P. 1
Aw Aw

Aw Aw

|Views: 4|Likes:

More info:

Published by: Merima Jusupović Pašalić on May 24, 2012
Copyright:Attribution Non-commercial

Availability:

Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less

04/09/2014

pdf

text

original

Annual Reviews www.annualreviews.

org/aronline

Ann. Rev. Psycho1 1981 3Z:Zl-SZ . Copyright B 1981 by Annual Reviews Inc. All rights reserved

.

HUMAN LEARNING AND MEMORY
Annu. Rev. Psychol. 1981.32:21-52. Downloaded from arjournals.annualreviews.org by Columbia University on 02/08/06. For personal use only.

9339

Wayne A . Wickelgren

'

Department of Psychology. University of Oregon. Eugene. Oregon 97403

CONTENTS
MEMORY CODING...........................................

Associative Memory ....................................................................................................
Direct access Speci/ic node encoding ............. .......................................................... Type and token nodes .............................................................................................. Nodes and codes ...................................................................................................... Types of links Chunking and Vertical Associative Memory Concepts .................................................................................... Concepts and words ................................................................................................ Concepts and images .............................. ............................................... Constituent vs propositional and pmcedural meaning .................................................. Concepts vs semantic features ....................... ............................................... Fuzzy logic .............................................................................. Family resemblance ......................................................................................... Prototypes Abstraction ............................................................................................................ Basic concepts ........................................................................................................ Episodic and semantic memory ................................................................................

...................................

............................................................................................................

.......................................................................................................... ............................................................

22 22 22 24 25 26 21 28 28 28 29
29

.......................................................................................................

Schemata .................................................................................................................. MEMORY DYNAMICS ...............................................
Spcing

Units of semantic memory ........................................................................................ Constituent structure ....... .......... Plans and the Coding of .....................................................................

Propositions ................................................................................................................

.........................................................................................

Chunking

Storage: Consolidation and Forgetting

........................

......................................................................

30 32 33 33 34 35 36 36 36 36 37 37 37 38 38 39 40 41 42 44

'Preparation of this article was supported in part by NSF grant BNS78-08878 .

21

0066-4308/80/0201-0021$01.00

Annual Reviews www.annualreviews.org/aronline

22

WICKELGREN
Retrieval .................................................................................................................... Speed-accuracytradeofffunct~ns .............................................................................. What know? dowe .................................................................................................. Short-TermMeraory .................................................................................................. 44 44 45 46

Annu. Rev. Psychol. 1981.32:21-52. Downloaded from arjournals.annualreviews.org by Columbia University on 02/08/06. For personal use only.

Doyou admire the insights into the nature of the human mindthat have beenachievedby theoretical and empirical research on learning and memory? Moretimes than I care to remember,I have heard mycolleagues denigrate our understanding of human learning and memory. Many the of best known researchers in this field claim that weunderstandvery little about learning and memory--thatat best we have madesomeprogress in improving experimentalmethods studying learning and memory, our for or at worst that wehave merely been amusingourselves at the taxpayer’s expense. This pessimisticviewis totally andtragically wrong. persistence Its in the minds so many of eminent researchers in cognitive psychology serves, for these individuals and those wholisten to them, as a self-fulfilling prophecy.Because they believe it to be impossible,they do not strive for precise andgeneraltheories that accuratelyintegrate the welter of specific facts we knowregarding learning and memory that bestow underand standingas a consequence. overall purposeof this article is to demonThe strate that a very high level of integrative theoretical understanding of human learning andmemory possible today. Many is aspects of this theoretical understanding changein the future. Indeed,the paceof change will will quickenas a consequence taking the best present-day of theories seriously. However, future changesin our theoretical understanding learning the of andmemory be evolutionary, incorporatingthe wisdom the past with will of the wisdom the future, not revolutionary in the sense of ignoring the of wisdom the past for an ever changingsuccessionof fads. of

MEMORY CODING
Associative Memory Koffka(1935) proposeda nonassociative theory of human long-term memory accordingto whichthe mindlays down continuousrecord of experia ence (trace column), much like a videotape. Popular accounts often talk about memory this nonassociativeway, and evensophisticated cognitive in psychologistssometimes credenceto such nonassociative give theories. This is a shame,becauseone of the importantsuccessesof cognitive psychology is that wecan confidentlyassert that this nonassociative theory of LTM is false (Wiekelgren 1972a;1977a, pp. 233-47). DIRECT ACCESS a functional standpoint, the most critical defining From feature of an associative memory the capacity for direct access retrieval is of traces withoutsearch. In artifieal intelligence, associative memories are

Annual Reviews www.annualreviews.org/aronline HUMAN LEARNING AND MEMORY 23

often called content-addressable whenthey have this direct access property that the most common location-addressable computer memorieslack. In a location-addressable computer memory,one must serially search through all memory locations to find the one containing a stored (trace) pattern matching the input pattern. However,if the properties of an input pattern determine exactly where in the memory that pattern will be stored, then the memory content-addressable, that is, associative. Oneis directly addressis ing the contents of the memory location, not just addressing the location number. Present-day computers can be programmed to achieve a very limited degree of assoeiativity (direct access to information), but in the future, computers with large parallel processing capacity mayachieve associative memories closer to humancapacity (Fahlman 1979). The main argument against human memoryretrieval being a random serial search of a loeation-addressable store is the incredible speed that would be required-~on the order of ten million locations searched per second. Since synaptic transmission appears to be the most basic functional time unit in the brain, and that is on the order of 1 msecper synapse, it is plausible to assume that searching one memory location would require at least 1 msee. This yields an upper limit on search speed of 1000 locations per second, which is at least 10,000 times too slow. Anotherretrieval theory that can be rejected is the hybrid search/direct access theory that postulates direct access to a category of memory locations, e.g. those storing species of birds, followedby serial search at modest speeds (tens of milliseconds per location) through this muchsmaller set locations. Corbett & Wiekelgren (1978) obtained evidence contrary to the versions of this theory that have been proposed by Rosch (1973) and Rips et al (1973). Furthermore, it is very unparsimoniousto postulate a direct access process that does most of the retrieval work and then have someslow search process finish the job. If you can have fast, accurate direct access to all of the bird storage locations, you should be able to haveat least equally fast and accurate direct access to the hummingbird location(s). The most plausible alternative to direct access is the hierarchical search theory in which search occurs first for the correct high-level category of locations, then within this set for a correct subordinate category of locations, and so on until the correct individual location is searched for and found (e.g. Greenoet al 1978, pp. 24-27). If searching among categories memory locations was on the order of 10 msee per alternative category, then a seven-stage hierarchical serial search process with an average of ten alternatives at each stage wouldresult in the capacity to search ten billion locations in 700 msec. However,the information necessary to accomplish an hierarchical search process could be used more simply and efficiently to implementa direct access retrieval process.

Annu. Rev. Psychol. 1981.32:21-52. Downloaded from arjournals.annualreviews.org by Columbia University on 02/08/06. For personal use only.

Specific localization theories assume that almost all (e. Rev. The difference between specific localization and more global theories of mental representation concerns what percentages of the w entries in the vector for any particular idea are zero. image. For personal use only. a vector of weights between zero and one ( .org/aronline 24 WICKELGREN Annu. Call this set of elements the node representing the idea.. direct access retrieval systems (Fahlman 1979).g. Anderson 1973). etc) that we can represent in our minds.) representing the degree to which each element in the memory participates in representing that particular idea (e. word.. Retrieval speed considerations argue strongly that the basic human longterm memory retrieval process is direct access. from -1 to 1. Froma structural standpoint. determining the form constituents of a complexpicture). An analogous problem exists for grouping(parsing) spatially distributed input (e. segment. e. 1979a). one must either possess an independent means of accurately segmentingtemporally distributed input (which has proved to be an intractable problem for speech recognition by serial computers) or else try potentially thousands or millions of alternative parsings into phonemes and words of even short phrases.annualreviews. for example. faster discrimination of a large bright circle from a small dim circle than of either a large from a small circle or a bright from a dim circle (Biederman & Checkosky 1970)~argues compellingly for parallel processing of retrieval cues. concept. For every idea (feature." There must be at least one node representing any idea we have . segment..32:21-52. the features of the input signal are processed in parallel to intersect at the locations (nodes in memory that represent each feature. after Johannes Mtiller’s similar neurophysiological doctrine of specific nerve energies. the more one appreciates the power of parallel processing. 99% more) of the entries for a particular node are zero.g.annualreviews. Psychologically.g. The more one considers the difficulties of serial search retrieval processes for recognition of any of the tens of millions of things humans beings can recognize in a matter of seconds. stands for) the idea.g. direct access appears to require the property I call specific node encoding or specific element representation (Wickelgren 1977a.g.Annual Reviews www. 1981. Global theories assumefew zero entries. proposition. Downloaded from arjournals.. With any serial search process. concept. a node means nothing more than "whatever represents an idea. there is assumedto be a particular set of elements that represents (encodes.org by Columbia University on 02/08/06.wi . More generally. the phenomenon of redundancy gain~. SPECIFICNODE ENCODING have emphasized direct access as the deI fining property of an associative memory from a processing standpoint. that is. Oneway or to add inhibition to such a specific node encoding theory is to allow w i entries to have negative values as well as positive values. Psychol. With direct access. a node is a fuzzy set of elements. Furthermore. and proposition signaled by the input.

Downloaded from arjournals. but in somewaythat is physically distinguishable from a stronger version of one trace. 234-43). encodedin our minds. There may be some good physical reasons for such redundant multiplicity. which.org by Columbia University on 02/08/06. there is no conflict betweenmultiple trace theories and specific node encoding. The single vs multiple link or trace issue is theoretically important and not resolved. to overcome limited connection capacity of a single neuron. is an established property of human memory. as discussed in the prior section.org/aronline HUMAN LEARNING AND MEMORY 25 Annu. the others are retrieved at the same time. then from a cognitive standpoint. Rev. The preceding arguments against nonassociative memory theories apply to this ex- .32:21-52.annualreviews. The identity of the codes for the multiple encodings can be determined by search and comparisonof the codes found in the separated locations. by definition. WhatI shall call mental multiple node encoding is the encodingof an identical idea in separated (not associated) locations memory. but we can ignore this and assumespecific node encoding at a mental level. Psychol.annualreviews. So the only alternative to specific node encoding is multiple node encoding of each idea.so long as the same idea is always represented by the same node. pp. then there wouldbe a conflict with the principle of specific node encoding. typically because the idea was used to encode experiences or thoughts at two different points in time (usually in two different propositions or images). Mental multiple node encoding can be rejected because it lacks the capacity for direct access retrieval. This evidence rules out the simple tape recorder theory of humanlong-term memory. 1981. to provide protection against cell death. Still a third alternative is that repetition strengthens the sameseries of links. Physical vs mental specificity If every occurrence of the sameidea is encoded in a separate memory location. all of ihe physically separate nodes are functioning as a single mental node. or else. but instead strengthens somenewseries of links.Annual Reviews www. but all of these locations are directly accessible in parallel.There is also a great deal of evidence that the encodingof an idea that is identical or similar to a previously encoded idea has different consequences from the encoding of a new idea (Wickelgren 1977a. Multiple nodesvs multiple links It is possible that repetition of an association between two nodes does not strengthen the same physical links connecting these nodes. TYPEANDTOKEN NODES a new token node were created for every If occurrence of precisely the sameidea in a newcontext and all tokens of the same type were not directly accessible from any one of them. we could not think about that idea. However.which has mental multiple node encoding. so that wheneverone is retrieved. ete. For personal use only.

g. Using unique token nodes for the relational concepts in each proposition solves this associative interference problem. 1981. The same idea node can be incorporated as a constituent of manydifferent propositions (e. It is not in a node.the codefor an idea/~ a node (or set of nodes).annualreviews. in conformity with the principle of specific node encoding and permitting direct access to each idea because it has a unique location in memory. It is doubtful that manyof the inventors of such semantic memory systems think this is a desirable feature for artificial intelligence or a true feature of humansemantic memory. According to the principle. there is no eonfllct betweenthe prlnelple of specific node encoding and findings regarding frequency (Hintzman& Block 1971) and recency (Flexser & Bower1974)judgments that support the hypothesis that somewhat different traces are sometimesestablished by the separate occurrences of the sameitem in different contexts (different lists or positions).It is an undesirable necessity that the limitations of existing computersforce themto live with. Norman& Rumelhart 1975) use a new token node for every occurrence of any relational concept (signaled by verbs and. The higher-order nodes encode compound concepts and propositions by associations to and from their constituent nodes. Psychol.32:21-52.org/aronline 26 WICKELGREN treme overuse of token nodes. Old ideas are always encoded by the same nodes.org by Columbia University on 02/08/06.g. NODES AND CODES brief but important message about terminology A is the subject of this section.Annual Reviews www. However. prepositions) in a newproposition. For example. of specific node encoding. Rev. .annualreviews. Finally. For personal use only. The philosophy of this latter approach to the introduction of new nodes into associative memory that new nodes are is introduced only to encode some new idea. word occurred twice in the second list. Annu.g.a more elegant way to solve this problem is to use a hierarchical associative memory that introduces new nodes to stand for new propositions and can also encode nonpropositional compoundconcepts (e. Downloaded from arjournals. Sometheories of semantic memory(e. In an associative memory. encoding "John hit Frank" and "Peter hit Bill" by associations with the sametype node for "hit" results in a memory that cannot tell whether John hit Frank or Bill. The purpose of this is to encode clearly the set of concepts that constitutes each proposition and avoid some associative interference problems. a green square) by the same higher-order node mechanism(Anderson & Bower 1973). Manyartificial intelligence models of semantic memory have token nodes without direct access to all other token nodes of the same type simply because they are implemented on existing serial digital computers. word i occurred manytimes in the first list. etc). word i occurred near the beginning of the first list.there is no such thing as a code in a node.

Findler 1979). 220-25. Psychol. For personal use only. 233-51. Downloaded from arjournals. Anderson & Bower 1973. that it is associative. Links bond idea nodes together to encode more complex sets and se- Annu. but the issue concerns the specific type of associative memorywe have. In cognitive psyat chology.annualreviews. nodes in associative memory were usually assumed be connected by only one type of excitatory link (associto ation).g. Wickelgren (1965. Prior to Quillian’s (1966) thesis. 1977a. least since the early 1960s. namely. but. 1981. 11-22. and there has been very little theoretical use of inhibitory links amongmemory nodes outside of the perceptual and neural areas. TYPES LINKS OF Until quite recently.Annual Reviews www. Virtually all semanticmemory nets.annualreviews. However. but irrespective of the variety of link types.org by Columbia University on 02/08/06.g. not whether or not human memoryis characterized by specific node encoding and direct access retrieval.are considered by their inventors to be associative memorynetworks to a greater or lesser degree. Rumelhart. The number different types of links is an important and as yet unsettled of theoretical issue.and that a great truth has been established regarding howthe mind works. p.32:21-52. an ever increasing numberof associative network models of humanand artificial semantic memoryhave been invented. Lindsay Norman1972. which are widely accepted as the critical defining properties of an associative memory the in fields of computer science and semantic memory should be so considand ered in all of cognitive psychology. nobodydid anything with it.org/aronline HUMAN LEARNING AND MEMORY There is no point several different nonassociative to possible codes in 27 in having a code in a node unless you could have any of codes representing different ideas. Direct access (sometimes marching under the name of"content-addressable memory") has been recognized by computerscientists as the defining property of an associative memory. This makesthe memory a greater or lesser degree depending upon the variety of each node. people occasionally discussed the possibility labeled associations (different types of links). pp. Rev. pp. contingent uponthe degree of direct access capability. namely.the definition of the concept of associative memory. whetherdesignedto be artificial intelligence models or models of humanmemory. 1979a. Cognitive psychologyshould recognize that a major theoretical problemhas been largely solved. with the notable exception of inhibitory links in neural net theories. 1972a. Greenoet al 1978. . Anderson 1976. virtually all of whichuse a large variety of link types to express a large variety of relations between concepts (e.beginning with Quillian. 21) is behindthe times. Anyone asserts that a critical defining propwho erty of an associative memory the assumption of a single type of link is betweenidea nodes (e. 6-11) has repeatedly pointed out that the critical defining features of an associative memory specific node encoding and the direct are access retrieval such coding makes possible.

This is a remarkable accomplishment. Psychol. Anderson & Bower 1973.org by Columbia University on 02/08/06. Rev. not growingwith experience.annualreviews. Intuitively. Johnson 1972. "house" can refer to any particular house you ever experienced). we are not finding it ever more difficult to think with these concepts. like the electrovalent and covalent bondsthat hold atomstogether to form molecules or the four types of force that hold the nucleus together. One day we will knowhowmanydistinct types of bonds there are. this horizontal associative learning process is supplemented by a vertical associative learning process.g. Downloaded from arjournals. 1976a. new nodes are probably added by strengthening synaptic connections to neurons that have not previously been functionally connected to associative memory. A great deal morecan be created in the near future if psychologists have the will to do it. Dictionaries list several meaningsfor the typical word.the set of idea nodes is fixed after maturation is complete. though the anatomical connections might already exist prior to the chunking learning process. George Miller 0956) originated the concept of chunking. but does not add any new nodes to the memory (e. Froma physical standpoint.b.b. the newchunk nodes do not appear out of thin air. words cannot be the atomic Annu. 1981. In a vertical associative memory.Annual Reviews www. Chunking and Vertical Associative Memory In a horizontal (nonhierarchical) associative memory.b). For personal use only. Hebb1949).It is surely an important reason for the seemingly boundless potential of the human mind to understand ever more about anything. The modernword for any particular meaning of a word or phrase is "concept. for ever more complexcombinationsof simpler ideas. Learning changes the strengths of the links connecting these nodes to each other. 1979a. it is clear that as we acquire concepts. but each of these refers to a large family of different specific meanings. that adds new nodes to associative memory.annualreviews. . 1977a. new nodes have been added to memory by the chunkingprocess.org/aronline 28 WICKELGREN quences of ideas. chunking.g. From a psychological standpoint. Someof that beauty can be appreciated today if you have the eye for it. Estes 1972. specifying new chunk nodes to stand for combinations of old nodes.g. Concepts CONCEPTS AND WORDS typical word probably has thousands of A different meanings(e. Of course. Psychology has as much potential for intellectual beauty as any other science. The new higher-level concept nodes allow us to think about complex subject matter just about as easily and efficiently as we could previously think about simpler subject matter using lower-level concepts." Because words do not have unique meanings. Wickelgren 1969a. and its meaning was extended by manyothers (e.32:21-52.

at least for concrete concepts. e. a meow. that is. Constituent meaning is a generalized and more psychological analog of the philosophical notion of extensional (referential) meaning. but. For example. It is of no interest whether there are any "true" synonymsfor any given speaker of a language. CONSTITUENT VS PROPOSITIONAL AND PROCEDURAL MEANING The meaning of a concept is given partly by the constituent words and images that activate it from below and partly by the propositions and procedures of which it is a constituent. also from nonverbal stimulus cues. For personal use only. the reader may mistakenly retrieve the former concept to the second occurrence of the word. Downloaded from arjournals. Rev. not words. The atoms of semantic memoryare either concepts or semantic features. Synonymsprobably exist to minimize the short-term memory interference problemthat derives from our et~icient use of tens of thousands of words to refer to millions of concepts. The constituents of concept nodes are the chunk nodes for the words and images that cue them. consider the phenomenon of synonymity (Herrmann1978). Understanding the relation between concepts and words and that concepts. 1981. Wordsare high-level structural units representing ordered sets of phonetic and graphic segments. the feel of fur. Otherwise. The interesting point is that any given concept can be expressed by so manydifferent words or phrases. are the atoms of semantic memory clarifies thinking about many problems. What should in you do in writing a paragraph whenyou have used the same word to refer to two different concepts? Use a synonym place of one of the occurrences in of the word. . If there are any. they are rare.org/aronline HUMAN LEARNING AND MEMORY 29 Annu.Annual Reviews www. elements of semantic memory. What could be the purpose of such duplication? One plausible answer derives from the multiplicity of concepts associated with each word and consideration of a likely role of short-term memory understanding speech and text. as cues to activate etc the eat concept.org by Columbia University on 02/08/06. which can activate it from above (Woods 1975. The combination of features that make up each of these cues constitutes an image. CONCEPTS IMAGES AND Concept nodes not only receive input from words.annualreviews. Propositional and procedural meaning are equivalent to what linguistic philosophers refer to as intensional meaning. Both are essential componentsof the meaning of concepts.32:21-52. Wickelgren 1979a). pairs of words (or phrases) which are associated to exactly the same set of concepts. Psychol.annualreviews.g. It seems clear and accurate to regard concept nodes as the first level of nodes that integrate verbal and nonverbalstimuli. because its strength of association to the word was temporarily increased by the prior pairing of word and concept. the shape of a cat.

Downloaded from arjournals. McCloskey Glucksberg (1979) suggest & just such a model. Psychol.. (1973). "Each word is represented by a set of attributes. 1970)..org/aronline 30 WICKELGREN CONCEPTS SEMANTIC VS FEATURES two principal theoretical The approaches to semantic memoryare the associative network theory used throughout this article and the semantic feature theory developed by Katz & Fodor (1963).Annual Reviews www.. Somenetwork theories have had trouble explaining whyhigh similarity trues are verified faster than low similarity trues.. Meyer (1970). although the mathematical formulation leaves a bit to be desired. which serve as the constituents of higher-level coneepts and/or the first concepts learned by children? (b) Are concepts other than semantic feature concepts represented by unitary nodes at all.. describes these semantic feature models. initiated in parallel by these tasks. a particular size. each with a characteristic asymptotic strength.that is. the lowest level concepts. 1981. but high similarity falses are contradicted more slowly than low similarity falses. If one assumesthat there are two processes.g. almost all of the semantic features anyone has discussed would be considered concepts in semantic memory. e.. a chickenis a bird (low similarity true).org by Columbia University on 02/08/06.. and Smith et al (1974). Schaeffer & Wallace (1969. The assumptionthat is rejected by these data is that semantic judgment times depend entirely or primarily upon the numberof links separating the concepts whose relation is being judged. However. can doubtless account for the recent antonymjudgment results as well (Glass et al 1979. and a rat is a (low similarity false). and as characteristic features. Herrmann al 1979). a robin is a bird (high feature similarity true). called semantic features . . For personal use only. a bat is a bird (high similarity false). or are Annu.From a coding standpoint there are two issues: (a) Are semantic feature concepts the most basic concepts of semantic memory." The experimental testing of semantic feature theories has focused on verification and contradiction of category-example relations between concepts. Such a two-process retrieval assumption. feature theory can et a be easily married to the sametwo-process retrieval theory and account just as well for these results. then network theories account for category-example judgment findings better than existing feature theories (e.annualreviews.g. ’bird’ wouldinclude as defining features.32:21-52.. Rev.annualreviews. What can be concluded concerning semantic features vs concept networks? First.No reasonable person could deny that there are nodes encoding these features somewhere in memory. verification and contradiction. married to the associative network theory. animate and feathered. As Smith (1978." "Robin" has manyfeatures in common with "bird" plus such features as "red-breasted. the results of Holyoak& Glass 1975 and Corbett &Wickelgren 1978)... and that judgments are made whenthe difference in retrieved strength of the two proeesses exceeds some critical value. Indeed. Rips et al (1973).

The research of Roschet al (1976) and Rosch(1977) makesit quite clear that these very general superordinate categories are not the basic level of concepts in adult semantic memory. they are best answeredin the opposite order. white. Rejection of the semantic feature theory in favor of the more general concept-node-and-link network theory does not mean that the general concepts referred to as "semantic features" do not exist in semantic memory. Semantic memory just where we need chunking the is most to avoid the enormous associative interference problem that would result from associating concepts only by associating their featural constituents. these general concepts are probably coded more complexly and are certainly acquired later than concepts of intermediate referential generality. . on the average. 500) have already pointed out this fundamental contradiction to the semantic feature theory. living. pp. Althoughthe abovetwo questions are best posed in the order I used. Clark & Clark (1977.Annual Reviews www. They surely do. feathered.g. or dog (see also Piaget 1954. there is probably a morespecific reason for the persistence of this inadequate semantic feature theory. retrieval of information by experts wouldbe less efficient than by amateurs and retrieval by adults would be less efficient than by children. These attributes are part of the constituent meaningof a concept. However. If the mind represented concepts only as sets of semantic features. animate. 1981.org by Columbia University on 02/08/06. each of which participates in thousands of other concepts and so would have thousands of competing associations. There is a great deal of evidence supporting the chunking capacity of the humanmind.org/aronline HUMAN LEARNING AND MEMORY 31 these concepts represented only as sets of semantic feature nodes?I believe that we can draw firm conclusions on both of these issues. etc. are they typically nor the first concepts learned by children. With respect to whether semantic features are the basic concepts of semantic memory. Besides the general failure of psychologists to try to decide theoretical issues. e. 126-32). Some these alleged semantic features of refer to sensory attributes of the stimuli that activate concepts from below.annualreviews. male. Indeed. or collie. delicious apple. Neither are the basic concepts highly specific (narrow referential generality). and they are certainly coded at levels below the concepts they activate in semantic Annu. de Villiers &de Villiers 1978. such as physical object. apple. As Roschdemonstrates. contrary to fact.32:21-52. answer is also clear. Rev. For personal use only. hammer.annualreviews. Downloaded from arjournals. Psychol. Semanticfeatures the are necessarily superordinate concepts with wide referential generality (large sets of examples). it wouldbe peculiar indeed if semantic memory did not. They are not.they do not appear to be more simply coded nor developmentally prior to other concepts. p. the basic concepts are of intermediate referential generality. Anglin1977. If other modalities of memory form new chunk nodes to represent combinations of moreelementary constituents. as discussed earlier. ball peen hammer. Brown 1958. namely.

org/aronline 32 WICKELGREN Annu. Werepresent red at several levels of codingin the mind. What more appropriate way to bury the semantic feature theory than to take away part of its name? Fuzzy logic (Zadeh 1965) is a generalization of standard logic that is more applicable to humanthinking. Fuzzy logic can express various relations between different concepts. namely. Rev. that the degree of membership an of event in the intersection of two categories is the minimum the degree of of membership either constituent category.They reject = the standard minimum rule for fuzzy intersection.annualreviews. Wickelgren 1979a. Instead of assuming that the degree of membershipof some instance in any concept (category) limited to the extreme values zero or one. such sensory features and semantic features are very different beasts. more generally. For example.org by Columbia University on 02/08/06. Consistent terminology might help. Downloaded from arjournals. Psychol. the concept red is activated by the word "red" as well as by red light.AB). Zadeh’s invention of fuzzy logic is an important intellectual accomplishment both for artificial intelligence and for cognitive psychology. that also include classical logic as a special case.namely. lising the example of GRUE law = GREEN BLUE.Failure to realize this causes confusion. using the example ORANGE RED AND YELLOW. especially if one relaxes the restriction that fuzzy logic be a simple extension of classical logic. Whateverthe value of classical logic to mathematics.32:21-52. For personal use only. I propose that we use the otherwise synonymous terms "feature" and "attribute" differentially to help remind us of the distinction between sensory features and conceptual attributes. However. In the second place.However. most of the alleged semanticfeatures are muchtoo abstract to be considered elementary sensory features.it seems to me that in describing the mind we ought not to restrict ourselves to logics that include classical logic. However. including it as a special case. 1981. 360-68). fuzzy logic assumes any degree between zero and one.Annual Reviews www. that the degree of membership in the union OR equals the maximum the degree of membership in either constituent of category.annualreviews. Fuzzy logic corrects one of the flaws of classical logic as theory of humanthought: the restriction to two-valued set membership FUZZY LOGIC . pp. Kay & McDaniel(1978) discuss a couple of different alternative rules for fuzzy intersection (C = A AND B). The concept red is not coded in the lower levels of the visual system where the sensory feature red is first encoded. They also discuss the standard in maximum for fuzzy union (C = A ORB). between any idea nodes in the mind. memory. is a multifacetedfailure as a description it of the "laws of thought" (Wason&Johnson-Laird 1972. there are a great variety of possible fuzzy logics yet to be invented. Bellman & Zadeh (1970) suggest a multiplicative rule for intersection (C = AB)and an additive rule for union (C = A + B . In the first place.

this is almost certainly not what we do. speech. Oden 1979). However. Odenhas done a numberof other pioneering studies on the application of fuzzy logic to semantic memory. see the superb application of fuzzy logic to speech recognition by Oden& Massaro(1978).g.and the strongest association will win most of the time. but I am hopeful this will change. Reed 1972). rather than being satisfactory as they stand. we say that the cue sets for a given concept have a family resemblance to each other. Rev. the prototype hypothesis in no way depends upon whether a real example of a concept or possible cue set for a concept actually matchesthe prototype. FollowingWittgenstein (1953). functions. which uses the multiplicative rule for combiningthe featural constituents of a speech segment after power function weighting of each featural constituent. The weights can be considered to be the associative strength of each feature to each segment. The various fuzzy logical operators that have been suggested are interesting to consider. probably as a point of departure. Psychol. Downloaded from arjournals. but there are many attributes that appear frequently in these cue sets.org/aronline HUMANLEARNING AND MEMORY 33 Annu. PROTOTYPES prototype hypothesis is that a concept can be characThe terized by an ideal set of attributes (Evans1967. which is a very strong point in its favor. For a striking exception.annualreviews. any cue set will have somestrength of association to every concept node in memory.org by Columbia University on 02/08/06. the prototype is an ideal that can never be experienced at one time. Instead. In any case. and reading (e. subject to randomerror. propositional. only a tiny subset of all the characteristic attributes of a conceptwill be sufficient to cause us to think of that concept.annualreviews. Posner &Keele 1968. What does the prototype hypothesis add to our characterization of concepts as family resemblances with fuzzy boundaries from other concepts? This is not entirely clear (Neumann 1977). Thus. most psychological applications of fuzzy set theory by cognitive psychologists have been banal or vague.32:21-52. it is probablycorrect to assert that for most natural concepts. FAMILY RESEMBLANCE Frequently. So far. will contain all of the ideal attributes.Annual Reviews www. fuzzy logic is completely compatible with the assumptionof graded strengths of associations betweennodes in the mind. The most compellingreason to prefer this associa- . then clearly no single exampleof most concepts. If it were true that activating a concept from a cue set involved matching that cue set to the ideal set (prototype) for each concept (Reed 1972). and procedural meaningas constituting the entire set of ideal attributes characteristic of a concept. For personal use only. Thereis nothing common all of the sufficient cue sets for a given to concept (Wickelgren 1969a). and certainly no single cue set. If we include both verbal and nonverbal constituent. then the prototype hypothesis would have importance. In general. 1981.

propositional. chairs are to sit on)." but wouldbe quite inadequate to learn superordinate concepts such as "fruit" and "mammal" that do not cluster about a single prototype. this must be the sort of process they have in mind. Psychol. A third reason to prefer the strength hypothesis derives from consideration of the abstraction process by which concepts are learned. the more adequate instance strength abstraction mechanismis also more natural for an associative memory.annualreviews. A second reason to prefer the strength hypothesis is that a theory with predictions that are probably very similar to it. and procedural meaning of concepts. ABSTRACTION Abstraction is the process by which the input links to a chunk node from other nodes are selectively strengthened and weakened through repeated experience with different instances of the concept. while the prototype abstraction mechanism would be much more complex to implement. Furthermore. Recently.g. An associative computer could be built to do this. the context theory of Medin & Schaffer (1978). These input links should include the specification of constituent. Rev. Such an average prototype abstraction mechanism could learn concepts whose examples cluster about a single prototype.org/aronline 34 WICKELGREN tive strength hypothesis to the prototype matching hypothesis is that the strength hypothesis is exactly what an associative memory with direct access predicts. since the relations between concepts often appear to be learned very early (e. This strength learning hypothesis is both reasone.Annual Reviews www.32:21-52.org by Columbia University on 02/08/06. Annu.annualreviews.ble and consistent with associative network theory. evidence has accumulated to support the hypothesis that for adults the best example of a concept is sometimes composedof the modal (most frequently occurring) values on each attribute dimension rather than the average values (Goldman& Homa1977. This permits determining the average value and the variance on each feature dimension. For example. For personal use only. Whenpeople speak of storing the average prototype without storing the examples. such as "apple" or "cat. Downloaded from arjournals. 1981. and add one to a number-ofinstances counter. though so far infants have been demonstrated to abstract only averages . Neumann 1977. has been shown by them account better for transfer from concept learning to classification of new instances than the independent cue type of prototype matching theory. It would be highly unparsimonious to opt for some prototype learning theory that cannot be incorporated easily within the more general associative network theory. but it is a less accurate wayto record the regions in a multidimensional attribute space that are associated with any given concept than the strength mechanismpreviously described. namely. add its square to another counter. one way to store an average prototype without storing examplesis to add the value of each attribute dimension to a counter. Strauss 1979).

pp. Rosch 1977) that we understood this phenomenonadequately. Downloaded from arjournals. For personal use only. Psychol. In addition. there is increasing acceptanceof the hypothesis that the learning of a concept derives from encoding each instance more or less faithfully. Whether adults abstract modesor averages depends on the discriminability of the values on a dimension--highly discriminable values cause the mode have the greatest strength of association to the concept.annualreviews. subject to selective attention in learning and to the interference and facilitation in learning. Brown suggested that the reason for this was greater frequency of use of these concepts by other humansin the child’s environment. This high degree of examplesimilarity meansextensive strength generalization in the .32:21-52. According to Rosch. subsequent research has confirmed this conclusion. (Strauss 1979). BASICCONCEPTS it hadn’t been so sad it would have been funny to If hear psychologists at one time asserting that the earliest concepts learned by children were the most general ones followed by increasing conceptual differentiation. storage. inhibition is also involved in some mannerso that the attributes that have the strongest association to a concept are those that discriminate best between this concept and other concepts (Rosch &Mervis 1975). basic concepts are the most general concepts whoseexamplesare sufficiently similar that they hover around a single prototype. p. Brown(1958) pointed out that neither extreme was true and that the earliest concepts learned by children are of intermediate referential generality. to myknowledge.annualreviews. and retrieval that inevitably results from encoding in an associative memory. 310) point out. Rev.Annual Reviews www. and at other times asserting that the earliest learned concepts were the most specific ones followed by concepts of increasingly greater generality.explains the abstraction of averages whendiscriminability is low and modeswhendiscdminability is high. as discussed earlier. this can be accounted for by old-fashioned stimulus generalization in an associative networktheory. See also Wickelgren(1979a. As Neumann (1977) and Wickelgren (1979a. to less discriminable values cause the average value to have the greatest strength of association to the concept. and. Noother unified theory.org/aronline HUMAN LEARNING AND MEMORY 35 Annu.The most plausible theory at present is that the abstracted constituent structure of a conceptis the result of accumulatedassociation of the attributes of events that cued that concept. Starting most recently with Brooks(1978).it was not until the elegant insight of Rosch(Rosch et al 1976. 311-12) for an associative explanation of this using the Spencianoverlapping excitatory and inhibitory generalization gradients mechanism. However. 1981. and this is surely an important component the explanation for why of concepts like "cat" and "apple" are basic instead of "mammal" "fruit" or on the one hand or "Siamese cat" or "Jonathan apple" on the other.org by Columbia University on 02/08/06.

uniqueexperiences will generally be forgotten unless they are frequently recalled. youhavethe best extant explanation of the relation betweenepisodic and semantic (generic) memory. Cunningham (1978) . cognitivelycorrect definition of a propositionis not known. in effect. Propositions UNITS SEMANTIC OF MEMORY A proposition contains a relational concept (verb) and one or moreargumentconcepts (agent. peoplehave but a very high degreeof agreement concerningthe analysis of sentences into component propositions. At the sametime.annualreviews. When are they recalled. integrative learning of separate examples. there is no episodic memory. 321-25). For personal use only.Annual Reviews www. object. Although probablydoes not apply it to networkswith redundantlink paths betweennodes. pp.32:21-52. Unique experienceswill by definition not be merged with subsequent experiences via the abstraction process. Wickelgren 1979a. Asa consequence. are this explanation makesit clear howexceptions could occur whena child has a great deal of experiencewith a moregeneral or less general concept than the corresponding level concept. pp.annualreviews.exceptfor the first recall of a uniqueexperience. it is parsimonious assume of to that propositions are fuzzy combinations their conceptconstituents as well (Oden of 1978.org/aronline 36 WICKELGREN Annu. CONSTITUENT STRUCTURE A proposition is a set of concepts. Thus. etc).I havetakena little license with basic Rosch’s idea to make compatible it with the associative-strength. increases the frequencyof pairing each attribute node with the concept node. Hence these concepts typically learnedfirst. which.the basic level conceptsare those wherethe two factors contributing to associative strength. they become subject to integration and abstraction with the recall experience. instancelearning theory of abstraction advocated the prior section.org by Columbia University on 02/08/06. so it has beenpossible to establish that propositions are units of codingabove concepts in semantic memory (Wickelgren 1979a. Thereis no reasonto believe that these are two different forms of memory with either different coding structure or different cognitive processes. Psychol. example frequency example and similarity. Given that concept nodes and phonetic and graphic segment nodes are fuzzy combinations their constituents. in EPISODIC ory AND SEMANTIC MEMORYIf you accept the instance mem- theoryof abstraction presentedearlier. but what kind of set is unclearat the presenttime. 317-21). 1981. Rev. only various degrees of generic memory. typically hit the maximum their combinedeffect on speed of concept in learning. Neitherpredicatenor case constituent structure ha~ proved completelysatisfactory (Dosher 1976). Downloaded from arjournals. Onthe average.

Wickelgren 1979a). MEMORY DYNAMICS Annu. Anderson 1976). Mytheory is that propositions are unordered sets of concepts. production systems (Newell 1973.Annual Reviews www.Another basic and obvious dis- . storage (consolidation and has forgetting). While propositions can encode all knowledge. 357-60. pp. this knowledgeis not in the proper form for the temporal control of human action (by which I meanall mental actions. 1976c. 1981. As a theory of the coding of order in phonetic and graphic segments in words. 325-27).1979a). For personal use only.annualreviews. Widespread understanding and acceptanceof the learning. Memory three temporal phases: learning.A more parsimoniousalternative is that a schemais a mental set of primed(partially activated) nodes (and links?) that results automatically from spreading activation from the activation of nodes representing prior thoughts (Wickelgren 1979a. Rev. Psychol.c. partially ordered) sets of concepts. Downloaded from arjournals. 1979c). Primingvia spreading activation is a natural part of an associtive memory.annualreviews. storage. pp.org by Columbia University on 02/08/06. including knowledgeabout actions and the temporal sequence of events. but it is far too early to concludeanythingabout its application to the encoding of syntax and other higher-order procedural knowledge. and retrieval distinction has greatly facilitated scientific research on memory. while plans are ordered (most generally. It maynot be useful to distinguish these two types of knowledge. although they could exist. not just those mental actions that directly control obvious physical actions by the muscles).org/aronline HUMANLEARNING AND MEMORY 37 has devised a very interesting scaling method for determining network structure from dissimilarity data. Schemata Somepeople believe that there is a level in the mind abovepropositions and plans that integrates sets of these into single units called schemata. There is no evidence supporting the hypothesis of unitary schema nodes.Plans are units of procedural knowledge. Plans have been described using augmented transition networks (Woods 1970). context-sensitive coding is incomparably superior to any other theory (Wickelgren 1969b. I currently believe is. there is considerable evidence supporting such and a hypothesis (Collins & Loftus 1975. and context-sensitive coding (Wickelgren 1979a. Plans achieve this control of sequences of mental actions contingent upon internal and external events. and retrieval (recall or recognition). 1972b. Plans and the Coding of Order Propositions are units of declarative knowledge. but in agreement with Anderson(1976).32:21-52.

Mierostorage is concerned with the subsequent consolidation and forgetting of such small atoms or molecules of learned information.annualreviews. but problemsolving. forgotten. as in free or ordered recall of lists or text. This distinction is analogous to that between macroeconomics and mieroeconomies. wherethere is little control over the control processes. pp. but no evidence compels this assumption.2 to 2 see depending upon the coding level (Wiekelgren 1979a. Oneelementary act of microretrieval takes about one second.org by Columbia University on 02/08/06. this is a complex retrieval process consisting of a sequence of elementary acts of recall and recognition. Learning REPETITION "Practice makes perfect"--or.32:21-52. it makes memory traces stronger. Basically. and retrieval. Rev. integrating the to-be- . minutes. proposition). Microretrieval is concernedwith a single elementary act of recalling or recognizing someunit of information (e. Downloaded from arjournals. one elementary act of microretrieval can be of a very complex chunk of information with numerous constituents. I tend to ignore all multiple recall (especially free recall) studies. 270-78). storage.org/aronline 38 WICKELGREN Annu. word. that between the macro and micro levels in the cognitive analysis of learning. There do appear to be sudden jumps in strength during learning. It is still possible that memory traces are composed of varying numbers of component traces each of which is learned. If the retrieval is extended over tens of seconds. creativity.Annual Reviews www. under manyconditions. because of chunking. it is simpler to characterize traces by continuous strengths. and perhaps someinference processes as well. and comprehension of large units of text are at the macrolevel. within the attention span.annualreviews. and recognition operations is at the micro level. varying from . if the retrieval takes place in a couple of seconds or less.g. The study of single inference operations or small sets or sequences of inference. you are studying microretrieval. and learning curves are almost always continuous incremental functions of study time. For personal use only. let us consider the retrieval process in more detail. To study memorydynamics independently from coding. and retrieval. which probably result from insights regarding good ways to code material (e. and/or retrieved in an all-or-none manner. Accordingly. in any case. retrieval cues. several separate (unchunked) traces can be retrieved in parallel. Psychol. or longer. storage. To explicate the distinction. given that forgetting and retrieval functions are always continuous incremental functions of storage and retrieval time.g. and retrieval time for each elementary act of retrieval. consolidated. Such tasks are studying macroretrieval. 1981. recall. This review is only concernedwith the micro level of learning. tinction is sorely needed. Fortunately. concept. Furthermore. Microlearning is concerned with the learning of a single association or a small set of associations encoding a single chunkor a small set of chunks.

Oneof the persistent problemswith almost all research on spacing effects is the failure to separately assess the effects of spacing on learning and retention. For personal use only. The increment to the memory trace contributed by second-trial learning can be estimated by subtracting the memory strength due to the first learning trial at the appropriate retention interval from the total learning strength obtained with two learning trials. contrary to the assertion that increased learning results only from adding traces at higher levels of processing.org/aronline HUMAN LEARNING AND MEMORY 39 Annu.org by Columbia University on 02/08/06.32:21-52. even with the same question each time (does the wordhave an r sound). in morestandard learning tasks. Reed 1977). This can only be done by obtaining retention functions for the single-trial learning condition and for each two-trial learning condition (Wickelgren 1972c. the end product of these processes can be activated without all of the preliminary coding required by first or spaced-second trials. not only can the second trial learning increment be . With a massedsecond trial.annualreviews. Wiekelgren 1977a. can classify theories of the spacing effect into We three categories: encodingvariability. Further support for the incremental nature of memory traces comes from Nelson (1977). Jacoby suggests that subjects do not go through the same recognition and other coding processes for a massedsecondtrial as for a first trial or a spaced secondtrial. learned material with related already-learned material).annualreviews. This can be done at a variety of retention intervals following the second learning trial. the levels (types) of coding and processing maybe identical. However. producedincreased learning. and since then an ingenious study by Ross & Landauer (1978) provides more evidence against all encoding varability explanations. whofound that repetition at the same phonetic level of coding. p. 1981.Annual Reviews www. This deficient coding hypothesisis doubtless the explanation of Jacoby’s results. but the speed of achieving the encoding is much faster for massedsecond trials. If increases in degree of learning are closely tied to the encoding process and not to the period of maintained activation after encoding. as shownby increases in strength ratings and in retrieval speed (Corbett 1977. 320). SPACING Spaced repetitions almost always benefit memorymore than massed repetitions. A novel experiment by Jacoby (1978) further supports the deficient second-trial learning hypothesis. Thus. Psychol. and deficient learning of the second presentation. then faster encoding maynecessarily produce less learning. Progress has been made in understanding the reasons for this phenomenon. even when a single jump in learning boosts retrieval from chance to 100% accuracy. However. deficientconsolidation or rehearsal of the first presentation. subsequent learning will further increase strength. which were obtained using an unusual learning task. Hintzman(1976) argued that the first two categories of theories could ruled out on the basis of prior evidence. Rev. Downloaded from arjournals.

This can produce advantages for shorter over longer spacings. perhaps indefinitely. 1981. retention of the second-trial trace is at least as good as that of the first trial (Wickelgren1972c and unpublished data). For example. course. Phonetic traces are what will help you perform best on a phonetic .annualreviews. Semantic processing by itself is no guarantee of a high level of learning. the concept of levels Of of coding (and processing) existed long before the fad started and will continue to be an essential. deciding that an ostrich is not a geographical location results in very little memory for having processed "ostrich. but the trace code that results from this processing. All of this tends to producea beneficial effect of greater spacing between learning trials under manycircumstances. Retention function analyses of the spacing effect produce the following conclusions: (a) Second-trial learning increases monotonically with spacing. estimated. (c) Consolidation and forgetting of the second-trial’s trace is probably identical to that for the first trial." compared deciding that an ostrich is a living thing to (Schulman 1971). One of the criticisms of the levels fad was that it failed to add to the precise theoretical specification of the levels concept in either coding or processing domains. It is foolish to do research on such a complex quantitative problem as the spacing effect without a mathematical modelingapproach and without obtaining complete retention functions. Rev.32:21-52. Second. concept for understandingthe sti-ucture of the mind. First. Another flaw with the faddish conception was that its specific claims about learning and memory turned out to be almost entirely false. high levels of performance on memory tests are not guaranteed by a semantic coding level either. This makesthe assumptionthat the second trial does not destroy or alter in any way the memory trace resulting from the first trial. However. thoughnot yet completelyspecified.org/aronline 40 WICKELGREN Annu. Downloaded from arjournals. certainly. though the maximum increases are achieved during the first 10 to 30 minutes after first-trial learning 0Vickelgren 1972eand unpublished data). Hintzman 1976). (b) The memory trace for the first trial is probably unaffected by second trial learning (Wiekelgren 1972c.annualreviews. but also the retention function for the second-trial learning increment can be determined for comparison across spacing conditions and to the first-trial retention function. LEVELS CODING OF AND PROCESSING levels of processing fad is The over in the field of learning and memory. The evidence reviewed by Hintzman(1976) and the results of Wickelgren (1972c) are in agreement with this assumption.Annual Reviews www. For personal use only. Whatcounts most is the relation between the information stored and the information questioned in retrieval.shorter spacings have the advantageof shorter retention intervals for the first learning trial at any given retention interval following the secondtrial. Psychol.org by Columbia University on 02/08/06. it is clearly not the level of processing that matters for degree of learning and retention.

etc (Morals. so the former conception appears to be left with only the multiple retrieval cues explanation for the effectiveness of elaboration-distinctiveness.g. I believe it left a useful legacy of facts and ideas generated in its disproof. as mentionedearlier. Friedman & Bourne 1978. King & Anderson 1976. Choosing between these alternatives is difficult. (b) Traces whoseconstituents have few interfering associations to other traces are more distinctive and thus remembered better because they are less subject to interference. Bower1978) that learning multiple associations to the same nodes causes interference. ELABORATION DISTINCTIVE/qESS AND Two other explanations of the variability in learning are the elaboration hypothesis (Craik &Tulving 1975. repetition of processing at lower structural levels further increases degree of learning. which is surely a critically important concept in understanding the mind. the levels of processing frameworkhas been largely abandoned. Third. Bransford & Franks 1977.despite the fact that embeddingthe material in a sentence or image. At least. 1981.org by Columbia University on 02/08/06. However. For personal use only. retrieval test. McDaniel.annualreviews. Jacoby & Craik 1979).The distinctiveness hypothesis has two principal variants. Stein 1978). For all these reasons. Anderson& Reder 1979) and the distinctiveness hypothesis (Klein &Saltz 1976. despite the plausibility of the hypothesis that ending is more distinctive at semantic than at structural levels and thus less subject to interference. and it generated interest in levels of coding and processing. Psychol. Rev. The elaboration hypothesis is that traces with manyconstituents or many associations to other traces are well learned and well remembered. nowno knowledgeable cognitive psychologist should have trouble explaining to students what it means to read a page of words and suddenly realize that you haven’t been processing the meaningsat all. which are complete opposites.Annual Reviews www.32:21-52. substantial learning takes place at lower structural levels of ending and.annualreviews.org/aronline HUMANLEARNING AND MEMORY 41 Annu. Anderson 1974. ]3ysenck 1979. Downloaded from arjournals. Onthe other side is the evidence that elaboration often aids learning and memory. though no one seems to have realized this before: (a) Elaborate traces are more distinctive and therefore have more effective possible retrieval cues and are also less subject to interference in storage and retrieval. relating the new material to schemata from existing . forgetting of many structural traces is not especially rapid and often there has been no difference in forgetting rate betweenstructural and semantic traces (Nelson & Vining 1978). it is not necessaryto process at semantic levels to achieve further increases in degree of learning (Nelson 1977). Wickelgren1977a. The latter conception of what makesa trace less subject to interference is clearly the correct one. semantic on a closely related semantic test. Onthe one side is all the evidence from decades of verbal learning research and more recently from semantic memoryresearch (e. Fourth.

Elaboration that merely adds to what must be learned does not increase the memorability of the other material and often decreases it. Wickelgren 1972c). but that multiple interactive images producedno better memory than a single interactive image. Downloaded from arjournals. when everything in some input material is already encoded in memory. Nelson et al found that interactive images were superior to noninteractive images for learning word pairs. Wickelgren Norman1971. 119-20). To add to the problemswith the elaboration hypothesis. Nickerson & Adams(1979) discovered that people have extraordinarily poor recall and recognition memory the visual details of a very familiar for object.org/aronline 42 WlCKELGREN Annu. it would be wrong to go overboard on what could Of be called the reduction hypothesis.but these results demonstrate that there is no huge benefit to redundant multiple associations connecting the same nodes. Elaboration that increases the amount of the other material that can be encoded using already learned schematais beneficial to learning. Psychol.Annual Reviews www.annualreviews. Learning curves always indicate diminishing returns after an earlier period of positive accderation. knowledge. CHUNKING course. etc. that wc learn best whenwc have the least to add to what is already in memory. This meansthat elaboration of the stimulus input maybe benefitting memory because it is requiring a less elaborate addition to memory than nominally simpler stimulus input.org by Columbia University on 02/08/06. Thousandsof experiences recognizing pennies clearly led to little learning of its features onceit became highly familiar on the basis of somesubset of features.In the extreme.we often do further strengthen the existing associations. Information that is in the part of sentence that the syntax signals to be newis better learned than information in the part the syntax signals to be given (Hornby 1974.g.annualreviews. Such interference has been demonstrated (Owens et al 1979). Probably the primary reason that maximallearning occurs for material with intermediate degrees of prior integration into chunks is that this is . For personal use only.32:21-52. Singer 1976). daboration has not always been found to provide a net benefit to memory (Nelson et al 1978. Stein et al 1978. a United States penny. appears to be increasing the potential for proactive and retroactive interference. The Stein et al and Morris et al studies indicate that certain kinds of elaboration increase memorability more than other kinds. Morris et al 1979). we knowfor certain that multiple associations are important for makinga piece of knowledgerelated to all that it should be related to in memory. Of course. All of these observations support the hypothesis that learning is maximalat intermediate degrees of prior chunking (integration) of the material to be learned (Wickelgren1979a. Rev. 1981. though only studies measuring learning on an unbounded strength scale are relevant for establishing such diminishing returns without ceiling artifact (e. but this is not when the greatest amount of learning occurs.

This binds the formerly free chunk node to represent the set of its constituent nodes.32:21-52. thus preventing further chunking. the links associating the new chunk node to its constituent nodes will be strengthened. Tyler et (1979) found that greater difficulty in finding the correct wordto complete a sentence or solve an anagramled to better memory the target words. Rev. the entire experience is too of complexto be integrated into one chunk at that time. This mayexplain whylearning of familiar words and pictures has often been enhancedby increasing the difficulty of processing the material: presenting incomplete pictures (Kunenet al 1979). Masson& Sala 1978). Psychol. the hippocampal chunking arousal system.org by Columbia University on 02/08/06.Annual Reviews www.annualreviews. preventing epilepsy (Milner 1957).annualreviews. (c) Lateral inhibition among cortical nodes limits the total numberof nodes (free or bound) that can be fully activated at any one time. Activating a set of boundnodes that is already well integrated under a boundtop chunk node produces a very high level of activation of these nodes. Boundnodes have strong input and output links to other nodes in cortical memory and weak links to the hippocampal chunking arousal system. For personal use only. as with some mylectures. but it will be of only one or a few subsets of the component features of the complexunfamiliar experience. Downloaded from arjournals. one way to obtain further chunking of a highly familiar entity wouldbe to focus attention on subsets of features and the relations between them. Chunkingmayoccur.The theory assumes the following: (a) There are two (fuzzy) sets of nodes in cortical associative memory. and as a consequence maybe difficult for an experimenter to measure. When the current experience is already integrated under one top chunk node. little or no further chunking occurs. Alongthe same line. This inhibits the activation of all other nodes. requiring completionof incomplete sentences or recognition of inverted text (Kolers & Ostry 1974. including free nodes. (b) chunking. for The chunking and consolidation theory proposed by Wickelgren (1979b) can be extended in an elegant way to account for a variety of learning phenomena.org/aronline HUMAN LEARNING AND MEMORY 43 where maximal new chunking (specification of new nodes) occurs. bound and free. 1981. . Accordingto this theory of chunking. Such partial chunking is under the control of the subject’s attentional and grouping strategies. presenting misspelled words for error correction (Jacoby et al 1979). Whenthe experience is an unfamiliar combination of too manyseparate (ungrouped or unrelated) parts. the chunking arousal system primes the free nodes so that they can competesuccessfully for activation against other boundnodes to which the set of bound nodes to be chunked mayalready be associated. and possibly to. The node or set of nodes maximally activated (by the weak links from the bound nodes to be chunked) will increase in activation so that by standard nodal contiguity conditioning (Hebb 1949). as in recognizing a penny.Activating less strongly asso- Annu. Free nodes have weakinput and output links to other cortical nodes and strong input links from.

Wickelgren et al 1980). Storage: Consolidationand Forgetting Annu. (d) Binding a chunk node to its constituents initiates a consolidation process that cumulatively disconnects the now bound chunk node from the chunking arousal system. typically at a point near the asymptote. most obviously subject’s ratings of trace strength. 1974. accurate. and. Retrieval SPEED-ACCURACY TRADI~OFFFUNCTIONS The study of the dynamics of memory retrieval. and indeed of all cognitive processes. and elegant than any other. A. Obtaining the entire speed-accuracy tradeoff function for each condition avoids this problem and permits comparison of two conditions without the possibility of an invalidating speed-accuracy tradeoff (Wickelgren 1977c). 1979b) is. I have nothing to add to the theory at this time. For personal use only. A variety of phenomena. 1977a. establish that there are manylevels of trace strength above the minimum level necessary to yield 100%correct performance at asymptote (unlimited retrieval time).Annual Reviews www. Unless those asymptotic (stored) strength differences are estimated and factored out using speed-accuracy tradeoff methods and. The theory of consolidation and forgetting described in Wickelgren (1972c.org by Columbia University on 02/08/06. 1976) had the insight that the phenomenon speed-accuracy tradeoff in reaction time could be used to of study the time course of all mental information processing (all of which is memory retrieval in the broadest sense). much more general. pp. where retrieval dynamicsis over (Wickelgren 1977c). (b) Obtaining the entire speed-accuracy tradeoff function provides muchmore extensive information concerning the time course of retrieval dynamicsthan does a reaction-time experiment. one cannot study re- . preventing it from being recruited to represent someother combinationof constituents. 1981.32:21-52. Rev. if necessary. but a lower accuracy level. Speed-accuracytradeoff functions have three major advantages over reaction-time measures (even accompanied by an accuracy measure): (a) It is not possible to meaningfully comparethe ditficulty of two conditions whenone condition has a shorter reaction time. V. Reed (1973. except for the modifications concerning chunking and spacing effects in learning discussed in the last section. 362-94.annualreviews.annualreviews. in my opinion. Psychol.org/aronline 44 WICKELGREN ciated sets produces less inhibition. Downloaded from arjournals. (c) Speed-accuracy tradeoff functions permit separate estimation of the strength of a memory trace in storage and its retrieval dynamicsparameters and functional form. Reaction-time measures completely confound storage and retrieval. permitting somefurther integration of these bound nodes by strengthening associations from various combinations of these boundnodes to free chunk nodes. incremental scaling (Wickelgren 1978. than another condition. has been given a powerful new experimental tool comparable in significance to the invention of the microscope. which provides the equivalent of a single point of such a function.

org by Columbia University on 02/08/06. Indeed. we can draw some conclusions about memoryretrieval dynamics with varying degrees of certainty: (a) Microretrieval in both recall and recognition is simple direct-access process.Annual Reviews www. Dosher 1976. not a search process (Corbett & Wickelgren 1978. cascade processing. but in a partially overlapping manner.Nevertheless. Wickelgren (1976a) calls it chain-parallel processing. coding level is perhaps the most important determinant of retrieval dynamics. The major uncertainty is just howlittle the loss is. Someof us now clearly understand that memory traces should be considered to have at least two extreme states: -etrieved (on our minds. we knowmuch less about retrieval dynamicsthan we ought to know. conscious. Finally. 276-78). Wickelgren Corbett 1977). subject’s ratings of trace strength establish that they do not. and there have been only a few tens of speed-accuracy tradeoff studies. Virtually all reaction-time studies of memory uninterpretaare ble with respect to retrieval dynamics because of this confounding storage of and retrieval. and howeach is affected by various conditions.annualreviews.annualreviews. not strictly serially. For personal use only. Remington 1977. Retrieval of many. Rev. speeding retrieval dynamics (Corbett 1977). 1981. Since this conclusion is supported by the arguments for associative memory described earlier. such that partial retrieval at a lower level initiates retrieval at the next higher level and continually updates its input to the next level as its ownretrieval becomesmore complete. (b) Tentatively. (e) Despite cascade processing.and sometimes all. levels is occurring in parallel. one of the most important accomplishments in the study of retrieval is purely conceptual.org/aronline HUMANLEARNING AND MEMORY 45 trieval dynamics. pp. because storage and retrieval are confounded. (d) Retrieval of each link in a chain of links occurs. what the span of attention is. There is quite a lot of evidence supporting cascade processing. in active memory) unretrieved (not and Annu. we can process several retrieval cues in parallel with little or no loss in efficiency (Dosher1976. pp. (c) Withinthe span of attention. Thus. we can be quite confident of it. Turvey(1973) calls this parallel-contingent processing. Psychol. in particular. not only in increased strength. Downloaded from arjournals.32:21-52. 273-76). WHAT WEKNOW? DO Since the thousands of reaction-time studies of memory retrieval dynamicstell us very little that is definitive. (f) Repeated retrieval of an associative chain in somecases can result. Wickelgren & Corbett 1977). just because subjects always respond correctly that "a robin is a bird" and "a chicken is a bird" does not meanthat these traces have equal strength in storage. Higher-level traces (longer associative chains from input to output) have slower retrieval dynamics (Wickelgren 1979a. and McClelland (1979) probably has the best namefor it. both theories of automatization are correct (Wickelgren 1979a. Thus. . microretrieval in recall is identical in dynamicsto recognition (Wickelgren&Corbett 1977). but also in a short-circuiting of the chain.

namely. active is (primary) memory. longand term memory. it is clear that there is one kind of short-term memory. Because we are not thinking all thoughts simultaneously and because time is required to retrieve new thoughts. believe that this is the only distinct form of short-term I memory.(c) There are three dynamically distinct types of traces: active memory. Downloaded from arjournals.namely.annualreviews. the traces associated to this attentional focus. The focus is what is being consciously attended to. Psychol.and there is the subset of this passive memory is currently in various states that of activation.32:21-52. Such active memory the only true form of short-term is memory. (b) There is only active short-term memory and associative long-term memory. presumably due to high levels of interference.This must be true.there are three principal theoretical alternatives: (a) There is only active short-term memory associative longand term memory.Active memory extends to encompass short lists of items (the span of immediatememory). . So-called short-term memory short lists is just long-term associative for memory under conditions where forgetting is rapid. and the halo includes the entire attentional set of traces primed(partially activated) by the traces the attentional focus. short-term memory. even the immediately previous one. Short-term is memory mediates the span of immediate memory. There is long-term associative memory(passive memory). active memory.and active memoryis limited to the trace(s) currently being thought of and. to a lesser extent. Previously activated thoughts. For personal use only. at least not to the same degree.It is clear that one form of short-term memory different from long-term memory. are not in active memory.org by Columbia University on 02/08/06. because we are not thinking all thoughts simultaneously. Short-Term Memory Annu.In bridging the gap from active memory the sort of memorythat every cognitive psychologist to agrees is long-term memory. Persistence in the retrieved (active memory) state is one type of short-term memory.annualreviews. Long-term memory accounts for everything else. but we have yet to establish this definitively. Active memory equivalent to the span of attention.org/aronline 46 WICKELGREN on our minds. modified to is include both a focus and an associative halo of decreasing activation.Annual Reviews www.Active memory limited to the currently activated thought is and its associative halo (the span of attention). 1981. unconscious. in passive memory). to regardless of whether the trace on which the memory based is the same is or different from associative long-term memory. Speed-accuracy tradeoff functions are most simply interpreted as indicating that the transition from the unretrieved state to the retrieved state is an incremental rather than an all-or-none process.Active memory limited to the span of attention. Rev. I will use "short-term memory" refer to any rapidly forgotten memory.

Downloaded from arjournals. A very remarkable property of the retention function for long-term memory is that it automatically includes a rapidly decaying short-term memory buffer without the need for a separate short-term memorysystem. This initially rapid fall in trace strength meansthat it is impossibleto push the strength of long-term memory traces up to their maximafor more than a fraction of a second.Besides allowing us to dial telephone numberswithout continually looking at the phonebook. I supported the second alternative for a time (Wickelgren 1979a). For personal use only. Rev. and this.annualreviews. It is also what accounts for "warm-up" effects whenwe sit downto think and write. this short-term aspect of associative memory doubtless plays an important role in speech recognition. It appears that active memory confined to the very last thought (and whatis ever it is associated to in long-term memory. in turn. This short-lasting increment to the long-term memory trace functions as a short-term memory buffer and is the basis for the span of immediate memory. 1979b). 1981. this strengthening dissipates to a large extent with time and interference. As we all know. meansthat there is always roomat the top of the current levels of long-term memory add a short-lasting to increment. Theretrieval dynamics all other items in the list wereidentical. The rest to of the span of immediate memoryis most parsimoniously attributed to associative long-term memory.32:21-52. At first we have trouble getting our thoughts to flow until we have retrieved some associations and strengthened them. Accordingly.Annual Reviews www. traces that are still in active memory some to extent (primed or partially activated) should have faster retrieval dynamics. but a recent experiment (Wickelgren et 1980) definitively rules out the second alternative in favor of the first. What we found was that the primingeffect on retrieval dynamics strictly limited to the very last item was in the list. active memory long-term memory and (Wickelgren 1973. we find ourselves spending a large fraction of our time warming up our minds reviewing old thoughts and not enough time producing new thoughts. of despite massive changes in asymptotic recognition memory accuracy. articulation. and reading. if our thinking on a particular problem is too fragmented in time. . The reason for this is that the rate of forgetting in long-term memory initially is very rapid and continually slowing downwith increasing trace age.org by Columbia University on 02/08/06. However. is clear that such short-term memory not as rapidly lost as it is Annu. then the decline in asymptotic recognition accuracy with more intervening items should have the same dynamics as the decline in the priming effect on retrieval dynamics.annualreviews.org/aronline HUMANLEARNING AND MEMORY 47 The third alternative simply cannot be supported at the present time because all of the known memoryphenomena can be accounted for more parsimoniously with but two traces. Hence. 1975. a lesser extent). Wickelgren et al reasoned that since retrieval converts traces from the passive to the active state. 1974. Psychol. if the basis of short-term memory lists (probe memory for span) is active memory.

17:141-64 Biederman. 1977. Bower. S. Psychol. and Conceptual Development. 1973. Decisionmaking in a fuzzy environment. 302 pp. there are not just two categories of memories. Hillsdale. There are memoriesthat last seconds. memoriesthat last days. NJ: Erlbaum. pp.. J. Psychol. 4:451-74 Anderson.they all derive from the same remarkable associative memory.. J. 1970. J. Zadeh. 1974. judging by trace longevity. there is currently no psychological reason to peel off the memories that last seconds and call them a different kind of memory trace. Associative long-term memory has solved the whole problem of having memories last for widely varying times. For example. failing to control degree of learning has often led to wrongconclusions. L. pp. Rev. 1970. R. A. Checkosky. presumably because such limited longevity is functional. J. list memory a typical probe memory in span experiment. F. Processing redundant information. J. Word.Even active memory but a change of state for traces is in the same associative memory system. 1975). 83:486-90 . Anderson. M. Anelaborative processing explanation of depth of processing. even with no difference in forgetting rate (Wickelgren 1977a. memories that last minutes. NewYork: Norton. E. 1976. Hence. Bellman. Memory. Object. 546 pp. 524 pp. except for active memory. but this state change is a dynamically different trace from passive long-term memory. and Thought.org by Columbia University on 02/08/06.. Psychol. Reder. The difference is presumably to differences in the susceptibility of the traces to interferdue ence (Wickelgren 1974. people often erroneously assumethat initial degree of learning is equal. short-term and long-term.. Functionally.annualreviews. J.Annual Reviews www. With the large amplification factor that prevails between degree of learning and trace longevity.J. Exp. M. H. Psychol. See Eysenck 1979. R. Although we mayfind evidence for somebiological separation into several stages of memory storage. Retrieval of propositional information from long-term memory. R. what appears to be a difference in "forgetting rate" is very often merely a difference in degree of learning. Downloaded from arjournals. memoriesthat last hours. Language. Cognit. there is a continuous spectrum of trace lifetimes. 1973. but the~e mayalso be important differences in the degree of learning that account for someof the differences in howlong the memory lasts as well. A. Anderson. whenthe differences could be quite substantial. 385-403 Anglin.annualreviews. R. 1979. Manage. Another remarkable property of long-term retention is that modest differences in degree of learning are amplified to produce enormousdifferences in trace longevity. and memories that last months and years. Sci.32:21-52. Human Associative Memory.org/aronline 48 WICKELGREN Annu. Rev. When initial degree of learning in two conditions is sufficient to produce nearly perfect performance on immediate retention tests. a factor of two in learning can increase trace longevity by a factor of 100. (J. R. 371-72). I. In any case. Theoretically. L. A theory for the recognition of items from short memorized lists. For personal use only. Washington DC: Winston. 80:417 38 Anderson. Literature Cited Anderson. 1981.

8:87-88 Eysenck. M. Learn. Comprehension of antonymyand the generality of categorization models. NewYork: Academic. F. A. The retrieval of sentences from memory:a speed-accuracy study. Behav. Rev. Learn. 1979. Aspreading-activation theory of semantic processing. 30:1-15 Craik. 1977. 82:407-28 Corbett. Learn. Bower. 608 pp. pp. J. K. ed. New York: Academic. 1975.. F. 1975. J. E. Am. J. pp. A. DaPolito. H. 328 pp. 1972. Learn. 1979. D. M. 103:706-16 Glass. Integrative and metric properties of abstracted information as a function of category discriminability. E. Chaffin. H. 479 pp. 1958.. Hebb. H. D. I. 1977. Mem.org by Columbia University on 02/08/06. B. Melton. NJ: Erlbaum. 88:297-306 Holyoak..I.. A.. J. M. K. 308 pp. 1977. J. Math. 85:490--512 Herrmann. ed. Z Exp. Mere.. Craik. F. 1979. S. In Coding Processes in HumanMemory. 1978. L. P. 1978. N. 16190. 1971. Wickelgren. T. A. Psychol. Behav. Verb. Lloyd. Bull. elaboration." A Cognitive Analysis. F. M. Block. ed. Learn.J. Psychol. J. Cambridge. The Organization of Behavior. Oninterpreting the effects of repetition: solving a problemversus remembering solution. R.. Moore. Psychol. Psychol. Glass. J. 241 pp. J. J. Semantic memoryretrieval: Analysis by speed accuracy tradeoff functions. G. Brown. Peters. E. P.. New York: Academic. 315 pp. R.org/aronline HUMAN Bower. J. Roseh. James. H. Verb. 247 pp. L. Hillsdale. 1967. 8:291-310 Estes. J. A. A brief statement of schema theory. 5:585-97 Hintzman. Begg. P. NJ: Prentice-Hall. Verb. Effects of decision difficulty on recognition and recall. L. I. Findler. 14:215-39 Hornby. 1978. Psychol. Craik. J. Clark. 1972. 1974. D. A. H: H. Exp. Dosher. Exp. 1981. 315 pp. A. Verb. Clark. Conti. Psychol. S. The roles of contradictions and counterexamples in the rejection of false sentences. A. Fahlman. Rev. Depth of processing and the retention of words in episodic memory. 1975. 1979. In The Psychologyof Learning and Motivation. 1978. 1978. Learn. Free trees and bidiJ. D. S. E. de Villiers.32:21-52. NewYork: Wiley. 4:171247. Behav. Psychon. W. V. Evans. 18:585-600 Johnson. Hintzman. B. D. Retrieval dynamics for rote and visual image mnemonics. I. D. S. 1978. Exp. NJ: Erlbaum. G. ed. Bower. What’s in a word?On the child’s acquisition of semantics in his first language. In Levels of Processing in HumanMemory. 104:268-94 Cunningham. 1976.-Hum. T. Verb. 1949. H. E. Robbins.. M. Q. Repetition and memory.. 1-21 Jacoby. R. Mass: Harvard. W. 312 pp. 1973. 169-211. 17:165-88 de Villiers. J. instance variability. Psychol-Gen.Language Acquisition. Associative Learning. N. In Cogfor nition and Categorization. 16:233--46 Corbett. New York: Free Press. A. Annu.. J. G. D. 1974. Behav. and experience. F. Interference paradigms for meaningful propositional memory. 1976. 91:575-85 Br~)oks. W. Psychol. Repetition and memory: Evidence for a multiple-trace hypothesis. J. See Eysenck 1979. Herrmann. Tulving. Psychol. Verb. Role of antonymyrelations in semantic judgments. Downloaded from arjournals. P. ed.. C. I. F. 1978. Behav. Associative Networks. Learn. In The Psychology of Learning and Motivation. O. 1979. How frequency affects recency judgments: A model for recency discrimination. L. 5:598-606 Goldman.. Effects of elaboration of processing at encoding and retrieval: trace distinctiveness and recovery of initial context. Martin. L. Cognit. 1979. Words and Things. Psychol.. T. Psychology and Language. B. J. Psychol. Holyoak. pp. A. Englewood Cliffs. J.Annual Reviews www. A. Bower. E. Craik. Psychol. L. Verb. Cermak. For personal use only. Mass: MIT. J. Exp. Surface structure and presupposition. An associative basis for coding and organization in memory. Depth. .-Hum. Hillsdale. Homa.. New York: Harcourt.annualreviews.. Kiser. 1978. P. An old problem for the new psychosemanties: synonymity. K. 1979. NETL: A System for Representing and Using Real-World Knowledge. Nonanalytic concept formation and memory instances. In Cognitive Development and the Acquisition of Language. LEARNING AND MEMORY 49 Flexser.. H. 89-118. NewYork: Academic. Polson. W. J. J. J. L. Verb.. Exp. L. ed. Verb. E. Psychol. The role of chunking and organization in the process of recall. NewYork: Wiley. L. G. 3:375-85 Greeno. and distinctiveness. 17:649-67 Jacoby. M. A. F.annualreviews. rectional trees as representations of psychological distance... G. 335 pp. Collins. Loftus. E.. I. 448 pp.ed. Learn. Clark. Sci. G. Exp. L. 462 pp. T. G. a Verb.. 13:530-38 Jacoby. pp. D. J. Cambridge. 278 pp. J. L. V.-Hum. 10:47-91. L. Mem.

1935. Keele. E. 1978. O.. Cognit. 4:16-30 Reed. Longterm memory for a common object.D. 430 pp. Sala. B. K. J. Learn. plus or minus two: somelimits on our capacity for processing information. Carnegie-Mellon Univ. Cambridge. D. B. 1974. Mem. 1956. Context theory of classification learning. 6:26-37 Oden. 4:198-209 retention.. J. Mere. W. NewYork: Harcourt. S. 1978. 1979. 1:242-99 Miller. 1954. 463-526. PsychoL Rev. PhD thesis. ¯ L Exp. J. Rev. For personal use only. T.. Pa. 1978. P. A. Verb. Vining.. B. Kolers. G. S. J... Mem. Explorations in Cognition.Z Hum. J. Science 181: 574-76 Reed. Mem. J. 6:156-64 Merlin. Psychol. 555 pp. A. Specifying the mechanisms in a levels-of-processing approach to memory... 1979. A. McDaniel. Psychol.Annual Reviews www. Bransford. O. Cognit. Men~ Cognit. Cognit.. Language 54:610-46 King.J. 64:242-52 Morals. Fodor. The Construction of Reality in the Child.. G.M. Learn. Onthe time relations of mental processes: an examination of systems of processes in cascade. Hatchett. D. Cognit. Longterm memory search: an intersecting activation process. I.annualreviews. 1976. J. Waterman. PsychoL 10:244-70 McClelland. Psychol. R. J. Cognit. Effect of multiple images on associative learning.. 16:519-33 Morals. Neumann. 315 pp. Learn. Bower. 440 pp. Learn. In Visual Information Processing. P. Ronk. M. S. 1979. K. Friedman. Decision processes in verifying category membership statements: Implications for models of semantic memory. D. 15:587-605 Klein. C. NewYork: Basic Books.-Hum. Downloaded from arjournals. Z Exp. Franks. 1968. of abstract ideas. Verb. S. Cognit. O. Verb. Psychol. Psychol. ed. W. 1975. Integration of featural information in speech perception. Behav. NewYork: Academic.. 1978. 1979. 1981. V.annualreviews.E. G. pp. Behav. V. J. M. D. Production systems: models of control structures. A. 1979. L. T. S. 1966. J. 1979. Principles of Gestalt Psychology. Behav. Anderson. Learn. G. L. 1957. Mem.. Psychol. Cognit.Prerequisites for the utilization of knowledgein the recall of prose passages. F.. San Francisco: Freeman. D. V. D.C. 1977. Schaffer. J.. Exp. 63:81-97 Milner. D. Verb. V. 16:693-98 Reed. Timecourse of loss of information regarding pattern analyzing operations. 85:207-38 Meyer. 11:1-37 MeDaniel. 1976. J. Language39:170210 Kay. J. A fuzzy logical model of letter identification. L. Posner. K... G.32:21-52. 5:187-97 Newell. Rumelhart. 5:336-52 Oden.. H.org/aronline 50 WICKELGREN Nelson. Ostry. 5:574-84 Masson. A. 1978. Rev. D. Psychol. Reed. Bransford. S. aev. 1963. eds. C. 11:287-307 Norman. C. Adams. 1978. Learn. K. Exp. Glucksberg. M. Green. Effect of semantic versus structural processing on long-term Learn. A. The "soap opera" effect in story recall.. Spread of encoding effects within the nonverbal domain. R. G. Oden. Percept. Interactive processes in sentence comprehension and recognition. Minsky. On the genesis77:353~3 chol. Pittsburg. Mass: MITPress. G. M. Psychol. A. Rev. Quantitative prediction of spacing effects in learning. Learn. Semantic constraints and judged preference for interpretations of ambiguous sentences. Psychol. 1973... K. Cognit. 13:599-612 Kunen. J. Greene. 2:671-79 Koffka. 1977. Rev. 6:337~tl Nelson. S. R. xp. 1970. Mem. The magical number seven. 16:151-71 Nelson. J.. Cognit. S. Psychol. E. 1976. J. W. Verb.. Repetition and depth of processing. Exp. Psychol. Black. 7:185-91 Piaget. 485 pp. D. J. A. Mere. G. 1978. Cognit. W. M.Hum. A. J. Saltz. M. 1978. Bourne. L. Pattern recognition and categorization. Merit 5:253-61 Annu. pp.. Learn. Mere.org by Columbia University on 02/08/06. Verb. P. PsyQuillian. (Reprinted in part in Semantic Information Processing. 85:172-91 Owens. 1968. M. Stein..D. PsychoL: Hum. 227-70. ed. Verb. A. D. C. . Verb. 1977. The linguistic significance of the meanings basic of color terms. T.. D. Behav. C. R. Igl.. Onthe representation and retrieval of stored semantic information. 3:3824O7 Katz. J. J. Nickerson. E. M. Remembering levels of inthe formation in words. 1979. Z Verb. Semantic memory.P. G. Chase. Verb. Levels of processing versus transfer appropriate processing. PsychoL-Hum. Perform. 1973. Massaro. 1972.J.. E. Speed-accuracy trade-off in recognition memory. The cell assembly: Mark II. 86:287-330 McCloskey.. M. List length and the time course of recognition in immediate memory. 1977. Behav. Visual prototype formation with discontinuous representation of variability. The structure of a semantic theory. Psychol.

Abstraction of prototypical information by adults and 10month-old infants. J. Context-sensitive coding in speech recognition. Lindsay. A. N. J. McCallum. Mervis. pp. B.. 1971... 5:618-32 Turvey. C. Processing of phonemes in speech: A speed-accuracy study.. A. D. Deutsch. 1972a. retrieval. PsychoL-Hum. E.. W. L. Donaldson. Wickelgren. A. 1969. Mere. P. M. In Handbookof Learning and Cognitive Processes. Verb. A. 9:418-55 Wickelgren. A. S. Cognit. Learn. Biophys. S. W. 1969c. Wallace. Learn.. 311 pp. K. Psychol. Wickelgren. E. Psychol. 1978. S. New York: Academic. E. Mass: Harvard. Verb. PsychoL Rev. Exp.Hum. Rosch. J. Constraints on effective elaboration. H. W. J. B. In Cognitive Development and the Acquisition of Language. Verb. Hertel. 1965. In Cognition in Learning and Memory. Depth of processing reexamined: the effects of the precision of encoding and test appropriateness. T. Smith.. 1981. Rips. B. A. Johnson.Annual Reviews www. Cambridge. M. 5:607-17 Wason.. Leibovic.. Gilbert. Sehaeffer. D. Behav.org by Columbia University on 02/08/06. Thematic structure and the integration of linguistic information. Learn. B. M. Psychol. K. ed.E. Ellis.549-58 Smith. In Short-Term Memory. H. 17:165-74 Stein.J. W. H. Psychol. 1969a. Trace resistance and the decay of long-term memory.ed. R. 1975. 31:123-42 Wickelgren. E. N. Wickelgren. Coding. D. Moore. Behav. D. C.org/aronline HUMAN LEARNING Remington. Exp. ed. A. 1978. Estes. In Information Processing in the NervousSystem. Morris. 1972b. 1978. B. 1977. Am. Verb. Recognition memory for targets from a scannedwordlist. J. W. Gary. Psychology of Reasoning: Structure and Content. 1970. Verb. 1973. Verb. 1976a. A. NewYork: Academic. Norman.. H. 82:343-46 Schaeffer. C. A. 197246. ed. 65-72. T. N. Bull’ Math. I. More on the long and short of memory. Theories of semantic memory.E. E. H. £ Verb. R. Br. J. J. 111-44. Downloaded from arjournals..Learn. J. P. Exp. 1972. ed. 8:382-439 Ross. 272 pp. Psychol. E. W. 15. M. 17:14-25 Wiekelgren. 1973. 1:1-49. 1973.annualreviews. and serial order in (speech) behavior.. Behav. Wallace. L. J. 315 pp. Network strength theory of storage and retrieval dynamics. A. Single-trace fragility theory of memorydynamics. Mere. Memory for at least one of two items: test and failure of several theories of spacing effects. Bransford. 7:573-605 Roseh. 1974. pp. Mere. Human categorization. . 1979. Short-term memory for repeated and non-repeated items. 81:214-41 Stein. C. New York: Springer.. ed. P. pp. 6:1-56. Verb. Soc. Psychol. The long and the short of memory. J. 1978.. 1974. Learn. NewYork: Academic. W. parallel processing in speech. Learned specification of concept neurons. 1969b. K. 315 pp. 1972. W. W. M.P. Rosch. NJ: Erlbanm. W. 308 pp. E. Learn. Tulving. Context-sensitive coding. C. E. 17:669-80 Rumelhart. Family resemblances: Studies in the internal structure of categories. ed. 1976. H. 2:775-80 Wickelgren. W. Psychol.. 80:1-52 Tyler. Boyes-Braem. Smith. W. Semanticdistance and the verification of semantic relations. 12:1-20 Rosch. Context-sensitive codingand serial vs. S. 1979. Subproblems of semantic memory: A review of Human Annu. A. W. R. H. Q.. Exp. In Advancesin Cross-Cultural Psychology. Onthe internal structure of perceptual and semantic categories. pp. and development. 62:335-46 Singer. Math. H. pp. Cognit. 83:466-78 Wickelgren. Hillsdale. Warren. 85-96. J. 264 pp. Semantic similarity and the comparison of word meanings. D. Psychol. D. Shoben. 1976. 411 pp. Landauer. Behav. Rev. Gregg. Structure and process in semantic memory: Afeatural modelfor semantic decisions. Johnson-Laird. PsychoL Bull. A. 1972c. B. associative memory. Learn. NewYork: Wiley. W. On peripheral and central processes in vision: inferences from an information-processing analysis of masking with patterned stimuli. 62:1279-90 Rips. 1977. NewYork: Academic. Deutsch. C. J. E. A.. J. P. Cognit. Rev. 237-62. L. Behav. E. E. J. The comparison of word meanings. 80:425-38 Wickelgren. PsychoLl~ev. NewYork: Academic.annualreviews. Z Acoust. Exp. J. T. B. Cognitive effort and memory. 76:1-15 Wiekelgren. J. For personal use only. 1976b. In Speech and Cortical Functioning. E.. Wickelgren. Shoben. E. 1973. In Organization of Memory.32:21-52. A. 423 pp. articulation. 1975. Basic objects in natural categories. D. W. and dynamics of multitrace associative memory. W. Verb. J.. 19-50. Psychol. W. pp. Wickelgren. 17:707-14 AND MEMORY 51 Strauss. Psychol. Mervis. 86:144-52 Schulman. Verb. 212 pp. A.. ed. PsychoL Rev.

. i977b. Wickelgren. J. What’s in a link: Foundations for semantic networks. A. Wickelgren. Corbett. In Representation and Understanding. A. pp. C. NJ: Prentice-Hall. Freidman. R. 436 pp. Chunking and consolidation: Atheoretical synthesis of semantic networks. Bull. A. 1977a. 1953.-Hum. 1971. Psychon. A. W. A. I liked the postcard you sent Abe and I: Context-sensi- . W. 10:232-34 Wickelgren. Downloaded from arjournals. 35-82. Behav. Anderson. Psychon. T. P. 1977. ed. A. Commun. A. A. Inf. A. 13:243-68 Wickelgren. 1977c. 42:81-82 Wickelgren. Carterette. W. New York: Academic. W. In Handbookof Perception. Norman. E. T. Cognitive Psychology. Soc. Priming and retrieval from short-term memory: a speed accuracy tradeoff analysis. W. Speed-accuracy tradeoff and information processing dynamics. Englewood Cliffs. ~. J.L. Sci. A. A. For personal use only. Bower. 1979c. by G. configuring in conditioning. l~erb.org/aronline 52 WICKELGREN tive coding of syntax and other procedural knowledge. Control. Psychol. D. Learning and Memory. Fuzzy sets. NewYork: Macmillan Woods.Annual Reviews www. IJobrow. Mere. Math. 1981. W. S-R vs Cognitive Learning. Rev. Psychol. 501 pp.. Associative interference and retrieval dynamics in yes-no recall and recognition.. Learn. Learn. In press Wickelgren. L. W. 7:227-64. Psychon. W. Psychol. A. D. Concept neurons: A proposed developmental study. Phonetic coding and serial order. NJ: Prentice-Hall. 1979a. lnvariance of forgetting rate with number of repetitions in verbal short-term recognition memory. A. W. NewYork: Academic. Dosher. normal forgetting. Psychol. A. 1980. B. W. 1979b. Philosophical Investigations. 1975. ed. Ettglewood Cliffs.annualreviews. 8:338-53 Annu. . Acta Psychol. 427 pp. and the hippocampal arousal system. A. Rev. W.. W. 1976c. H. Corbett. A. A. M. Soc.32:21-52. L.annualreviews. 13:61-63 Wickelgren. Exp. Associative Memory J. 1978. 448 Wickelgren. Verb. Acta Psychol. 3:189-202 Wickelgren.org by Columbia University on 02/08/06. Zadeh. A. 41:67-85 Wickelgren. Wickelgren’s neglect. Transition network grammars for natural language analysis. 1970. Collins. 22:363-64 Wittgenstein. A. 86:44--60 Wickelgren. W. ACM13:591-606 Woods. 1965. the amnesic syndrome.

Psychol.org by Columbia University on 02/08/06.annualreviews. Downloaded from arjournals. For personal use only. 1981.Annu.32:21-52. . Rev.

You're Reading a Free Preview

Download
scribd
/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->