Long term memory stores
our knowledge of the world. It enables us to recall events, solve problems,
comprehend, recognize patterns, etc.
(permastore).
Bahrick et al. - memory for names, faces, and matching of names with faces of yearbook portraits (3.3 months to 48 years):
1. free recall of names
2. picture recognition test
3. name recognition test
4. match names with pictures
Findings: face recognition - 90%; ability to match names with faces - 90%
Bahrick & Hall (1991) high school Spanish was well remembered 30 years later
Types of LTM: episodic, semantic, procedural (Tulving)
Episodic - autobiographical memory, time and place dependent, it conveys the conditions of occurrence. Knowledge about personally experienced events and their temporal relations. Flashbulb Memories - extremely vivid episodic memories, usually attached to a significant or vivid event (surprising or emotional event)Recognition and Recall - retrieval of information from LTMSemantic - knowledge of concepts, meaning, all the information needed to use language, knowledge independent of time or place, verbal information, visual - spatial information, etc. Decontextualized knowledge.(Procedural and Propositional)
Procedural - connection chains between stimuli and responses. "How to" knowledge. Consists of a production system: 1) a condition which specifies a set of features that must be true of memory; 2) an action that specifies a sequence of changes which should be made to memory if the condition is or is not met, e.g. crossing the street at the light: if light is red and (cond.) wait (action); if light is green and (cond) cross (action) (State = want to cross). Usually cannot be explicitly stated.
Threshold Hypothesis - both recog, and recall depend on the strength of the item in memory. Before an item can be recognized, it must have a certain strength value - Recognition Threshold; similarly there is a Recall Threshold which is assumed to be higher than the recognition threshold.
This means three things:
1) items of low strength will be neither recognized or recalled
2) items of higher strength will be recognized but not recalled
3) items of an even higher strength will be both recognized and recalledThis theory accounts for the finding that recognition is a more sensitive measure of memory.
IF the threshold hypothesis is correct - that only item strength affects the difference between recog. and recall - THEN variables affecting recog. in any way should affect recall in the same way. There is much evidence supporting this hypothesis:
1) the number if items occurring between test item presentation and test;
2) rate of presentation;
3)serial position effect.
All of these variables push the strength of the item up or down and thus affect recog. and recall in the same way.
HOWEVER,if a variable is found that affects recall and recog. differently then the threshold hyp. is called to question:
Kintsch (1970) found that word frequency differentially affected recognition and recall: recogniton is impaired by using frequently occurrring words in the TBR list and recall is enhanced
Dual-Process Hypothesis (Generate-Recognize Model) (Anderson & Bower, 1972: Kintsch, 1970)Assumption: recall includes recognition as a subprocess. Recall is made up of two processes: search and decision; recogniton corresponds to the decision process. Search requires generation.
Evidence: variable manipulations separately affect the search and decision processes:
Kintsch (1968) memory for categorized lists using recog. or recall tests.
He manipulated the structure of the lists: high (words strongly associated with the category name) and low structure (words weakly associated with the category name).
Findings: low structure led to poorer recall but nodifference between groups for recognition;
Conclusion: list structure affects the search and not the decision processesProblems for the two process theory: Watkins (1973) there are times when recall is superior to recognition. Presented pairs of items "EXPLO-RE" and "SPANI-EL" Tested by either cued recall "EXPLO-?" or of recognition "RE". Recall was dramatically higher 67% vs. 9%.
Recognition failure of recallable words or recognition failure . Ss were unable to recognize the word leprosy but were able to recall it when given its definition (Tulving & Thompson, 1973) According to a two process theory, recog. failure should almost never happen:
Researchers argue that you can account for these problems by the fact that people normally store both the tbr information and contextual information at the same time. Both recall and recogntion tend to be better when the contextual information present at learning is also present at time of the memory test. Both recognition and recall are higher when the study and the test contexts are the same (Tulving & Thompson, 1973 - the Encoding Specificity Principle). Tulving and Thompson argued that there are some differences between recogntion and recall: While the amount of informational overlap between information in the memory trace and in the retrieval environment is crucial to both recog. and recall, a greater amount of informational overlap is required for successful recall than for successful recognition. Recall requires naming a previous event whereas recog. involves only a familiarity judgment.
Baddeley(1982) proposed a distinction between intrinsic and extrinsic context. Intrinsic context has a direct impact on the meaning or significance of the tbr item (e.g. strawberry vs. traffic as intrinsic context for jam) whereas extrinsic context (e.g. the room in which the test is given) does not.
According to Baddeley, recall is affected by both intrinsic and extrinsic context, whereas recognition is only affected by intrinsic context. (Baddeley and Godden - learning on land or in the water)
Organization of information in LTM
Research and theories
on the organization of information in LTM can be looked at in two ways: simple
organization and complex organization. Early research focused on simple
organization: associative clustering, categorical clustering, and hierarchical
organization.
Task: Experimenter imposed organization (Jenkins, Russell, & Mink)constructed lists by manipulating the association value of words in the list. The words were presented randomly. Recall the words in any order. Findings: Ss recalled the words in terms of their associations. Conclusion: Ss organized the list in a way that reflects their internal organization (associative clustering)Bousfield and Bousfield (1951) - presented a list of words wherein the words were categorically related. The list was presented in random order and the subjects recalled them in categories. Conclusion: Ss organized the list in a way that reflects their internal organization (categorical clustering)
Some-or-none law - if Ss recalled one item from a category, they recalled more than one. WHY?
Bousfield explained his results thusly: in learning a categorized list , all the instances of the category will become asociated to a higher-order structure representing the category itself. Later, the recall of just one of the instances tends to activate the superordinate structure which will, in turn, facilitate the recall of other instances of the category. Clustering is greater when words in a list are both categorically and associatively related (cat and dog) than when the relationship is only associative (Cofer, 1965).
Hierarchical organization - superordinate, subordinate, instances
Subjective organization: the subject builds his/her own structure into the list . Task: the experimenter purposely presents a disorganized list (no inherent organization).
Mandler and Pearlstone (1966) card sorting task, presented Ss with a set of 52 cards. Each card had a word printed on it. Ss were instructed to sort the cards in 2 to 7 categories in any manner they wished. The cards were then scrambled again and the Ss were to organize them again, over and over until he came up with the same organization twice in succession.
Recall. Findings: Ss recalled about 5 words for each category sorted . The greater the number of categories, the greater the number of words recalled.
Encoding Specificity Principle - Tulving and Thompson - organization is important in both encoding and retrieval."What is stored is determined by what is perceived and how it is encoded, and what is stored determines what retrieval cues are effective in accessing what is stored"
Recall and recognition are the result of an interaction between encoding and retrieval. To get at information stored in memory, the same information that was available at encoding (storage) should be available at retrieval (the cue permits direct access to the stored information).
Tulving and Pearlstone (1966) gave Ss lists of words in categories with the category name at the top of each grouping . At test half of the Ss received the category names and the other half did not. FIndings: the Ss who received the category names at test recalled more of the words.Tulving and Thompson (1970)
The effectiveness of a retrieval cue depends on the properties of the encoded event, not on the characteristics of the target word in semantic memory, e.g. words with a single meaning were frequently not recognized although they were recallable through descriptive phrases presented along with the word at input - leprosy recalled from the cue "found in colonies during biblical times". The phrase was a more effective cue than the word itself.State dependent learning - when Ss encode and retrieve information in the same physiological or psychological state
Other kinds of evidence for the importance of meaning as a basis of organization of LTM come from studies more directly concerned with the organization of material that the subject already knows:
Meyer and Schvanaveldt (1971) presented a series of letter strings, one at a time, is this a word (lexical decision task) (nurse or narse) RT taken
1) meaning affected rt speed
2) if two semantically related words occurred in sequence, the response to the second word occurred more rapidly.Meyer & Schvanaveldt (1978) showed that activation spreads so fast that 2 words presented simultaneously prime each other; findings: associated pairs (doctor-nurse) were reponded to faster than unrelated pairs.
Memory benefits from both distinctive(encoding the differences) and organizational(encoding the similarities) processing.
Epstein, Phillips, & Johnson (1975); Begg (1978)
Ss were given related or unrelated word pairs and either asked to rate their similarities or their differences (beer-wine); (beer-dog). Memory for the pairs was tested.
Findings: related pairs are remembered better when attending to differences; unrelated pairs rem'd better when attending to similarities
Conclusion: encoding involves attention to certain aspects of events, some shared some different. Optimal memory requires both. (think of faces)
Self- generation effects in memory
GENERATION EFFECT - (Slamecka & Graf, 78) - we remember things better if we generate our own retrieval cues (do it yourself)
Slamecka & Graf - group one asked to generate an antonym; group two given an antonym (FAST_ S___) vs. (FAST SLOW). The test was to recall the second word of each pair
Generation effects occur under different memory conditions (recall, recognition); they are obtained with many types of materials; and they work with non-semantic as well as semantic tasks.
Glisky & Rabinowitz (85) what is important in generation; the process or the product?
Group 1 - generated the completion of word fragments
Group 2 - read the same words
Test: Ss saw both words and fragments. Fragments were to be completed and then a recognition decision made about the completed word as a member of the study list. Complete words simply had to be recognized.
READ_READ ELEPHANT - ELEPHANT
GENERATE _- GENERATE -LEP--NT/ -LEP--NT
READ - GENERATE ELEPHANT / -LEP--NT
GENERATE - READ -LEP--NT / ELEPHANT
RESULTS: GENERATION AT STUDY LED TO BETTER RECOGNITION OF THE ITEMS THAN READING AT STUDY REGARDLESS OF TEST CONDITION (THE GENERATION EFFECT)
RECOGNITION WAS BETTER FOR ITEMS IN THE GEN/GEN COND. THAN IN THE GEN/READ COND
Implications: the generation process contributes to memory over and above the effect of a generated product. The psychological process of generation is at least as important as the structural representation produced by the process in accounting for generation
Organization at encoding produces the general information being described in the retrieval process. Organization is the extraction of shared information. Results from organization can be a useful starting point for retrieval. Information specific to the event in question within the general organization is also extracted at encoding. This specific or distinctive information is necessary if we are to move from general information to specific event memory in retrieval. (ESP) What can be retrieved is what has been stored. And what has been stored can only be retrieved under the appropriate circumstances.(remember the studies on general vs. specific questions about some autobiographical information)
Mantayla (1986) the power of self-generated retrieval cues. Ss were instructed to generate three properties for each word. when they were again given those properties as retrieval cues they performed at the 90% level.
McDaniel, Wadill, & Einstein (1988) the generation effect can be explained as the "enhancement of both relational and distinctive processing". Generating information is analogous to problem solving. Ss use whatever cues are available to solve the problem including cue words, word fragments,, and even other read or generated words in the studied list. In the course of problem solving, the task focuses the subject on relational or distinctive aspects of the tbr information (remember from the domains of processing/ not more processing at the same level but more types of processing leads to better memory)
Is memory organized around wording or meaning?
Is memory for something heard or read simply a verbatim account of that data or is it different?
Sachs (1967)
Kintsch, McKoon, Keenan (1977)
These findings do not mean to imply that we cannot retain verbatim information, clearly we can memorize information in verbatim form however, most of the time when we are listening to information we are getting the gist (Bartlett, 1932 - war of the ghosts)> We tend to summarize the content and integrate the details. Integrating details requires detecting relationships ( the theme or the general idea). Guidance provided by the theme facilitates integration but at the risk of distorting one's understanding of the material. An increase in cognitive efficiency often comes at the risk of inaccuracy. (e.g. telling rumors)
Bransford and Franks (1971) "The girl who lives next door broke the large window on the porch". Ss were presented with shorter less complex sentences that contained either one, two, or three of the sentence propositions. They were later tested with "old" and "new"sentences which could be derived from the "old"> FIndings: Ss were incapable of discriminating "old" from "new" sentences therefore integration must have occurred because the complex sentences were never presented but were recognized with confidence.
One can conclude that we store related ideas as a single unit
(Organization). Ideas expressed by sentences are integrated into a single representation of the general idea expressed by a group of sentences.Complex organization - higher order units of information: SCHEMA
Schema - a large body of organized information that we have about various concepts, events or knowledge domains; a general description or framework through which we comprehend information; scripts - (Schank & Abelson, 1977) schema for routines (temporally based) e.g. a restaurant script, a classroom script, etc..each particular detail fills in or "instantiates" the script. The schema provides a stereotyped or prototypical decription of the "typical event". Scenes refer to the physical or spatial relations among objects in space (e.g. furniture placement). Brewer & Treyens (1981) what a typical office contains will influence a person's memory for a particular office.
Memory Organization Packets ( Schank, 1982) MOPs
generalized clusters of events called scenes. Scenes are collections of the high level components of scripts , organized sets of scenes and added specific contextual informationSchema theory assumes that what is encoded in memory is strongly influenced by what you already know (schema). We use these schema to select and interpret input so that it is reasonalbly consistent with what we know (our schema).
Bartlett (1932) the war of the ghosts, abstract knowledge structures aided recall and reconstruction of past events. Recall of the story differed substantially from the actual story.
Thorndyke (1977) story grammars - a story has a series of components: 1) goal state; 2) an episode or episode sequence, 3) resolution .
Seifert & Black (1983) if Ss are given a story with a common TAU (Thematic Abstraction Unit, Dyer, 1983) and are asked to write similar stories , most write stories that match the TAU
Seifert, McKoon, ABelson, & Ratcliff (1986)
Conclusion: "during the reading of an episode, thematic information may be encoded so as to lead to activation of similar episodes, but such encoding is not automatic and depends on the subject's strategies and task difficulty"Raskin (1985) scripts for joke comprehension.
TOPs Thematic Organization Points higher levels of organization which are less tied to the particular set of situations (theme)- the general character of the series of episodes (Romeo and Juliet and West Side Story same theme - mutual goal pursuit against outside opposition
The theme anchors the information (the general idea). The theme is the most likely information to be remembered and is less subject to forgetting over time. Theme is central to comprehension and serve as the focus of organizing memory for discourse.
Thematic effects (context provided before or after a text is presented) - Dooling and Lachman(1971)
Sulin and Dooling(1974) Helen Keller vs. Carol Harris, Adolph Hitler vs. Gerald Martin
Construction and distortion - memory for meaning rather than wording, distort info based on prior knowledge and expectations
Bartlett (1932) memory for info is constructed in terms of one's background knowledge or schema and this construction leads to distortion (e.g. rumors). Sometimes this is referred to as "Reconstructive Memory" "we remember by combining elements from the original material together with exisitng knowledge (assimilation and accomodation)". The original passge did not fit with their existing schema so they made it fit (canoe became dory, etc.). What we know determines what we will know.
Does construction occur at encoding or at retrieval?
Baggett (1975)
Spiro
Bob and Margie, Ss read a story about Bob and Margie. In one condition Bob did not want children, he delayed telling this to his fiance Margie and he was ocncerned. When he finally told her she strongly agreed with him or (in the second condition) strongly disagreed with him. After a series of other stories Spiro casually informed the Ss that "by the way 1) Bob and Margie are happily married or 2) they broke up. 2-6 days later the Ss were asked to recall what they could remember about Bob and Margie. Findings: Ss given the happy situation distorted their recall in that direction to accomodate the inconsistencies and the same for the unhappy ending. Conclusion: Ss assimilated the info into a more general preexisting schema for relationship information resulting in a loss of information specific to the story and a greater use of the schema at recall. At recall , the Ss reconstructed the story from the info available (making outcomes and events consistent).
Bransford and Franks (1971) found that Ss often go beyond the information that is given in recall:
Conclusion: they abstracted the combined content of the sentences that had occurred. They integrated the info communicated by sets of individual sentences to construct wholistic semantic ideas yielding a composite memory, a memory for the meaning.
Inference - the process of filling in the gaps
Comprehension requires our active involvement in order to supply information which is not explicitly contained in the text. We are especially likely to draw inferences if a sentence doesn't appear to fit the current context. We generate backward inferences (Thorndyke, 1976) In a study on story memory, the sentence "The hamburger chain owner was afraid his love for french fries would ruin his marriage" was followed by a few sentences and then by the final sentence "The hamburger chain owner decided to join weight watchers in order to save his marriage"
Thorndyke tested to see whether Ss had been more likely to draw appropriate inferences (the hamburger chain owner was fat and his wife didn't like it) than inappropriate inferences in a recognition memory task. Had various sentences appeared in the original story? Verbatim 85%/appropriate inferences 58%/ inappropriate inferences 6%Brewer (1974)Ss falsely recognized the pragmatic implications of sentences more frequently than the sentences that they actually heard (Dennis sat in Santa's chair heard as Dennis sat on Santa's lap).
Kintsch (1977) recall of meaningful prose is constructive, reproductive, and reconstructive. Immediately it is reproductive, constructive during comprehension, reconstructive during retrieval.Contextual influences activate specific schema which bias our comprehension:
Anderson & Ortony (1975) there are a number of ways that a an object may be classified, e.g. a piano as furniture or as amusical instrument. How the object is classified is determined by context. "Fist" is an effective retrieval cue for " The accountant pounded his desk" whereas "hammer" is an eff. ret. cue for "The accountant pounded the stake". Differennt inferences or asumptions are made about the pounding instrument in each case.
Anderson & Pritchard (1978) altered story perspective. Ss read the same story from the perspective or either a homebuyer or a burgler
Loftus - eyewitness testimony, leading questions, construction based errors
Eyewitness Testimony
Loftus and Palmer (1974) "how fast were the cars going when they (smashed 40.8mph, collided 39.3mph, hit 34mph, contacted 31.8mph) each other? Where do you think the highest estimates of speed were made? Ss who got the word "smashed" remembered the presence of broken glass - when there was none. Ss were influenced by presuppositions invoked by the words.
McClosky & Zaragosa (1985) argue that misleading post-event information does not affect the original memory but simply biases a person's tendency to respond in a particular way.Inferences - going beyond the given information and drawing implications. An inference is knowledge that is activated once the information is understood. There are logical inferences that must follow from the prior statement. E.g. "John's actions forced Mr. Jones to fire him" logically implies that John was fired. Unlike a presupposition, you do not have to think "John was fired" in order to understand the assertion, but the assertion demands the inference be made. A second type of inference is a Pragmatic inference - does not have to follow from an assertion but is reasonable based on world knowledge. E.g. Bill and Mary were looking at engagement rings" does not necessarily imply that they are getting engaged; however, this is a reasonable inference.
Characteristics of schema (Rumelhart & Norman, 1983):
1) vary enormously in the kinds of information they contain
2) organized in a hierarchical manner
3) operate in a top-down mannner to facilitate comprehension or interpretation of the world . These top-down processes interact in various ways with bottom-up processes
4) schemata have slots, some of which have fixed values and others that have optional values. Slots often have default values meaning that plausible guesses are made if the relevant information is not explicitly supplied (inferences)
Connectionist interpretation of schema :
Schema emerge at the moment they are needed from the interaction of the large number of parallel processing elements all working together. There is no construct which is a schema but only patterns of activation which produce the effects that are attributed to schema