Difficulty: Introductory | Prerequisites: None, though familiarity with basic cognitive psychology concepts (attention, perception) is helpful.
Working memory is one of the foundational topics in cognitive psychology. It sits at the intersection of attention, short-term storage, and executive control, and it underpins nearly every higher cognitive task you will study later in the course, from language comprehension to reasoning to decision-making. This set of notes covers what short-term memory is, how it differs from long-term memory, the major models (Baddeley's multi-component model in particular), and the key experimental evidence you are expected to know. If you have missed the earlier lectures on attention and perception, you can still follow this material, but you will get more from it if you understand Sperling's iconic memory experiments and the basics of how attention filters information.
Short-term memory holds roughly 7 (plus or minus 2) chunks of information and lasts only about 10 seconds without rehearsal. Working memory is the broader system that actively manipulates that information, and Baddeley's model breaks it into three components: a phonological loop for verbal/acoustic material, a visuospatial sketchpad for images and spatial layouts, and a central executive that directs attention between them. The phonological loop handles linguistic information specifically, not just sound, as shown by research with deaf signers whose memory errors follow sign-form similarity rather than acoustic similarity.
Short-term memory (STM)
A temporary storage system that holds a small amount of information in an accessible state for a brief period (roughly 10 seconds without rehearsal). Think of it as the mental space where you keep a phone number just long enough to dial it.
Working memory (WM)
The broader cognitive system responsible for temporarily holding and actively manipulating information. In simple terms, STM is the storage; working memory is the storage plus the processing.
Chunk
A unit of meaningful information in STM. A chunk can be a single letter or an entire word, because what counts as one "item" depends on how you organise the material. Think of it as a mental container whose size adjusts to meaning.
Rehearsal
The process of repeating information to keep it active in STM. In simple terms, it is the mental equivalent of juggling: the items stay up only as long as you keep cycling through them.
Phonological loop
The component of Baddeley's working memory model that maintains verbal and acoustic information through subvocal rehearsal. Think of it as your inner voice repeating words on a loop.
Visuospatial sketchpad
The component of Baddeley's model that maintains visual and spatial information. In simple terms, it is your mind's eye, holding a brief mental image or spatial layout.
Central executive
The attentional control component of Baddeley's model that coordinates the phonological loop and visuospatial sketchpad, directing resources where they are needed.
Proactive interference (PI)
When previously learned information interferes with the ability to remember new information. Think of it as old memories getting in the way of forming new ones.
Word-length effect
The finding that words taking longer to pronounce are harder to remember in STM, because they take more time to rehearse in the phonological loop. In simple terms, shorter words are easier to juggle.
Brown-Peterson paradigm
An experimental method for measuring STM duration: participants hear items to remember, count backwards (to prevent rehearsal), then try to recall the items. It showed STM lasts roughly 10 seconds without rehearsal.
George Miller's landmark paper established that STM can hold roughly seven items (plus or minus two). This limit shows up across short-term memory tasks, attention tasks, and identification tasks alike. Sperling's experiment on iconic memory demonstrated this constraint: participants could not report all letters in a briefly flashed display, suggesting a bottleneck in how much reaches short-term storage.
The crucial question is what counts as an "item." STM can hold 7 individual letters or 7 entire words, because each item is a chunk of meaningful information. Chunking is the process of grouping smaller units into larger meaningful units, and it effectively expands STM capacity. For example, the letters F, B, I, C, I, A are six items, but if you recognise them as FBI and CIA, they become two chunks, freeing slots for other material.
STM represents information primarily in an acoustic or verbal format, maintained through rehearsal (silently repeating material to yourself).
Conrad (1964) study: Even when items were presented visually, recall errors were acoustic in nature. Participants confused letters that sound alike (e.g., C recalled as P, not O) and words that rhyme (e.g., "boat" recalled as "vote," not "ship").
Conrad also found that lists of acoustically similar letters (e.g., B, C, D, G, P, T) were harder to remember than acoustically distinct letters, further supporting acoustic coding in STM.
This pattern of acoustic confusion is strong evidence that STM relies on a sound-based code, regardless of whether the original input was visual or auditory.
STM lasts as long as you keep rehearsing. The juggling metaphor is useful here: items remain in STM as long as you actively cycle through them, and they drop as soon as you stop.
What happens when rehearsal is prevented?
The Brown-Peterson paradigm tests this directly:
Hear three letters to remember
Count backwards for a set interval (this prevents rehearsal)
Try to recall the three letters
Results show a steep decline in recall accuracy as the counting interval increases. Without rehearsal, STM contents decay in roughly 10 seconds. By 18 seconds, recall is near floor.
A common misunderstanding about "bad short-term memory": People who complain about having poor short-term memory are almost never talking about STM in the technical sense. They usually mean they cannot remember names or facts after 5 or 10 minutes, which is a long-term memory retrieval problem, not a short-term storage problem.
Baddeley proposed that working memory is not a single store but a system with three components:
Phonological loop: Maintains verbal and acoustic information through subvocal rehearsal. This is the component doing the work when you repeat a phone number to yourself. It is responsible for the acoustic coding Conrad demonstrated.
Visuospatial sketchpad: Maintains visual and spatial information, like a brief mental image. Think of it as a quick visual flash you can hold and inspect.
Central executive: The attentional control system that coordinates the other two components. It decides which type of maintenance to engage and allocates processing resources.
The model explains why you can simultaneously hold a mental image and rehearse a phone number (different subsystems) but struggle to rehearse two verbal lists at once (same subsystem, the phonological loop, overloaded).
Word-length effect: Longer words (or words that take longer to pronounce) are harder to remember in STM. For instance, recall is better for short-vowel words like "bit" than for long-vowel words like "beet." This is evidence for the articulatory loop specifically, because the bottleneck is pronunciation time, not the number of items. STM capacity is constrained by the amount of material you can rehearse, not simply the count of items.
What is the phonological loop for?
It is probably tied to language processing and language learning. An interesting test of this idea comes from studying congenitally deaf individuals who use sign language. They clearly have no "phonological" loop in the auditory sense, but the question is whether they have a corresponding linguistic memory system.
This study tested both hearing and deaf participants who used sign language.
Hearing subjects heard words; deaf subjects saw the same words in sign.
Both groups recalled items by writing down the English word.
Results:
Hearing speakers made the expected acoustic/phonological errors (e.g., cook recalled as crook, boat as vote).
Deaf signers made errors based on sign-form similarity, not meaning. Around 70% of their errors involved signs that differed by only one component (e.g., name confused with egg, socks with star, bird with newspaper). These pairs look nothing alike semantically but are nearly identical in how the hands move.
Conclusions:
The phonological loop is designed to handle linguistic information, not sound per se.
For hearing people, linguistic information is coded in terms of pronunciation.
For deaf signers, linguistic information is coded in terms of sign components (hand shape, movement, location). Their representations are still linguistic in nature; the "pronunciation" is physical rather than auditory, but the underlying principle is the same.
This is an important finding because it shows the loop is a linguistic system, not merely an acoustic one.
Proactive interference (PI) is the phenomenon in which previously learned semantic information intrudes on current STM performance. This seems to suggest that meaning plays a role in STM, but the picture is more nuanced.
Keep in mind that any short-term recall task can draw on both the articulatory loop and long-term memory. Long-term memory does involve meaning, and that is probably where the PI effect comes from. STM itself probably does not encode meaning.
Key takeaway: No memory task likely involves just one kind of memory. Working memory is involved in most cognitive tasks, and WM measures often predict individual differences in task performance as well as IQ. Adding a working-memory load to any task generally hurts performance on that task.
Students often think "short-term memory" and "working memory" are the same thing. They are not. STM is a passive store; working memory is the active system that includes STM plus executive processing.
Students often confuse STM duration with everyday forgetting. Forgetting someone's name after five minutes is a long-term memory problem, not an STM problem. STM duration is measured in seconds, not minutes.
Students often assume the 7 ± 2 limit means seven individual items regardless of complexity. The limit is on chunks, and chunk size depends on how meaningfully you can group the material.
Students often think the phonological loop is about sound. The Newkirk, Klima and Bellugi study shows it handles linguistic information more broadly, including sign-based representations in deaf individuals.
⚠️ Be able to explain Miller's 7 ± 2 and the concept of chunking. Exam questions often ask you to explain why STM can hold 7 letters or 7 words.
⚠️ Know the Brown-Peterson paradigm in detail: procedure, results, and what it demonstrates about STM duration.
⚠️ Be prepared to draw or describe Baddeley's three-component model and explain what each part does.
⚠️ The Conrad (1964) study and the Newkirk, Klima and Bellugi (1979) study are commonly tested. Know the methodology and what each demonstrates about the nature of coding in STM.
⚠️ The word-length effect is evidence for the articulatory loop specifically, not STM in general. This distinction appears on exams.
⚠️ Understand the relationship between working memory capacity and general cognitive performance (IQ). WM load effects on task performance are frequently tested.
True or False: STM capacity is limited to exactly 7 items regardless of how the items are organised. (False: the limit is on chunks, and chunk size varies with meaningfulness.)
True or False: The Brown-Peterson paradigm prevents rehearsal by asking participants to count backwards. (True.)
Fill in the blank: In Baddeley's model, the ______ coordinates the phonological loop and the visuospatial sketchpad. (Central executive.)
True or False: Conrad's study showed that STM errors are based on visual similarity. (False: errors were acoustic/phonological.)
Fill in the blank: The word-length effect shows that STM capacity depends on the ______ of material, not just the number of items. (Amount, or pronunciation time.)
Q: What is the capacity of short-term memory, and what determines what counts as a single item?
A: STM holds approximately 7 ± 2 items. An item is a chunk of meaningful information. A chunk can be a single letter or an entire word, depending on how the material is organised. Chunking lets you fit more raw information into the same number of slots.
Q: Describe the Brown-Peterson paradigm and state what it demonstrates.
A: Participants hear three letters, then count backwards for a set interval to prevent rehearsal, then try to recall the letters. Recall drops sharply with longer intervals. This demonstrates that STM duration without rehearsal is roughly 10 seconds.
Q: What did Conrad (1964) find about the nature of errors in short-term recall?
A: Conrad found that recall errors were acoustic in nature, even when items were presented visually. Participants confused letters that sound similar (C with P) and words that rhyme (boat with vote), rather than items that look similar. This demonstrates that STM uses an acoustic/verbal code.
Q: Name and describe the three components of Baddeley's working memory model.
A: The phonological loop maintains verbal/acoustic information through rehearsal. The visuospatial sketchpad maintains visual and spatial information briefly. The central executive is the attentional control system that coordinates the other two and decides which type of maintenance to use.
Q: What did Newkirk, Klima and Bellugi (1979) demonstrate about the phonological loop?
A: They showed that deaf signers' STM errors were based on sign-form similarity (confusing signs that differ in only one component), while hearing participants made the expected acoustic errors. This demonstrates that the phonological loop handles linguistic information broadly, not just sound. Deaf signers encode linguistic material in terms of sign components, which is still linguistic in nature.
This material connects directly to long-term memory (covered in the next set of notes), because the encoding processes that move information from STM into LTM (levels of processing, elaboration) depend on how working memory handles the material in the first place.
Baddeley's central executive also links to the broader study of attention and executive function. If you go on to study cognitive control, dual-task paradigms, or individual differences in intelligence, you will see working memory capacity appear as a key predictor.
The finding that the phonological loop is linguistic rather than purely acoustic connects to psycholinguistics and the cognitive science of sign language.
Short-term memory, STM, working memory, WM, Miller's law, magical number seven, 7 plus or minus 2, chunking, rehearsal, maintenance rehearsal, phonological loop, articulatory loop, visuospatial sketchpad, central executive, Baddeley model, Baddeley and Hitch, Brown-Peterson paradigm, Brown-Peterson task, Conrad 1964, acoustic coding, verbal coding, word-length effect, Newkirk Klima Bellugi 1979, sign language memory, proactive interference, PI, Sperling iconic memory, cognitive psychology, memory capacity, memory duration