1. Short Term Memory Professor Julie Yoo Fall 2008

2. Contents 1.Background 1.1 The Phenomenon of STM 1.2 “Short Term” v “Working” Memory 2.STM 2.1 Capacities and Limits 2.2 Forgetting: Decay and Interference 2.3 Retention: Recall and Rehearsal 2.3 Retrieval 3.Forms of Stored Information 3.1 Verbally Based Codes 3.2 Semantic Codes 3.3 Visual Codes 3.4 Other Codes 4.Working Memory 4.1 Problem with Traditional Definition of STM 4.2 Components of Working Memory 4.3 Support for Baddeley’s Model

3. 1.1 The Phenomenon of STM Aka: “working memory,” “primary memory,” “supervisory attention system,” and “temporary memory.” Illustrations from Our Everyday Life: – Remembering a phone number from directory assistance. – Remembering to pick up the three missing ingredients for the recipe you’re making for tonight’s dinner. – Remembering the directions you got from the gas station to get to your destination. – Remembering what was said 15 seconds ago to keep you up with the conversation, lecture, or discussion. Our goals: – What is STM? – What does it do, exactly? – How does the mind do it? What are the processes that support STM?

4. 1.2 “Short Term” vs. “Working” Memory These are the two main terms that get used, and they refer to the same part of the overall human information processing system, but they differ in their historical connotation. Quick Overall Characterization: – The first stage of the the input and storage of new information that involves conscious effort and whose content we are aware of (unlike the contents of VSM). – Operationally, it is when no more than 15 – 20 s passes for information retention, very little of which transfers over to long-term storage memory (LTM). Terminological Difference – STM: This term is used when focusing on rapid recall and the memorization of new information. – WM: A “mental workbench” where conscious effort is applied to process information. Example: When word meanings are retrieved from LTM and put together, WM is where this putting together happens.

5. Contents 1.Background 1.1 The Phenomenon of STM 1.2 “Short Term” v “Working” Memory 2.STM 2.1 Capacities and Limits 2.2 Forgetting: Decay and Interference 2.3 Retention: Recall and Rehearsal 2.3 Retrieval 3.Forms of Stored Information 3.1 Verbally Based Codes 3.2 Semantic Codes 3.3 Visual Codes 3.4 Other Codes 4.Working Memory 4.1 Problem with Traditional Definition of STM 4.2 Components of Working Memory 4.3 Support for Baddeley’s Model

6. 2.1 STM: Its Limits and Capacity The Bottleneck Phenomenon of STM: – There is only so much that this information retrieval system can hold; if you try to give it more than it can handle, it loses accuracy. Example: Number of random digits we can immediately recall is around seven. Like a juggler who can competently juggle seven balls, but lose all the balls if even one more is tossed in. – This is quite unlike LTM, which is also limited, but can store vastly more than STM. Overcoming the Bottleneck: – Chunking: By grouping the informational units together into “chunks” as small as one unit or larger, it becomes easier for us to retain the information. Ex: 3-4 groupings of phone numbers, 3-2-4 of SS numbers. – Closure: The successful formation of a chunk.

7. Bottleneck and Recoding: – Miller 1956 famously claimed that number of chunks STM can handle is 7 +/– 2. (But we will see later that this entire way of thinking about the capacity of STM rests on an oversimplification.) Recoding: When STM is called upon to handle an ever increasing amounts of information, the information in the existing chunks gets further enriched and regrouped. Recoding thus allows to retain the small number of units. Recoding is not always easy; it sometimes takes considerable mental effort to regroup the increasing information so that you can still retain everything accurately. But it is quite ubiquitous in our everyday mental activities. But it is quite ubiquitous in our everyday mental activities. Involvement of LTM in Recoding: – Two important things for recoding: We need sufficient time or mental resources for recoding. We can recode if the recoding scheme is well-learned or over-learned, like a mnemonic device. This calls upon LTM for two things: – The specific facts – The recoding strategy

8. 2.2 Loss in STM: Interference and Decay Brown-Peterson Task: How does information in STM get lost? Experiments conducted by Brown 1958 and Peterson and Peterson 1959 – now called the Brown-Peterson task – tested for loss of information in STM. – The Experiment: Subjects were given a simple three-letter stimulus, followed by a three-digit number. This insured that STM was never overloaded (units never exceeded 2 chunks). Subjects were told to attend to the stimulus, then to count backwards by threes from the number given. – Result: Memory of the three-letter stimulus was only slightly better than 50% after 3 s of backwards counting, and gradually dwindled to 5% after 18 s of counting. – Analysis: Decay or interference? Brown 1958 and Peterson and Peterson 1959 hypothesized that it was due to decay. – Decay: Forgetting caused by passage of time before rehearsal. The information is not engaging the attention mechanism.

9. Waugh and Norman 1965 argued that the loss was due to active interference, not just the prevention of rehearsal. – Interference: The initial three-letter stimulus for forgotten because the distractor task (counting backwards) used the same attention mechanism. In addition, it turns out that there are two types of interference, depending on when the loss is caused to occur: Proactive Interference (PI): When older material interferes in time with your recollection of the current stimulus. Ex: Continued trials of the Brown-Peterson task caused performance to worsen with each task. Retroactive Interference (RI): When newer information interferes backward in time with the recollection of older items. Ex: Incoming task or information “erases” previously deposited information. – Release from Proactive Interference (PI): Wickens 1972 noted that switching to a new stimulus released subjects from PI. To show this Wickens switched to a new stimulus on the 4 th trial of the Brown- Peterson task ; though the recall declined over the first three trials, subjects could recall the new stimulus with 90% accuracy.

10. Simple decay is not readily testable. Why? New Decay Theory: This doesn’t mean that interference is the only way for information to get lost in STM. Altman and Gray 2002 have proposed that decay is a forgetting process that helps the cognitive system deal with PI. – Experiment: Subjects were shown a single number at a time, drawn from number 1 – 4 and 6 – 9. Subjects were then told to judge the number according to certain instructions – odd/even instruction or small/large instruction. The group that switched instructions frequently had a higher forgetting rate than the group that used the same instruction. – Result: With the higher rate of instruction change, successfully remembering the current information depends upon successful decay of recent information. More generally, if the likelihood of interference increases (e.g. lots of speed limit changes over a short period of time), then the rate of decay-based forgetting increases.

11. 2.3 Retention in STM: Recall and Rehearsal How do we access the information in STM when it does not face interference or decay? There are a couple of ways. One is through rehearsal and the other recall. Studies involving recall have revealed some interesting features of STM: – Free vs. Serial Recall: When people are given a list of items to store in STM, they find free recall easier than serial recall. Free Recall: Recall of items in any order that comes to mind. Serial Recall: Recall of items in the order of presentation. This is harder because it calls for focus on the items to be recalled as well as their position in the list. – Serial Position Effects: In assessing the capacities of STM, a common test is the serial position test, a list of items presented in an order. Primacy Effect: Accuracy of recall for the early items on the list. Recency Effect: Accuracy of recall for final items on the list.

12. – Analysis: Subjects showed accuracy for recalling the latest items on the list. Recalling the items in the middle portion had the lowest accuracy. These results are probably due to the fact that there is no interference to erase the memory of later items are in STM for immediate recall, earlier items get a chance to get rehearsed and stored in LTM for recall, and the middle items do not get a chance to be rehearsed at all. If we want to understand what STM is for, what function it serves, we look to its ability to rehearse material. – Rehearsal: The mental repetition or practicing of material to be learned, whereby through the process of repetition, the material in gets briefly held in STM or transferred over to LTM. – The Rehearsal Buffer: This term is used to describe part of what STM is for, namely, to serve as a mental recycling system for holding information temporarily. On standard interpretations, it is an optional control process invoked by STM having two main properties : Recycling: Rehearsal that maintains a short-term memory trace for a brief period of time. Transferring to LTM: Rehearsal that transfers the information for long-term storage.

13. 2.4 STM: Retrieval To understand how information is accessed from STM, psychologists have used the method of measuring reaction times (RT) for performing certain mental tasks and then making an inference about the nature of the underlying mechanisms based upon the patterns of observed RTs. Sternberg Account of STM Retrieval – E xperiment: In what has come to be known as the “Sternberg Task” (Sternberg 1966, 1969, 1975), subjects were given a short list of letters ranging from 1 to 6 letters (or numerals), called “the memory set,” and asked to determine whether a new letter, called “the probe” was a part of the given memory set or not. (Sample Sternberg Task) TrialMemory SetProbeResponse 12341234 L G S N M B H Y R K O P V LNMPLNMP YYNYYYNY

14. – Results: A linear increase in RT as the memory set got larger, for both “Y” and “N” responses, with each mental scanning process taking just under 38 ms, which is impressively fast. In other words, the search rate through STM is about 38 ms per item. – Sternberg’s Model of STM Retrieval: Sternberg proposed the following flow chart to describe the mechanism for STM retrieval: (Adapted from Sternberg 1969) START TIME 38 MS PER CYCLE – Analyses: Three hypotheses about how the search process / scanning happens: Serial Self-Terminating Search: The items are scanned one by one until a match is found. Not predictively accurate: If correct, the “N” responses should take longer than “Y” responses since all items need to be scanned for a “N,” but the data didn’t match this prediction; the “Y”s were no faster than the “N”s. Memory Set Items Encode Probe Search Process: Scan Memory Set and Compare with Probe Decide: “Y” / “N” Motor Response

15. Parallel Search: All the items in the memory set are scanned simultaneously – “in parallel.” Again not predictively accurate: If correct, then it should take the same amount of time to scan 6 items as it does 2, but the data didn’t bear this out; bigger memory set took longer to scan. Serial Exhaustive Search: The memory set is scanned one at a time (serially), and the entire set is scanned on every trial, both for when a match was or was not found (exhaustive). Since the times for “Y” and “N” responses were pretty much the same, Sternberg argued that both used the same search method – the serial exhaustive search. – Alternative Analyses Parallel Search: Some have argued that the the longer RT for bigger memory sets may be due to an overall diminishment of parallel processing with each increase in items to be scanned. Overlapping Mental Stages: As can be gleaned from the model, Sternberg assumes that the general retrieval process is a sequential one, with one task to be completed before the next one begins. McClellan 1979 has proposed that the tasks may overlap and proceed in a cascade fashion.

16. Contents 1.Background 1.1 The Phenomenon of STM 1.2 “Short Term” v “Working” Memory 2.STM 2.1 Capacities and Limits 2.2 Forgetting: Decay and Interference 2.3 Retention: Recall and Rehearsal 2.3 Retrieval 3.Forms of Stored Information 3.1 Verbally Based Codes 3.2 Semantic Codes 3.3 Visual Codes 3.4 Other Codes 4.Working Memory 4.1 Problem with Traditional Definition of STM 4.2 Components of Working Memory 4.3 Support for Baddeley’s Model

17. 3.1 Verbally Based Codes Verbally Based Codes: STM can handle information in many forms – sensory, verbal, semantic, and others. – Experiment: Conrad 1964 found that when subjects were presented with a string of letters visually, the errors they made in recall suggested that they processed the information verbally or aurally rather than visually. – Results: In the errors, subjects would substitute the more similar- sounding “D” for “P” rather than “F” for “P,” even though “F” is more visually similar to “P.” – Analysis: The nature of these errors suggest that subjects code certain information in a verbal form.

18. Semantic Codes – Experiment: Wickens 1972 ran some variations upon the Brown- Peterson task to examine how STM might use semantic codes. All subjects had 4 successive B-P trials, each trial consisting of a triad of words to be remembered. The words in each trial were grouped into a) vegetables, b) fruits, and c) professions. – Results: Subjects showed a stronger release from PI when the material switched to a more different topic – from fruits to professions than from fruits to vegetables. The control group ended their fourth trial with no switch, and displayed the effects of PI with their continual decline in recall. – Analysis: These results seem to suggest that the information, in these trials, is coded in linguistically meaningful terms, since the strength of the release and level of performance tracks semantic differences in the remembered words. 3.2 Semantically Based Codes

19. Visual Codes – Experiment: Shepard and Metzler (1971) and Cooper and Shepard (1973) asked subjects to compare perspective drawings in two forms and judge whether they were the same. – Result: The more the forms varied in angular rotation (120 o as opposed to 60 o ) the longer it took for them to mentally rotate the image. – Analysis: This seems to be evidence for holding a mental image in STM. 3.3 Visually Based Codes

20. Other Sensory Codes – Hearing – Touch – Odor and Taste Kinesthetic Code – Example 1: Knowing what it physically feels like to ride your bicycle. – Example 2: Deaf persons who know both English and ASL made errors based on cherological similarity in ASL rather than acoustic similarity in English. 3.4 Other Types of Codes

21. Contents 1.Background 1.1 The Phenomenon of STM 1.2 “Short Term” v “Working” Memory 2.STM 2.1 Capacities and Limits 2.2 Forgetting: Decay and Interference 2.3 Retention: Recall and Rehearsal 2.3 Retrieval 3.Forms of Stored Information 3.1 Verbally Based Codes 3.2 Semantic Codes 3.3 Visual Codes 3.4 Other Codes 4.Working Memory 4.1 Problem with Traditional Definition of STM 4.2 Components of Working Memory 4.3 Support for Baddeley’s Model

22. 4.1 Problem with Traditional Dfn of STM Capacity to handle several forms of information in different modes calls for notion of “working memory:” – We are capable of juggling several codes at once: auditory + visual + verbal. But patterns in certain errors we make suggest that Miller’s 7 (+/- 2) is much too simple to characterize the workings of STM. Example 1: Complex math problems or syntactically difficult English sentences can place demands on STM that don’t conform to Miller’s rule. Example 2: Patients who can only remember 1 or 2 chunks of information and yet suffer little learning and comprehension impairment. – Solution: Treat traditionally defined STM as one part of a larger system that is known as working memory.

23. 4.2 Components of Working Memory Baddeley’s Model of the WM System: Three major components, a central executive system, and slave systems. ( Adapted from Baddeley 1986) E XECUTIVE C ONTROL S YSTEM (central pool of mental resources) Activities: -Initiate control and decision processes -Integrate incoming information -Reasoning -Language comprehension -Transfer info to LTM -Planning future actions -Recency effects V ISUO - SPATIAL S KETCHPAD Activities: -Visual imagery tasks -Spatial, visual search tasks -(Difficult visual tasks drain ECS) A RTICULATORY R EHEARSAL L OOP (short-term buffer) Activities: -Recycling items for immediate recall -Articulatory Processes -(Difficult articulation tasks drain ECS)

24. Components of the Model – Executive system, somewhat like the Supervisory Attentional System (SAS), the most cognitively involved attentional mechanism under the broad heading of “Controlled Attention.” – Slave Systems: Responsible for fairly low-level processing. Domain specific, i.e., specialized for handling specific types of information or codes. Have their own pool of attentional resources but are very limited; if it is overburdened, it eats into the resources of the executive control system.

25. Baddeley’s Use of the Dual Task Method – This method, originally used for tracking attention upon one of two (or more) incoming stimuli. – Here, it is used to track which component of WM is dedicated to the simultaneous tasks and how the components, in action, influence each other. Brown-Peterson Task Brook’s capital “F” study – Working Memory and Other Mental Functions WM and Reasoning WM and Language Comprehension WM and Visual Processing Several details may be amended to the basic model – invoke more slave systems, for instance. The research is till working on this. 4.3 Support For Baddeley’s Model