17.6 Models of Attention & Working Memory Architecture
Key Takeaways
Selective attention filter models range from early physical filtering (Broadbent) to attenuation (Treisman) and late semantic selection (Deutsch-Norman); Treisman's attenuation model best accounts for the cocktail party effect, where personally meaningful stimuli (e.g., hearing one's name) overcome attenuation thresholds to reach conscious awareness.
Baddeley's multicomponent working memory model replaces unitary short-term memory with a central executive that coordinates slave storage systems—the phonological loop (phonological store and articulatory rehearsal), visuospatial sketchpad, and episodic buffer—preventing cross-modal interference.
The Atkinson-Shiffrin model routes information through three stores of increasing durability: sensory memory holds a large but near-instantaneous trace, short-term memory holds roughly seven items for tens of seconds unless maintained by rehearsal, and chunking raises effective capacity by recoding items into larger meaningful units before transfer to long-term storage.
17.6 Models of Attention & Working Memory Architecture
Note
The Information Processing Bottleneck: The human brain receives millions of bits of sensory information every second, yet conscious working memory can actively process only a minute fraction of that stream. Cognitive architecture relies on selective attention filters to prioritize inputs, active multi-store workspaces to manipulate data, and synaptic plasticity to consolidate durable memory traces.
Mastering the cognitive and biological foundations of memory requires understanding both theoretical psychological models and the molecular synaptic transformations occurring within hippocampal circuits.
Models of Attention: Selective versus Divided Processing
Attention is the cognitive mechanism by which an organism allocates limited processing resources toward specific environmental inputs, internal thoughts, or behavioral goals.
Selective Attention & The Cocktail Party Effect
Selective attention involves focusing on one specific stimulus while filtering out or ignoring competing ambient inputs. The classic illustration is the cocktail party effect: an individual immersed in a boisterous room can selectively tune in to a single conversational partner, yet immediately notice if someone across the room mentions their name. This capacity demonstrates that unattended sensory channels are not completely discarded; they are processed at preconscious levels.
The Three Theoretical Filter Models
To explain how and where sensory filtering occurs, cognitive psychologists developed three competing models:
- Broadbent's Early Selection Filter Model:
- Stimuli from all sensory channels enter an unlimited sensory buffer based on raw physical characteristics (e.g., sensory modality, voice pitch, ear of presentation, spatial location).
- A selective all-or-nothing bottleneck filter immediately intercepts the stream, permitting only the attended physical input to pass through to higher-level perceptual and semantic processing systems.
- Unattended signals decay rapidly and are lost completely.
- Limitation: Fails to explain the cocktail party effect or why conditioned fear responses (or galvanic skin responses) occur when emotionally conditioned words are presented to the unattended ear in dichotic listening experiments.
- Treisman's Attenuation Model:
- Rather than an all-or-none filter, Treisman proposed an attenuator ("volume dial").
- Unattended inputs are not completely eliminated; their signal strength is merely turned down (attenuated).
- Incoming signals then pass to a dictionary unit containing words with pre-assigned activation thresholds. High-priority, personally salient words (such as one's own name, words denoting biological threat like "fire!", or contextually primed concepts) possess exceptionally low activation thresholds. Even an attenuated, faint signal possesses sufficient energy to trigger threshold activation, penetrating conscious awareness.
- Deutsch-Norman Late Selection Model:
- Proposes that the bottleneck occurs after full semantic analysis.
- All incoming inputs—attended and unattended—undergo full perceptual and semantic interpretation.
- The selective filter operates late, immediately prior to conscious awareness and working memory storage, deciding which semantically decoded inputs enter executive awareness.
Sensory Inputs ──> [Sensory Buffer]
│
Broadbent: [All-or-Nothing Filter] ──> [Semantic Analysis] ──> [Consciousness]
│
Treisman: [Attenuator (Volume)] ──> [Dictionary Unit] ──> [Consciousness]
│
Deutsch-Norman: [Semantic Analysis] ──> [Late Selection] ──> [Consciousness]
Divided Attention: Controlled vs. Automatic Processing
Divided attention is the ability to perform multiple distinct tasks concurrently. Resource allocation models dictate that human attention is a finite capacity pool:
- Controlled Processing: Required for novel, complex, or non-routine tasks. It demands conscious, active executive attention, operates slowly and sequentially, and is easily disrupted by competing tasks.
- Automatic Processing: Developed through extensive practice and overlearning. Familiar tasks (e.g., an experienced driver braking, reading printed words) proceed effortlessly, rapidly, and in parallel with minimal conscious demand on central attentional capacity.
Memory Architecture: The Atkinson-Shiffrin Multi-Store Paradigm
The Atkinson-Shiffrin model conceptualizes memory as flow through three structural storage systems:
- Sensory Memory: High capacity, ultra-brief storage storing an exact physical replica of the proximal stimulus.
- Iconic Memory: Visual sensory store lasting approximately 0.5 seconds. George Sperling demonstrated its true capacity using the partial report technique: when shown a 3x3 matrix of letters for 50 milliseconds, subjects in whole-report could name only 4–5 letters. However, if signaled by a high, medium, or low pitch tone immediately after matrix offset directing them to report a single row, subjects achieved near 100% accuracy, proving that the entire visual array is momentarily retained in iconic memory before rapid decay.
- Echoic Memory: Auditory sensory store lasting significantly longer, approximately 3 to 4 seconds, enabling comprehension of continuous human speech.
- Short-Term Memory (STM): A limited-capacity store retaining conscious information temporarily.
- Capacity: Traditionally estimated by George Miller as chunks (discrete units of meaningful information). Modern cognitive paradigms that prevent rehearsal estimate the true un-chunked working capacity closer to items.
- Duration: Approximately 15 to 30 seconds without active rehearsal, as demonstrated by the Brown-Peterson distractor task.
- Long-Term Memory (LTM): An essentially limitless storage repository holding knowledge, skills, and past experiences indefinitely.
Baddeley's Multicomponent Working Memory Model
Alan Baddeley and Graham Hitch rejected the concept of short-term memory as a static, unitary passive buffer, reconceptualizing it as an active, dynamic Working Memory (WM) workspace that actively manipulates, transforms, and coordinates information.
┌────────────────────────────────┐
│ Central Executive │
│ (Prefrontal Cortex Regulation) │
└───────┬─────────┬──────────────┘
┌──────────────────┘ └──────────────────┐
▼ ▼
┌────────────────────────┐ ┌────────────────────────┐
│ Phonological Loop │ │ Visuospatial Sketchpad │
│ ┌────────────────────┐ │ │ ┌────────────────────┐ │
│ │ Phonological Store │ │ │ │ Visual Cache │ │
│ │ ("Inner Ear") │ │ │ │ (Form/Color) │ │
│ └─────────┬──────────┘ │ │ └─────────┬──────────┘ │
│ ▼ │ │ ▼ │
│ ┌────────────────────┐ │ │ ┌────────────────────┐ │
│ │Articulatory Control│ │ │ │Inner Scribe (Motion│ │
│ │ ("Inner Voice") │ │ │ │ & Spatial) │ │
│ └────────────────────┘ │ │ └────────────────────┘ │
└───────────┬────────────┘ └───────────┬────────────┘
│ │
│ ┌────────────────────────┐ │
└──────>│ Episodic Buffer │<──────────┘
│(Multimodal Binding/LTM)│
└───────────┬────────────┘
▼
┌────────────────────────┐
│ Long-Term Memory │
└────────────────────────┘
The Four Components of Baddeley's Model
- Central Executive: The supervisory control system located within the dorsolateral prefrontal cortex. It does not store information; instead, it coordinates attentional allocation, toggles between tasks, inhibits irrelevant distractors, and regulates the subordinate slave systems.
- Phonological Loop: Handles acoustic and verbal information. Partitioned into:
- Phonological Store ("Inner Ear"): Passively holds speech-based sounds for 1 to 2 seconds before acoustic decay.
- Articulatory Rehearsal Component ("Inner Voice"): Actively recycles verbal representations through silent, subvocal speech. Explains the word-length effect (recall is lower for lists of long, multisyllabic words because they take longer to subvocally articulate within the decay window) and the phonological similarity effect (words that sound similar like cat, bat, mat are harder to recall than phonologically distinct words).
- Visuospatial Sketchpad ("Inner Eye"): Stores and manipulates visual imagery and spatial coordinates (e.g., mentally rotating an object, navigating a campus). Divided into the visual cache (form and color) and the inner scribe (spatial pathways and movement planning).
- Episodic Buffer: Added by Baddeley in 2000 as a temporary multimodal workspace with limited capacity. It binds phonological, visual, and spatial information into coherent, chronologically ordered episodic chunks, serving as the critical bidirectional gateway between working memory and long-term memory.
The Three Models of Selective Attention
The models differ on one question only: how far unattended input is processed before it is filtered out.
| Model | Filter location | Fate of unattended input | Key supporting evidence |
|---|---|---|---|
| Broadbent early filter | Immediately after sensory registration | Blocked entirely on physical characteristics before meaning is extracted | Participants report pitch and voice but no content of the unattended channel |
| Treisman attenuation | Early, but the filter turns input down rather than off | Weakened, yet salient items still break through | Hearing one's own name in the unattended ear (the cocktail party effect) |
| Deutsch-Norman late selection | After semantic analysis, at the response stage | Fully processed for meaning, then discarded | Semantic priming from unattended words |
In a classic dichotic listening experiment, a participant shadows a recorded prose passage presented to their right ear while an unrelated spoken passage is presented to their left ear. In the middle of the trial, the unattended left ear suddenly presents the participant's full name spoken in a normal conversational volume. The participant immediately halts shadowing and reports hearing their name. Which theoretical model of attention most accurately accounts for this finding?
Broadbent's Early Selection Filter Model, because all physical inputs are fully analyzed semantically prior to sensory buffering.
The Resource Allocation Model, because automatic processing enables infinite capacity without cognitive constraints.
Treisman's Attenuation Model, because personally salient stimuli possess low activation thresholds that are triggered even by attenuated signals.
The Deutsch-Norman Model, because sensory stimuli are discarded at the peripheral receptor level before reaching the central nervous system.
Sections you finish are checked off in the contents.