1265. Saah, M. I., Waxman, S. R., & Johnson, J. (1996). The composition of children’s early lexicons as a function of age and vocabulary size. Paper presented at the 21st Boston University Conference on Language Development, Boston, MA. Shipley, E. F. (1993). Categories, hierarchies, and induction. The Psychology of Learning and Motivation, 30, 265-301. Smith, L. B. (1995). Self-organizing processes in learning to learn words: Development is not induction. In C. A. Nelson (Ed.), Basic and applied perspectives on learning, cognition, and development. The Minnesota Symposia on Child Psychology, Vol 28. (pp. 1-32). Mahwah, NJ: Lawrence Erlbaum Associates. Smith, L. B., & Heise, D. (1992). Perceptual similarity and conceptual structure. In B. Bums (Ed.), Percepts, concepts and categories: The representation and processing of information. Advances in psychology, Vol. 93 (pp. 233-272). Amsterdam: North-Holland.
290
Sandra R. Waxman
Smith, L. B., Jones, S. S., & Landau, B. (1992). Count nouns, adjectives, and perceptual properties in children’s novel word interpretations. Developmental Psychology, 28, 273-286. Soja, N. N., Carey, S . , & Spelke, E. (1991). Ontological categories guide young children’s inductions about word meaning: Object terms and substance terms. Cognition, 38, 179-21 1. Soja,N. N., Carey, S., & Spelke, E. (1992). Perception, ontology, and word meaning. Cognition, 45, 101-107. Spelke, E. S . (1993). Object perception. In A. I. Goldman (Ed.), Readings in philosophy and cognitive science (pp. 447-460). Cambridge, MA: MIT Press. Talmy, L. (1985). Lexicalization patterns: Semantic structure in lexical forms. In T. Shopen (Ed.), Language typology and syntactic description, Vol. 3 (pp. 57-149). Cambridge: Cambridge University Press. Tardif, T. (1996). Nouns are not always learned before verbs: Evidence from Mandarin speakers’ early vocabularies. Developmental Psychology, 32(3), 492-504. Tardif, T., Shatz, M., & Naigles, L. (in press). Caregiver speech and children’s use of nouns versus verbs: A comparison of English, Italian, and Mandarin. Journal of Child Language. Taylor, M., & Gelman, S. A. (1988). Adjectives and nouns: Children’s strategies for learning new words. Child Development, 59, 411-419. Tomasello, M. (1992). First verbs: A case study of early grammatical development. Cambridge: Cambridge University Press. Tomasello, M., & Akhtar, N. (1995). Two-year-oldsuse pragmatic cues to differentiate reference to objects and actions. Cognitive Development, 10, 201-224. Tomasello, M., & Olguin, R. (1993). Twenty-three-month-old children have a grammatical category of noun. Cognitive Development, 8(4), 451-464. Ward, T. B., Becker, A. H., Hass, S. D., & Vela, E. (1991). Attribute availability and the shape bias in children’s category generalization. Cognitive Development, 6, 143-167. Waxman, S. R. (1990). Linguistic biases and the establishment of conceptual hierarchies: Evidence from preschool children. Cognitive Development, 5(2), 123-150. Waxman, S. R. (1991). Convergences between semantic and conceptual organization in the preschool years. In S. A. Gelman & J. P. Byrnes (Eds.), Perspectives on language and cognition: Interrelations in development (pp. 107-145). Cambridge: Cambridge University Press. Waxman, S. R. (1994). The development of an appreciation of specific linkages between linguistic and conceptual organization. Lingua, 92, 229-257. Waxman, S . R. (1995). Characteristicsof word learners at 12- and 30-months:Early emergence and modification of the noun-category linkage. In D. MacLaughlin & S. McEwen (Eds.), Proceedings of the 19th Boston University Conference on Language Development, Vol. 1 (pp. 667-678). Somerville, MA: Cascadilla Press. Waxman, S. R. (in press). The dubbing ceremony revisited: Object naming and categorization in infancy and early childhood. To appear in D. L. Medin & S. Atran (Eds.), Folkbiology. Cambridge, MA: MIT PresslBradford Books. Waxman, S. R., & Braig, B. (1996). Stars and starfish: How far can shape take us? Paper presented at the International Conference on Infancy Studies, Providence, RI. Waxman, S. R., Bredart, S., Nicolay, A., & Hall, D. G. (1998). The semantic distinction between count nouns and adjectives in preschool-aged children: Cross-linguistic evidence. Manuscript in preparation. Waxman, S. R., & Hall, D. G. (1993). The development of a linkage between count nouns and object categories: Evidence from 16- to 21-month-old infants. Child Development, 64, 1224-1241.
Linking Object Categorization and Naming
291
Waxman, S. R., & Kosowski, T. (1990). Nouns mark category relations: Toddlers’ and preschoolers’ word-learning biases. Child Development, 61(5), 1461-1473. Waxman, S. R., Lynch, E. B., Casey, K. L., & Baer, L. (1997). Setters and samoyeds: The emergence of subordinate level categories as a basis for inductive inference. Developmental Psychology, 33(6), 1074-1090. Waxman, S. R., & Markow, D. B. (1995). Words as invitations to form categories: Evidence from 12-month-old infants. Cognitive Psychology, 29, 257-302. Waxman, S. R., & Markow, D. B. (in press). Object properties and object kind: 21-monthold infants’ extension of novel adjectives. Child Development. Waxman, S. R., & Namy, L. (1997). Challenging the notion of thematic bias in young children. Developmental Psychology, 33(3), 555-567. Waxman, S. R., Philippe, M., & Branning, A. (1997). A matter of time: Novel nouns mark object categories when delays are imposed. Manuscript submitted for publication. Waxman, S. R., & Senghas, A. (1992). Relations among word meanings in early lexical development. Developmental Psychology, 28(5), 862-873. Waxman, S. R., Senghas, A,, & Benveniste, S. (1997). A cross-linguistic examination of the noun-category bias: Its existence and specificity in French- and Spanish-speaking preschool-aged children. Cognitive Psychology, 43, 183-218. Waxman, S. R., Stote, R., & Philippe, M. (1997). Count nouns and object categories: Modifications in word-learners’ expectations from infancy through the preschool years. Poster presented at the meeting of the Society for Research in Child Development, Washington, DC. Wetzer, H. (1992). “Nouny” and “verby” adjectivals: A typology of predicate adjectival constructions. In M. Kefer & J. Van der Auwera (Eds.), Meaning and grammar: Crosslinguistic perspectives (pp. 223-262). New York: Mouton de Gruyter. Wierzbicka, A. (1986). What’s in a noun? (or: How do nouns differ in meaning from adjectives?). Studies in Language, 10(2), 353-389. Wilcox, T., & Baillargeon, R. (in press). Object individuation in infancy: The use of featural information in reasoning about occlusion events. Cognitive Psychology. Xu, F. (1997, November). Distinct labels provide pointers to distinct sortals for 9-monthold infants. Paper presented at the 22nd Boston University Conference on Language Development, Boston, MA. Xu,F., & Carey, S. (1996). Infants’ metaphysics: The case of numerical identity. Cognitive Psychology, 30, 111-153. Younger, B. A., & Cohen, L. B. (1986). Developmental change in infants’ perception of correlations among attributes. Child Development, 57, 803-815.
This Page Intentionally Left Blank
INDEX
A Absolute accuracy, of metacognition, 225 Abstraction-inhibition theory distractor representation in, 19 effect of attended stimuli, 19-20 negative priming and, 18-19 Abstractions, memory as, 8-10 Accuracy, in metacognitive tasks absolute, 225 agreement, 225 description of, 223 relative, see Relative accuracy Adaptivity definition of,120-121 extrinsic, 123-126 intrinsic, 123-124 strategic, to individual differences in arithmetic game show paradigm, 119-120 individual differences description of, 126-128 intrinsic versus extrinsic adaptivity, 128-131 retrieve versus compute, 117-118 strategy selection adaptivity in, 120-122 components of, 122-126 extrinsic adaptivity, 123-126, 128-131 intrinsic adaptivity, 123-124, 128- 131 methods for studying, 118-119
awareness and, 138 in complex, dynamic tasks Kanfer-Ackerman Air Traffic Controller Task, 139-144 static versus dynamic tasks, 139 in controlled environments advantages of, 131 Building Sticks Task, 131-138 description of, 131 cross-task stability, 150 description of, 117 importance of, 117 Adjectives count nouns and, distinction between in acquisition, 257-258 crosslinguistic studies, 268-271 for object categories, 256-258 specificity of, developmental changes in, 264-265 Agreement accuracy, of metacognition, 225 Amygdala-centered system, of memory, see also Hot memory system damage-related effects, 192 description of, 192 emotional fragmentary stimuli, 194-195 emotional impairments, 192-193 fear responses generated by, 193-194 hippocampal memory and, interaction between description of, 195-196 modulation hypothesis, 195 stress-related effects, 204-205 Animal learning, see Species typical cues Anticipation effect, 92-93
293
294
Index
Arithmetic, strategy adaptivity approach to individual differences in description of, 126-128 game show paradigm, 119-120 intrinsic versus extrinsic adaptivity, 128-131 retrieve versus compute, 117-118 strategy selection adaptivity in, 120-122 components of, 122-126 extrinsic adaptivity, 123-126, 128131 intrinsic adaptivity, 123-124, 128-131 methods for studying, 118-119 Associative priming, 197
B Backward blocking, 47 Before-disruption effect cause of, 36 definition of, 32 experiments of, 32-33 Behavioral psychology ethology and comparisons, 155-156 species typical cues for studying methodological integration, see Species typical cues general-process approach, 155-156 origins of, 155 Bias definition of, 226 intertemporal choice constructive preferences, 109-111 generality of, 93-94 psychophysics of, 108-109 as measure of metacomprehension accuracy description of, 227, 244 experiments, 235 test performance, 240 Brier score, 226,228 Building Sticks Task, for studying strategic adaptivity differences description of, 131-133 explicit awareness and extrinsic adaptivity, relationship between, 137-138 principles of, 131-133
study methods, 133-134 study results, 134-138
C Calibration, as measure of relative accuracy, 245 Categorization, of objects, see Object categorization CHARM model, of human memory, 1%-199 Childhood phobias, reemergence in adulthood, hippocampal-based explanation for, 201 Cognition, see Metacognition Composite holographic associative recall recognition memory model, see CHARM model Compound training, cue competition in the absence of associate literature findings, 61 backward blocking effect, 63 blocking by context effect, 47-48 definition of, 45 description of, 47-49 effect of retrieval cue, 65-66 experiments, 49-55,63-66 forward blocking experiment, 45-46 in judgmental tasks “allergy” task, 66-68 experiments, 68-73 temporal context effects, 69-73 Rescorla-Wagner model description of, 47-48 revising of, 55-59 during retrieval, 62 temporal contexts, 63-65 unlearning explanation, 55-59 Conceptually-drivenprocesses, 5-6 Conditioning, see Sexual conditioning Conspecific as conditioned stimulus, for studying species typical cues early use of, 158-159 sexual conditioning description of, 159-160 experiments, 159-161 learning to discriminate the sex of conspecific, 163-170 mechanisms of learning, 161-163
Index Pavlovian conditioning, 163 social proximity response, 160-163 Context similarity, negative priming and description of, 14 effect on location-specific negative priming, 24 Cool memory systems adaptive significance of, 213-214 anatomical elements, 189 case study of, 188-189 description of, 187 episodic binding description of, 189 hippocampal damage effects, 189192 function of, 187-189 hot memory systems and framework implication of, for memory under stress low levels of stress, 206-207 moderate to high levels of stress, 207-210 traumatic stress, 210-213 relation to other views of human memory CHARM model, 196-199 implicit memory-explicit memory view, 199-200 procedural vs. declarative, 200 taxon-locale view, 200-201 interaction between description of, 195-196 modulation hypothesis, 195 stress-related effects, 205-206 stress-related effects, 202-204 as “taxon” system, 201 Counterconditioning, 59-60 Count nouns adjectives and, distinction between in acquisition, 257-258 for object categories, 256-258, 279283 specificity of, developmental changes in, 264 Cue competition in the absence of compound training associate literature findings, 61 backward blocking effect, 63 blocking by context effect, 47-48
295 definition of, 45 description of, 47-49 effect of retrieval cue, 65-66 experiments, 49-55, 63-66 forward blocking experiment, 45-46 in judgmental tasks “allergy” task, 66-68 experiments, 68-73 temporal context effects, 69-73 Rescorla-Wagner model description of, 47-48 revising of, 55-59 during retrieval, 62 temporal contexts, 63-65 unlearning explanation, 55-59 inhibitory association, 74-75 theories of interference between outcomes and extending of models, 73-76 similarities between, 59-66
D Data-driven processes, 5-6 Delay effect, in discount rates, 86-88 Discounted utility theory discount rate, see Discount rate normative, 84 principles of, 83-84 Discount rate anticipation effect, 92-93 definition of, 83 delay effect, 86-88 high and variable, 85-86 inconsistent performance associated with variations of, 84 magnitude effect, 88 positive, 84 sequence effect, 89-90 sign effect, 88-89 spreading effect, 90-92 Dissociations, effect on implicit memory, 5
E Endogenous negative priming, 21-22 Episodic memory binding, role of hippocampus in description of, 189 hippocampal damage effects, 189-192 retrieval of, factors that influence, 16-17
296
Index
Episodic retrieval theory as cause of negative priming, 12-14 description of, 12-14 effect of attended stimuli, 19-20 experiments of, 17-18 modified, 34 Ethology appetitive and consummatory behavior, 157 behavioral psychology and comparisons, 155-156 integration efforts, 157 methodological integration overview, 158 species typical cues, see Species typical cues origins of, 155 Exogenous negative priming, 21-22 Explicit memory conceptually-driven,5-6 definition of, 5 tasks classified using hot-cool memory systems, 199-200 Extinction description of, 59-60 effect of temporal context on, 65 Extrinsic adaptivity individual differences in in arithmetic tasks, 128-131 in complex, dynamic tasks description of, 139 Kanfer-Ackerman Air Traffic Controller Task, 139-144 static versus dynamic tasks, 139 description of, 123-126 intrinsic adaptivity and, 149
F Fear conditioning effect of epinephrine and norepinephrine levels, 210 “hot” memory system classification of, 193-194,200 Feeding, Timberlake’s behavioral approach to, 157 Feeling-of-knowing (FOK) description of, 223 gamma as measure of, 224-225
judgments, 225 measures of, 224-225, 239
G Gamma description of, 224-225 as metacomprehension measure description of, 246 experiments. 237-239 test performance, 240
H Hippocampus, see also Cool memory system amygdala and, developmental differences in, 201 damage to effect on episodic memory binding function, 189-192 reemergence of childhood phobias and, 201 episodic memory binding function description of, 189-192 stress effects, 202-204 stress-related effects, 202-204 taxon systems and, 201 Hot memory systems adaptive significance of, 213-214 case study of, 188-189 characterization of emotional fragmentary stimuli processing, 194-195 emotional modulation, 192-193 fear conditioning, 193-194 cool memory systems and framework, relation to other views of human memory CHARM model, 196-199 implicit memory-explicit memory view, 199-200 procedural vs. declarative, 200 taxon-locale view, 200-201 interaction between description of, 195-196 modulation hypothesis, 195 description of, 187 function of, 187-188 stress-related effects, 204-205 as “taxon” system, 200-201
Index
I Immediate judgments of learning, 244 Implicit memory data-driven, 5-6 definition of, 2, 4 dissociations that influence, 5 tasks classified using hot-cool memory systems, 199-200 word-fragment completion task, 5 Individual differences parameters approach description of, 115 elements of, 115-1 16 processing speed, 115-116 working memory capacity, 115116 strategies approach description of, 116 simple form of, 116 strategy adaptivity approach in arithmetic description of, 126-128 game show paradigm, 119-120 intrinsic versus extrinsic adaptivity, 128- 131 retrieve versus compute, 117-118 strategy selection adaptivity in, 120-122 components of, 122-126 extrinsic adaptivity, 123-126, 128- 131 intrinsic adaptivity, 123-124, 128- 131 methods for studying, 118-119 awareness and, 138 in complex, dynamic tasks Kanfer-Ackerman Air Traffic Controller Task, 139-144 static versus dynamic tasks, 139 in controlled environments advantages of, 131 Building Sticks Task, 131-138 description of, 131 cross-talk stability, 150 description of, 117 importance of, 117 Inhibition of return definition of, 26 description of, 25-28, 35
experiments, 26-38 location-specific negative priming and, relationship between, 26-27 Instances, memory as, 8-10 Interference between outcomes, theories of, 59-66 proactive, 2-3, 61 retroactive, 2, 61 Intertemporal choice biases constructive preferences, 109- 111 generality of, 93-94 psychophysics of, 108-109 description of, 83-85 violations of anticipation effect, 92-93 delay effect, 86-88 high and variable discount rates, 85-86 magnitude effect, 88 sequence effect, 89-90 sign effect, 88-89 spreading effect, 90-92 Intrinsic adaptivity description of, 123-124 extrinsic adaptivity and, 149 IOR, see Inhibition of return
J JOL, see Judgments-of-learning Judgments-of-learning,223
K Kanfer-Ackerman Air Traffic Controller Task, for measuring individual differences in extrinsic adaptivity base-rate manipulations, 141-144 description of, 139-141 design of, 139-141 micro-level extrinsic adaptivity, 144147
L Location-specific negative priming description of, 22-25 inhibition of return and, relationship between, 26-27
297
298
Index
M Magnitude effect, 88 Memory abstraction versus instances, 8-10 amygdala-centered system of, see Hot memory system episodic binding, role of hippocampus in description of, 189 hippocampal damage effects, 189-192 retrieval of, factors that influence, 16-17 explicit conceptually-driven,5-6 definition of, 5 tasks classified using hot-cool memory systems, 199-200 implicit data-driven, 5-6 definition of, 2, 4 dissociations that influence, 5 tasks classified using hot-cool memory systems, 199-200 word-fragment completion task, 5 Memory systems “cool,” see Cool memory systems “hot,” see Hot memory systems Metacognition accuracy of absolute, 225 agreement, 225 description of, 223 measures of, comparison of, 224-228 relative, 225 correlation measures feeling-of-knowing,223 judgments-of-learning,223 judgment-performance relationships, measures of, 226-227 Metacomprehension effect of expectancy and processing on accuracy manipulations, 229 bias, 235, 244 description of, 228-229 experiments, 230-243 measures of, 223-224 Metamemory accuracy of, 242 test performance and, 239-240
N Naming, relationship with object categorization cross-cultural evidence, 251,268-271 description of, 252, 271 developmental and crosslinguistic considerations adjectives, 256-258 count nouns, 256-258 experiments, 255-256 effect of learner constraints and role of environment, 250-251 empirical evidence crosslinguistic, 268-271 developmental studies, 259-268 integration of, 271-273 general expectations description of, 272-273 lexical development and, 274-275 progression from, to more specific expectations, 273-275 perceptual and conceptual contributions, 276 role of basic-level object categories, 277-279 word learning and conceptual development, 253-255 Negative priming attended stimuli as cause, 19-20 causes of, 11-14, 37 context similarity and, 14 dependence of, 13-14 duration of, 15-16 endogenous, 21-22 exogenous, 21-22 experiments on, 10-13 location-specific description of, 22-25 inhibition of return and, relationship between, 26-27 modified theory of, 34-35 object-specific,22-25 origins of, 10 pre-prime interval effects, 16-17 short-term versus long-term instances, 16 in studies, 2 Norepinephrine, stress-related effects on synthesis of, 204-205
Index Normative theory principles of, 85 violations of anticipation effect, 92-93 delay effect, 86-88 high and variable discount rates, 85-86 magnitude effect, 88 sequence effect, 89-90 sign effect, 88-89 spreading effect, 90-92 Nouns, count adjectives and, distinction between in acquisition, 257-258 for object categories, 256-258, 279-283 specificity of, developmental changes in, 264 Novel words, effect on object categorization cross-cultural evidence, 268-271 description of, 259-263 meaning associated with, 279-283 object category formation, 271
0 Object categories basic-level vs. global-level, 277-279 description of, 249,251 early formation scope of, 276 similarities to formation by older adults, 275-276 novel words and, 259-263 object shape vs., effect of count nouns in, 279-283 scope of, 252 words and, linkage between, 256-258 Object categorization cross-cultural use of, 249 definition of, 252 description of, 249 in language of development of child, 252 naming and, relationship between cross-cultural evidence, 251, 268-271 description of, 252, 271 developmental and crosslinguistic considerations adjectives, 256-258 count nouns, 256-258 experiments, 255-256
299
effect of learner constraints and role of environment, 250-251 empirical evidence crosslinguistic, 268-271 developmental studies, 259-268 integration of, 271-273 general expectations description of, 272-273 lexical development and, 274-275 progression from, to more specific expectations, 273-275 perceptual and conceptual contributions, 276 role of basic-level object categories, 277-279 word learning and conceptual development, 253-255 novel words and, 249,259-263 Iutcomes, competition between, 60-62
P Paired associate learning, 2 Pearson r correlations, as metacomprehension measure, 237-238 Phobias, childhood, reemergence in adulthood, 201 Positive priming, 11 Preference reversals, 87 Priming associative, 197 definition of, 1 repetition, see Repetition priming Proactive interference in associative learning literature, 61 definition of, 2 studies of, 2-3 Processing, see Transfer-appropriate processing; Transfer-inappropriate processing
R Rapid serial visual presentation, 28-29 Reinstatement, 62 Relative accuracy, in metacognitive tasks description of, 225 measures of calibration, 245 resolution, 245 Repetition, performance and, 2
300 Repetition blindness definition of, 28-29 description of, 35 effect on discrimination, 29 memory-retrieval phenomenon and, 30 perceptual theories regarding, 30-32 in rapid serial visual presentation, 29-30 token-individuation hypothesis of, 30 transfer-appropriate processing, 35 Repetition priming, 197 delinition of, 1 effect of distractor onsets on, 14 Rescorla-Wagner model, of cue competition description of, 47 revising of, 55-59 Resolution, as measure of relative accuracy description of, 245 experiments of, 237 test performance, 240-241 Retrieval blocks, 22 Retroactive interference, 2 in associative learning literature, 61 RSVP, see Rapid serial visual presentation
S Sequence effect, 89-90 Sexual conditioning, studies of species typical cues in quails using description of, 159-160 experiments, 159-161 learning to discriminate the sex of conspecific, 163-170 mechanisms of learning, 161-163 Pavlovian conditioning, 163 social proximity response, 160-163 Sign effect, 88-89 Species typical cues, for studying methodological integration of behavioral psychology and ethology conspecific as conditioned stimulus early use of, 158-159 sexual conditioning description of, 159-160 experiments, 159-161 learning to discriminate the sex of conspecific, 163-170 mechanisms of learning, 161-163
Index
Pavlovian conditioning, 163 social proximity response, 160163 potentiation of responding to description of, 175-176 significance of, 180-181 using conditioned contextual cues, 176-179 using punctate conditioned stimuli, 179-180 special properties of, as conditioned stimulus for sexual conditioning copulatory behavior conditioning, 170-173 overview of, 170 resistance to blocking effect, 174175 sensitization of response with or without female cues, 173 Spreading effect, 90-92 Stimulus generalization, 164 Stimulus-repetition effect, definition of, 1 Strategy adaptivity approach, to individual differences in arithmetic description of, 126-128 game show paradigm, 119-120 intrinsic versus extrinsic adaptivity, 128-131 retrieve versus compute, 117-118 strategy selection adaptivity in, 120-122 components of, 122-126 extrinsic adaptivity, 123-126, 128-131 intrinsic adaptivity, 123-124, 128-131 methods for studying, 118-119 awareness and, 138 in complex, dynamic tasks Kanfer-Ackerman Air Traffic Controller Task, 139-144 static versus dynamic tasks, 139 in controlled environments advantages of, 131 Building Sticks Task, 131-138 description of, 131 cross-talk stability, 150 description of, 117 importance of, 117
Index
Stress cool memory system description of, 202-204 hot memory system and, -mework of, implication of, for memory under stress low levels of stress, 206-207 moderate to high levels of stress, 207-210 traumatic stress, 210-213 definition of, 202 glucocorticoid release, 202-203 hot memory system cool memory system and, framework of, implication of, for memory under stress low levels of stress, 206-207 moderate to high levels of stress, 207-210 traumatic stress, 210-213 description of, 204-205 physiologic effects, 202-203 response to, factors that affect, 201
T Time preferences in health and money domains description of, 94-95 domain independence, 95 emotions, 97-99 familiarity with decision domains, 96-97 fungibility of health and money, 99-101 utility functions, 95-96 measures of, agreement among description of, 101 discount rates and, 103-108 future consequences, consideration of, 102
301
TIP-TAP theory, see Transfer-inappropriate processing-transfer-appropriate processing theory, of priming Transfer-appropriate processing circularity of, 3-4 definition of, 3 dissociations and, 5 episodic or persistence activation of abstract representations as cause of, 36 performance facilitatory effects, 33-34 in repetition blindness, 35 studies of, 3-8 Transfer-inappropriate processing behaviors before-disruption effect, see Beforedisruption effect inhibition of return, see Inhibition of return negative priming, see Negative priming repetition blindness, see Repetition blindness definition of, 2-3 episodic nature of, 36 performance inhibitory effects, 33-34 Transfer-inappropriate processing-transferappropriate processing theory, of priming, 33-36
W Word learning, conceptual development and, 253-255 Words, novel, effect on object categorization cross-cultural evidence, 268-271 description of, 259-263 meaning associated with, 279-283 object category formation, 271 Working memory capacity, for individual differences, 115-116
This Page Intentionally Left Blank
Volume 27 Deriving Categories to Achieve Goals Lawrence W. Barsalou Learning and Applying Category Knowledge in Unsupervised Domains John P. Clapper and Gordon H.Bower Spatial Mental Models Barbara Tversky Memory’s View of Space Timothy P. McNamara Made in Memory: Distortions in Recollection after Misleading Information Elizabeth F. Loftus Cognitive Processes and Mechanisms in Language Comprehension: The Structure Building Framework Morton Ann Gernsbacher Temporal Learning John E. R. Staddon and Jennifer J. Higa Behavior’s Time Peter R. Killeen Index
Volume 28 Conditioned Food Preferences Elizabeth D. Capaldi Classical Conditioning as an Adaptive Specialization: A Computational Model C. R. Gallistel Occasion Setting in Pavlovian Conditioning Peter C. Holland Pairings in Learning and Perception: Pavlovian Conditioning and Contingent Aftereffects Shepard Siege1 and Lorraine G. Allan
Evolutionary Memories, Emotional Processing, and the Emotional Disorders Susan Mineka Investigations of an Exemplar-Based Connectionist Model of Category Learning Robert M. Nosofsky and John K. Kruschke Reconstructing the Past: Category Effects in Estimation Janellen Huttenlocher and Larry V. Hedges Index
Volume 29 Introduction: A Coupling of Disciplines in Categorization Research Roman Taraban Models of Categorization and Category Learning W. K. Estes Three Principles for Models of Category Learning John K. Kruschke Exemplar Models and Weighted Cue Models in Category Learning Roman Taraban and Joaquin Marcos Palacios The Acquisition of Categories Marked by Multiple Probabilistic Cues Janet L. McDonald The Evolution of a Case-Based Computational Approach to Knowledge Representation, Classification, and Learning Ray Bareiss and Brian M. Slator 303
304
Contents of Recent Volumes
Integrating Theory and Data in Category Learning Raymond J. Mooney Categorization, Concept Learning, and Problem-Solving: A Unifying View Douglas Fisher and Jungsoon Park Yo0 Processing Biases, Knowledge, and Context in Category Formation Thomas B. Ward Categorization and Rule Induction in Clinical Diagnosis and Assessment Gregory H. Mumma A Rational Theory of Concepts Gregory L. Murphy Concept Structure and Category Boundaries Barbara C. Malt Non-Predicating Conceptual Combinations Edward J. Shoben Exploring Information about Concepts by Asking Questions Arthur C. Graesser, Mark C. Langston, and William B. Baggett Hidden Kind Classifications Edward Wilson Averill Is Cognition Categorization? Timothy J. van Gelder What Are Concepts? Issues of Representation and Ontology William F. Brewer Index
Volume 30 Perceptual Learning Felice Bedford A Rational-ConstructivistAccount of Early Learning about Numbers and Objects Rochel Gelman Remembering, Knowing, and Reconstructing the Past Henry L. Roediger 111, Mark A. Wheeler, and Suparna Rajaram The Long-Term Retention of Knowledge and Skills Alice F. Healy, Deborah M. Clawson, Danielle S. McNamara, William R. Marmie, Vivian I. Schneider, Timothy C. Rickard, Robert J. Crutcher, Cheri L. King, K. Anders Ericsson, and Lyle E. Bourne. Jr.
A comprehension-Based Approach to Learning and Understanding Walter Kintsch, Bruce K. Britton, Charles R. Fletcher, Eileen Kintsch, Suzanne M. Mannes, and Mitchell J. Nathan Separating Causal Laws from Causal Facts: Pressing the Limits of Statistical Relevance Patricia W. Cheng Categories, Hierarchies, and Induction Elizabeth F. Shipley Index
Volume 31 Associative Representations of Instrumental Contingencies Ruth M. Colwill A Behavioral Analysis of Concepts: Its Application to Pigeons and Children Edward A. Wasserman and Suzette L. Astley The Child’s Representation of Human Groups Lawrence A. Hirschfeld Diagnostic Reasoning and Medical Expertise Vimla L. Patel, Jose F. Arocha, and David R. Kaufman Object Shape, Object Name, and Object Kind Representation and Development Barbara Landau The Ontogeny of Part Representation in Object Concepts Philippe G. Schyns and Gregory L. Murphy Index
Volume 32 Cognitive Approaches to Judgment and Decision Making Reid Hastie and Nancy Pennington And Let Us Not Forget Memory: The Role of Memory Processes and Techniques in the Study of Judgment and Choice Elke U. Weber, Wiliam M. Goldstein, and Sema Barlas
Contents of Recent Volumes Content and Discontent: Indications and Implications of Domain Specificity in Preferential Decision Making William M. Goldstein and Elke U. Weber An Information Processing Perspective on Choice John W. Payne, James R. Bettman, Eric J. Johnson, and Mary Frances Luce Algebra and Process in the Modeling of Risky Choice Lola L. Lopes Utility Invariance Despite Labile Preferences Barbara A. Mellers, Elke U. Weber, Lisa D. Ordbilez, and Alan D. J. Cooke Compatibility in Cognition and Decision Eldar Shafir Processing Linguistic Probabilities: General Principles and Empirical Evidence David V. Budescu and Thomas S. Wallsten Compositional Anomalies in the Semantics of Evidence John M. Miyamoto, Richard Gonzalez, and Shihfen Tu Varieties of Confirmation Bias Joshua Klayman Index
Volume 33 Landmark-Based Spatial Memory in the Pigeon Ken Cheng The Acquisition and Structure of Emotional Response Categories Paula M. Niedenthal and Jamin B. Halberstadt Early Symbol Understanding and Use Judy S. DeLoache Mechanisms of Transition: Learning with a Helping Hand Susan Goldin-Meadow and Martha Wagner Alibali The Universal Word Identification Reflex Charles A. Perfetti and Sulan Zhang
305
Prospective Memory: Progress and Processes Mark A. McDaniel Looking for Transfer and Interference Nancy Pennington and Bob Rehder Index
Volume 34 Associative and Normative Models of Causal Induction: Reacting to versus Understanding Cause A. G. Baker, Robin A. Murphy, and FrCdCric VallCe-Tourangeau Knowledge-Based Causal Induction Michael R. Waldmann A Comparative Analysis of Negative Contingency Learning in Humans and Nonhumans Douglas A. Williams Animal Analogues of Causal Judgment Ralph R. Miller and Helena Matute Conditionalizing Causality Barbara A. Spellman Causation and Association Edward A. Wasserman, Shu-Fang Kao, Linda J. Van Hamme, Masayoshi Katagiri, and Michael E. Young Distinguishing Associative and Probabilistic Contrast Theories of Human Contingency Judgment David R. Shanks, Francisco J. Lopez, Richard J. Darby, and Anthony Dickinson A Causal-Power Theory of Focal Sets Patricia W. Cheng, Jooyong Park, Aaron S. Yarlas, and Keith J. Holyoak The Use of Intervening Variables in Causal Learning Jerome R. Busemeyer, Mark A. McDaniel, and Eunhee Byun Structural and Probabilistic Causality Judea Pearl Index
Volume 35 Distance and Location Processes in Memory for the Times of Past Events William J. Friedman
306
Contents of Recent Volumes
Verbal and Spatial Working Memory in Humans John Jonides, Patricia A. Reuter-Loren, Edward E. Smith, Edward Awh, Lisa L. Barnes, Maxwell Drain, Jennifer Glass, Erick J. Lauber, Andrea L. Patalano, and Eric H. Schumacher Memory for Asymmetric Events John T. Wixted and Deirdra H. Dougherty The Maintenance of a Complex Knowledge Base After Seventeen Years Marigold Linton Category Learning As Problem Solving Brian H. Ross Building A Coherent Conception of HIV Transmission: A New Approach to Aids Educations Terry Kit-fong Au and Laura F. Romo Spatial Effects in the Partial Report Paradigm: A Challenge for Theories of Visual Spatial Attention Gordon D. Logan and Claw Bundesen Structural Biases in Concept Learning: Influences from Multiple Functions Dorrit Billman Index
Volume 36 Learning to Bridge Between Perception and Cognition Robert L. Goldstone, Philippe G. Schyns, and Douglas L. Medin The Affordances of Perceptual Inquiry: Pictures Are Learned From the World, and What That Fact Might Mean About Perception Quite Generally Julian Hochberg Perceptual Learning of Alphanumeric-Like Characters Richard M. Shiffrin and Nancy Lightfoot Expertise in Object and Face Recognition James Tanaka and Isabel Gauthier Infant Speech Perception: Processing Characteristics, Representational Units, and the Learning of Words Peter D. Eimas Constraints on the Learning of Spatial Terms: A Computational Investigation Terry Regier
Learning to Talk About the Properties of Objects: A Network Model of the Development of Dimensions Linda B. Smith, Michael Gasser, and Catherine M. Sandhofer Self-organization, Plasticity, and Low-Level Visual Phenomena in a Laterally Connected Map Model of the Primary Visual Cortex Risto Mikkulainen, James A. Bednar, Yoonsuck Choe, and Joseph Sirosh Perceptual Learning From Cross-Modal Feedback Virginia R. de Sa and Dana H. Ballard Learning As Extraction of Low-Dimensional Representations Shimon Edelman and Nathan Intrator Index
Volume 37 Object-Based Reasoning Miriam Bassok Encoding Spatial Representation Through Nonvisually Guided Locomotion: Tests of Human Path Integration Roberta L. Klatzky, Jack M. Loomis, and Reginald G. Golledge Production, Evaluation, and Preservation of Experiences: Constructive Processing in Remembering and Performance Tasks Bruce W. A. Whittlesea Goals, Representations, and Strategies in a Concept Attainment Task: The EPAM Model Fernand Gobet, Howard Richman, Jim Staszewski, and Herbert A. Simon Attenuating Interference During Comprehension: The Role of Suppression Morton Ann Gernsbacher Cognitive Processes in Counterfactual Thinking About What Might Have Been Ruth M. J. Byrne Episodic Enhancement of Processing Fluency Michael E. J. Masson and Colin M. MacLeod At a Loss From Words: Verbal Overshadowing of Perceptual Memories Jonathan W. Schooler, Stephen M. Fiore, and Maria A. Brandimonte Index
