Chapter 5: Short-term Memory and Working Memory

In this chapter, we explore the intricacies of short-term memory (STM) and working memory (WM), two fundamental concepts in cognitive psychology. While these terms are often used interchangeably, they refer to different aspects of how we temporarily store and manipulate information.

Information Processing Model

To understand STM and WM, it is essential to grasp the broader framework of the information processing model (see Figure 5.1). This model outlines how information is taken in, processed, and stored in the brain. It includes:

1. Sensory Memory: The initial stage where sensory information is briefly held.
2. Short-term Memory (STM): The system for temporary storage of information.
3. Long-term Memory (LTM): The system for more permanent storage of information.

Sensory Memory Recap

Before diving into STM and WM, let's briefly recap sensory memory. Sensory memory holds information from the senses for a very short duration. For example:

• Iconic Memory: Visual sensory memory, large in capacity but brief in duration (about 250 milliseconds), and pre-categorical.
• Echoic Memory: Auditory sensory memory, which lasts about four seconds and may contain some semantic information.

Figure 5.1. This figure shows the standard modal model of information processing where information from the outside world enters sensory memory and then this information flows to short-term memory. Information from short-term memory can be sent or retrieved from long-term memory and attention is presented in an oval, rather than a box, to illustrate that this is a process and not a structure. Responses are made from short-term memory.

"Modal model with STM highlighted." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0

Short-term Memory (STM)

Definition and Characteristics

STM is a limited-capacity store that holds information temporarily. It is crucial for tasks that require the retention and manipulation of information over short periods.

• Capacity: George Miller's famous "7 ± 2" theory suggests that STM can hold about seven chunks of information, where a chunk is a meaningful unit of information.
• Duration: Without rehearsal, information in STM seems to last about 18-30 seconds. However, determining the size and duration of STM is difficult because STM is affected by interference (which is a point we return to).

Chunking

Chunking involves organizing information into meaningful groups, which enhances the amount of information we can retain in STM.

Example of Chunking

Consider the following sequence of letters: IRS CIA NBA FBI USA. When presented as individual letters (or meaningless groups like IR, SCI, ANB, etc), this sequence is challenging to remember. However, when chunked into meaningful units (IRS, CIA, NBA, FBI, USA), it becomes easier to retain since this is only 5 chunks.

Chunking and Expertise

Expertise in a domain can significantly enhance one's ability to chunk information. For example, chess experts can remember the positions of pieces on a board much better than novices, but only when the pieces are arranged in a meaningful pattern.

Chess Example

A chess expert shown a board for a brief period can recall the positions accurately if the arrangement is logical. However, if the pieces are placed randomly, their performance drops to the level of non-experts. This demonstrates the role of chunking and long-term memory in expertise.

Forgetting in STM

Forgetting in STM can occur due to two primary reasons:

1. Decay: The theory that information fades over time such that time causes the forgetting.
2. Interference: More compelling evidence suggests that forgetting is primarily due to interference rather than decay (it is not time that causes forgetting but the events that happen during the passage of time).

Types of Interference

• Retroactive Interference: New information interferes with the memory of old information.
• Proactive Interference: Old information interferes with the memory of new information.

Evidence for Interference

• Brown-Peterson Task: Participants are given three letters to remember, followed by a distraction task (e.g., counting backward by 3s). Performance drops significantly after about 18 seconds of counting backward. This result could be caused by decay where forgetting increases as more time passes. However this could also be caused by interference. Retroactive interference in this task could occur if the numbers from the counting task (new information) makes it difficult to retrieve the letters that had been given (old information). Additionally, since people perform many trials in a Brown-Peterson task, rather than a single memory trial, proactive interference occurs if the letters from earlier trials (old information) makes it difficult to retrieve the letters that had been given most recently (new information).

Probe Digit Task

Participants hear a list of digits at different rates (4 digits per second or 1 digit per second) and must recall what follows a repeated digit. Results support the role of retroactive interference, as memory declines with the number of intervening items, not with the passage of time alone. So, if a person hears the digits: 4, 9, 14, 16, 13, 8, 1, 6, 12, 3, 10, 5, 2, 7, 15, 11, and 3 then the correct response is 10 since 10 follows the repeated 3 (see Figure 5.2). Note that this item (10) was presented with 5 interfering digits and the number of digits can be manipulated. In addition, if this were shown at a rate of 4 digits per second the item to recall had been presented closer in time than if the items are presented at a slower rate. Results show that as the number of interfering items increases performance declines and the rate of presentation does not matter (see left-hand side of Figure 5.3). If time were the cause of forgetting, performance would be the same across speed conditions when the same amount of time had elapsed. For example, 1 interfering item shown at a rate of 1 per second and 4 interfering items shown at a rate of 4 per second both occurred 1 second ago, so performance would be equivalent in these cases. Additionally, performance would improve when less time had passed compared to when more time had passed. (see right-hand side of Figure 5.3).

Figure 5.2. This image shows an example sequence in the probe digit task.

"Depiction of probe-digit task sequence." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0

Figure 5.3. This image shows the pattern of data in the probe digit task. On the left is the pattern expected if forgetting is caused by retroactive interference. If this happens then accuracy decreases as the number of items between the item and end of the list increases. On the right is the pattern expected in forgetting is caused by time. If this happens then accuracy decreases as time increases. So, accuracy is better in the 4-per-second condition because less time has passed. The data support the conclusion that forgetting is caused by retroactive interference (i.e., the left-hand pattern).

"Depiction of probe-digit task results." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0

Proactive Interference

In addition to retroactive interference, Keppel and Underwood (1962) presented findings that questioned the idea that forgetting in short-term memory is due to decay. Keppel and Underwood suggested that forgetting became more pronounced with additional trials because earlier trials created interference. Essentially, when trying to recall the current trigram, participants might mistakenly retrieve information from previous trials, causing confusion. In tests of this they observed that significant forgetting occurred only after participants completed multiple trials. On the first trial, participants’ memory for the three-letter trigram was terrific. This phenomenon, known as proactive interference (PI), occurs when older information interferes with the ability to recall newer material.

Release from Proactive Interference

Wickens and colleagues demonstrated that proactive interference can be overcome by changing the type of material participants are asked to remember. In their study, participants completed three Brown–Peterson trials using either words or numbers. Initially, performance was high on the first trial, but it dropped significantly by the third trial. On the fourth trial, the researchers switched to a completely different category of items—participants who had been recalling words were given numbers, and those recalling numbers were given words. This change led to a dramatic improvement in performance, with accuracy on the fourth trial returning to the initial levels. The extent of this recovery, known as the "release from proactive interference," depended on how distinct the new items were from the previous ones (see Figure 5.4). For example, participants who initially recalled fruit names experienced a greater release if the fourth trial involved flower names compared to vegetables, and an even larger release if the new items were professions. The greater the semantic difference between the old and new items, the more effective the release from interference, which suggests that short-term memory contains semantic information (i.e., meaning-based information).

Figure 5.4. This image shows results of an experiment testing the release from proactive interference. Here the first three lists contain fruit names and the final list varies from fruit names to professions.        

"Depiction of release from proactive interference." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0

Understanding the mechanisms behind the release from PI is crucial for appreciating how our memory systems function. A classic study by Gardiner, Craik, and Birtwistle helped researchers better understand whether this phenomenon is caused by better encoding of the information, better retrieval from memory, or a combination of the two.

In their study, participants were asked to remember lists of items from different categories. For instance, one group of items might be cultivated flowers (e.g., "carnation," "orchid," "rose"), which are followed by a distractor task such as counting backwards by threes for nine seconds. After repeating this process with similar lists from the same category, participants received a new list with a subtle category shift, such as wildflowers (e.g., "dandelion," "poppy," "buttercup"). This shift aimed to see if participants would experience a release from PI, indicated by improved recall performance.

The study involved three groups of participants:

1. Control Group: This group was not informed about the category shift. As expected, their performance showed a typical buildup of PI, with decreasing recall accuracy across trials.
2. Informed Group: This group was informed about the category shift from the beginning. They knew that the last set of flowers would be wildflowers and the first three would be cultivated flowers. The knowledge potentially allowed for better encoding of the information when it was given but this also could allow for better retrieval from memory (remember the wildflowers this time), resulting in a noticeable release from PI.
3. Post-encoding Informed Group: This group was told about the category shift only after they had encoded the last list but before they recalled the last list. This scenario tested whether the release from PI was due to encoding or retrieval processes. If the release from PI were caused by encoding then this should not show a release from PI because then encoded the list just like the control group. However, if the release from PI were caused by retrieval then this should show a large release from PI, like the group that was informed from the start.

The findings revealed that the Post-encoding Informed Group performed similarly to the Informed Group, indicating that the release from PI is primarily a retrieval-based phenomenon. The critical point is that participants could better discriminate between cultivated and wildflowers during retrieval, thereby reducing PI. In other words, having this information helped them to retrieve the correct memories.

Serial Position Effect

When people are given a list of items to remember, the serial position effect emerges, showing better recall for items at the beginning (primacy effect) and the end (recency effect) of the list, compared to items in the middle (see Figure 5.5).

1. Primacy Effect: Items presented at the beginning of a list are recalled more accurately because they are more likely to be transferred to long-term memory. During the initial presentations, participants can focus more on each item, enhancing encoding and rehearsal.
2. Recency Effect: Items at the end of the list are still in short-term memory at the time of recall, making them more accessible. This effect is evident when recall occurs shortly after presentation.

Figure 5.5. This image shows results of an experiment testing the serial position effect. The elevated performance for the first few items (primacy effect) is presumably caused by retrieval from LTM while the elevated performance for the last few items (recency effect) is presumably caused by retrieval from STM.        

"Depiction of serial position effect." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0

The distinct explanations for these effects become evident under various conditions:

• If a distractor task, such as counting backwards, is introduced after the list and before recall then the recency effect diminishes, as the short-term memory trace is disrupted.
• The primacy effect remains intact because those items have been consolidated into long-term memory.

Serial position effects occur in various memory tasks, including recalling personally meaningful events (episodic memory) and general facts (semantic memory). For example, when asked to recall phone numbers from different

periods of their life, people often remember their first and most recent numbers easily but struggle with those in between. Similarly, when tasked with naming U.S. presidents, individuals tend to recall the first and most recent presidents more easily than those in the middle.

In these examples, the primacy and recency effects are evident, but they do not seem to involve distinct memory systems. This challenges the typical explanation that primacy results from long-term memory storage while recency is due to short-term memory retrieval. Instead, these findings suggest that other factors may contribute to serial position effects in general.

Scanning Short-term Memory

Saul Sternberg's experiment provides insights into how we scan items stored in short-term memory. Participants were asked to memorize a set of items (e.g., letters) and then decide if a probe item was part of the original set. To do the task participants needed to scan their memory to determine (yes or no) whether the probe letter was one of the letters they were holding in memory. Sternberg proposed three potential methods of scanning (see Figure 5.6):

1. Serial Self-terminating Search: Items are scanned one at a time, and the search stops as soon as the target is found. This method predicts a linear increase in reaction time (RT) with the number of items, but the RT for "yes" responses would be about half the slope of the "no" responses, as the search terminates early for positive matches.
2. Parallel Search: All items are scanned simultaneously, leading to a constant RT regardless of the number of items. The "yes" and "no" response times should overlap, showing no dependence on the set size.
3. Serial Exhaustive Search: All items are scanned one by one, and the search continues to the end regardless of when the target is found. This method predicts a linear increase in RT with the number of items, with "yes" and "no" responses showing similar slopes, indicating no early termination.

Figure 5.6. This image shows results from Sternberg’s memory scanning task if scanning takes place in a serial serial-self-terminating (top), parallel (middle), or serial-exhaustive (bottom), manner.        

"Depiction possible results from Sternberg’s memory scanning task." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0

Sternberg's findings supported the serial exhaustive search model, where reaction times increased linearly with the number of items, and "yes" and "no" responses were virtually identical. This result suggests that even if the target is found early, the search process continues through all items in the set, possibly due to a fixed cognitive strategy or a requirement to ensure accuracy.

In addition, the equation that describes the pattern was RT = 38X + 397. You may remember that in the equation of a line, 38 is the slope and 397 is the y intercept. Since the x axis is set size, the slope tells us that for every item added to memory, it takes an additional 38 ms. Stated differently, it takes people 38 ms to scan each item in memory (this is very fast). The y-intercept here is then the time needed to do everything else. In Sternberg’s model he felt people need to encode the probe letter (i.e., identify what I am searching for), scan memory, arrive at a yes/no decision, and then execute the motor response. These steps are shown in Figure 5.7 as stages A, B, C, and D. So, for every item added to memory it takes 38 ms longer to complete step B (the scan stage), and it takes 397 ms to complete steps A, C, and D (the encoding, decision, and response stages).

Figure 5.7. Sequence of stages in Sternberg’s memory scanning task.        

"Sequence of stages in Sternberg’s memory scanning task." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0

Age and STM

Research has shown that aging affects various stages of cognitive processing, including encoding, scanning, and decision-making. Older adults exhibit slower encoding, which results in a higher y-intercept, and scanning rates also increase, resulting in steeper slopes. This suggests that aging impacts the efficiency of STM processes across the board.

Concussions and STM

Concussions or closed head injuries have been found to specifically affect the encoding and decision stages of STM, without altering the scanning rate. This leads to a larger intercept in reaction time tasks but no change in the slope, indicating that while the initial encoding takes longer, the scanning process remains unaffected.

Working Memory (WM)

Definition and Characteristics

WM extends the concept of STM by incorporating the manipulation of information. It is not just a passive store for holding information but an active system involved in manipulating the information for use in various tasks.

Differences Between STM and WM

• STM: Focuses on the temporary storage of information.
• WM: Includes both the storage and manipulation of information, integrating attention. In this way, some people might consider working memory to be short-term memory plus attention.

Working Memory

Working memory (WM) represents a more dynamic understanding of short-term information processing, emphasizing the manipulation and active maintenance of information. This concept was developed to address the limitations of the traditional STM model, highlighting the interplay between memory and attention.

Baddeley's Model of Working Memory

Alan Baddeley's model of WM includes three main components: the central executive, the phonological loop, and the visuospatial sketchpad (see figure 5.8). A fourth component, the episodic buffer, was later added. This is the part of working memory that integrates information from various modalities and sources to recall old episodic memories and create new episodic memories.

1. Central Executive: This is the control system responsible for attention, decision-making, and coordination of the subsidiary systems. It allocates resources and manages tasks such as problem-solving and planning.
2. Phonological Loop: This subsystem is specialized for verbal and auditory information. It consists of:

1. Phonological Store: Holds sound-based information for a short duration.
2. Articulatory Control Process: Rehearses and refreshes the information in the phonological store, akin to an "inner voice."

3. Visuospatial Sketchpad: Manages visual and spatial information, allowing for the manipulation of images and spatial relationships. It is crucial for tasks such as navigation and mental rotation.
4. Episodic Buffer: Integrates information from the phonological loop, visuospatial sketchpad, and LTM, creating a coherent episodic representation. It allows for the integration of different types of information and supports the formation of long-term memories.

Figure 5.8. Visual depiction of Baddeley's model of working memory. The model has three components: the visual-spatial sketchpad, the phonological loop, and the central executive. Here the words dog, table, and shoe are being sub-vocally repeated and held in the phonological loop.        

"Baddeley's model of working memory." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0

Evidence Supporting the Phonological Loop

Several phenomena support the existence and function of the phonological loop:

1. Phonological Similarity Effect: Memory performance is poorer for items that sound similar (e.g., P, G, T) compared to items that sound different (e.g., R, H, X). This suggests that items are stored in a phonological code and similar-sounding items are more likely to be confused.
2. Word Length Effect: Shorter words are remembered better than longer words. This is because shorter words can be rehearsed more quickly, allowing for more efficient use of the phonological loop.
3. Unattended Speech Effect: Background speech, especially if phonologically similar to the items being remembered, can disrupt memory performance. This indicates that irrelevant auditory information can enter the phonological store and interfere with the rehearsal process.
4. Articulatory Suppression: Repeating an irrelevant sound (e.g., "the") while trying to remember items disrupts memory performance. This prevents the articulatory control process from rehearsing the items, leading to rapid decay of the information in the phonological store.

Dual-Task Studies

Baddeley's dual-task studies demonstrate the independence of the phonological loop and visuospatial sketchpad. For instance, participants can perform a verbal task and a visual task simultaneously with minimal interference, suggesting separate resources for each subsystem. However, two tasks that tap into the same subsystem (e.g., two verbal tasks) lead to significant performance decrements, indicating competition for the same limited resources.

Visuo-spatial sketchpad

The visuo-spatial sketch pad is a component of working memory specialized for processing and maintaining visual and spatial information. It enables tasks like holding mental images or determining where a location is on campus when you seated in a classroom (e.g., for students at Bates College you can point in the direction of the puddle when you are indoors and to do this you are using your visuo-spatial sketch pad). Key phenomena illustrate its properties, including Kossly’s studies of mental imagery and studies of mental rotation.

Kosslyn's Mental Imagery Experiments

Stephen Kosslyn's research on mental imagery has significantly contributed to our understanding of how visual information is processed in short-term memory. Kosslyn examined how the time it takes to respond to a question depends on the mental distance one has to scan within an image. For instance, when individuals learn the layout of an island and later respond to questions about its features, their reaction times correlate with the distance they need to mentally traverse the map. This finding suggests that mental imagery involves spatial representations similar to those used in actual perception.

Kosslyn's experiments further explored mental image size by having participants visualize a rabbit next to either an elephant or a fly. When asked questions about the rabbit, participants imagining it next to a fly responded faster than those imagining it next to an elephant. The difference in reaction times indicates that the perceived size of the mental image affects how quickly details can be accessed, highlighting the dynamic nature of visual short-term memory.

Mental Rotation Experiments

 One of the most compelling pieces of evidence for the visuo-spatial sketch pad's role in working memory comes from studies on mental rotation. In these studies, participants are shown pairs of three-dimensional objects and asked to determine whether they are the same or different (see figure 5.9). The challenge lies in the fact that the objects are presented at different angles, requiring the participant to mentally rotate one object to match the orientation of the other in order to make the same/different decision. The results from these studies consistently show that the more an object needs to be rotated, the longer it takes for the participant to make a judgment (see figure 5.10). For example, it takes more time to compare objects rotated by 120 degrees than by just 60 degrees, mimicking the time it would take to physically rotate the objects.

Figure 5.9. Example of three-dimensional objects presented at slightly different angles. Participants are asked to determine whether they are the same or different. Here the two objects are the same.

"Three-dimensional objects." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0 

Figure 5.10. Pattern of results from mental-rotation task. Reaction times linearly increase with the angle of rotation.

"Results from mental rotation task." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0

Selective Interference Effects, Cognitive Resources, and Double Dissociations

Research on selective interference effects has provided compelling evidence for the separation of cognitive resources dedicated to different types of memory tasks. One classic example involves the rotary pursuit task, which requires participants to keep a wand on a moving dot. This task significantly disrupts the ability to perform mental rotations but has minimal impact on holding a list of words in memory. Conversely, articulatory suppression, which involves repeating an irrelevant sound, impairs the retention of verbal information but does not affect mental rotation tasks.

This double dissociation supports Alan Baddeley's model of working memory (see Figure 5.11), which posits distinct subsystems: the visuospatial sketchpad for visual and spatial information, and the phonological loop for verbal information. Rotary pursuit interferes with the visuospatial sketchpad, while articulatory suppression disrupts the phonological loop, confirming their functional independence.

Figure 5.11. Table showing double dissociations where task A selectively interferes with one cognitive ability 1 and task B selectively interferes with one cognitive ability 2. Below this it is shown that rotary pursuit selectively interferes with mental rotation while articulatory suppression selectively interferes with holding words in memory.

"Double dissociations." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0

Double Dissociation in Experimental Tasks

Further evidence for separate cognitive resources comes from studies by Seagal and Fusella, who investigated the interference effects of forming mental images on the detection of faint signals. Participants imagining a visual scene, such as a tree, were less proficient at detecting faint visual signals but performed well in detecting auditory signals. Conversely, those imagining an auditory scene, like a ringing telephone, showed the opposite pattern. These selective interference effects reinforce the notion of distinct subsystems within working memory, each specialized for different types of information processing (visual/spatial vs. phonological/auditory).

The Central Executive and Complex Decision-Making

The central executive, as proposed by Baddeley, plays a crucial role in managing and coordinating cognitive processes, particularly in complex decision-making tasks. It functions as an overarching control system that allocates resources between the phonological loop and the visuospatial sketchpad. For example, in tasks such as playing chess or registering for classes, the central executive integrates visual and verbal information, strategizes, and makes decisions based on multiple factors.

Although empirical support for the central executive is less extensive compared to other components of working memory, its importance is evident in tasks requiring the integration and manipulation of diverse types of information.

The Episodic Buffer: Integrating Information and Sequencing Events

Baddeley's model also includes the episodic buffer, a more recent addition that facilitates the integration of visual, spatial, and verbal information with temporal sequences. The episodic buffer is essential for tasks that require the coherent sequencing of events, such as storytelling. Research with amnesic patients demonstrates that despite impairments in verbal and/or visual working memory, individuals can often recall and sequence events correctly, indicating the robustness of the episodic buffer.

Attentional Refreshing: A New Component of Working Memory

Valerie Camos' concept of attentional refreshing introduces yet another dimension to our understanding of working memory. Refreshing involves briefly bringing information back into focus without active rehearsal. This process enhances memory performance, suggesting that attentional mechanisms play a vital role in maintaining information in short-term memory (see Figure 5.12).

In conclusion, these studies and theoretical advancements illustrate the complex, multifaceted nature of short-term memory. By understanding the distinct cognitive resources and mechanisms involved, we gain deeper insights into how information is processed, maintained, and manipulated in our minds.

Figure 5.12. Baddeley's model of working memory with all components (central executive, phonological loop, visuo-spatial sketchpad, and episodic buffer) along with the potential addition of Valerie Camos’ concept of attentional refreshing.

"Depiction of working memory model." by Kahan, T.A. is licensed under CC BY-NC-SA 4.0