Phonological Loop

Abstract

The phonological loop is the speech-based subsystem of working memory, the store that briefly holds words or digits for immediate use. In Alan Baddeley and Graham Hitch's multicomponent model it has two parts: a phonological store that holds sound-coded traces for a second or two, and an articulatory rehearsal process that refreshes them by inner speech before they fade. This article sets out that architecture and the clean evidence for it — the phonological similarity effect, the word-length effect, articulatory suppression, and the irrelevant speech effect — then turns to the neuropsychology and imaging that localise the system, its proposed role in learning the sound patterns of new words, and the dispute over whether a dedicated rehearsal loop is needed. Three interactive demonstrations trace acoustic confusability, drive the two-second limit of the loop, and show which manipulation abolishes which effect.

Keywords: phonological loop, verbal working memory, articulatory rehearsal, word-length effect, vocabulary acquisition

The phonological loop is the component of working memory that stores and rehearses speech-based information, and it is the best understood part of the most influential model of the short-term mind (Baddeley, 2003). When a person repeats a telephone number under their breath until it can be dialled, or holds the beginning of a spoken sentence while its end arrives, the loop is what does the holding. It was proposed as one of two subsidiary stores in the model that Baddeley and Hitch put forward to replace the single short-term buffer of earlier theory, and in the decades since it has accumulated a body of converging evidence — behavioural, neuropsychological, and developmental — that is rare for a hypothetical cognitive mechanism (Baddeley & Hitch, 1974; Baddeley, 2012). The sections below set out what the loop is, the four laboratory effects that reveal its inner structure, the brain systems that implement it, its surprising function in language learning, and the theoretical challenges that keep it a live topic rather than a settled one.

What the Phonological Loop Is

The phonological loop is defined by the kind of information it holds rather than by the modality through which that information arrives. Its currency is the phonological code: a representation of speech sounds, of how a word is pronounced rather than how it looks or what it means. Spoken material enters the store automatically, but printed words and even nameable pictures can also be recoded into this sound-based form and held there, which is why the loop governs verbal memory span whether the list is heard or seen (Baddeley, 2003). It is one of the two subsidiary systems of the multicomponent model, the counterpart of the visuospatial sketchpad, and it is subordinate to the central executive, the attentional controller that has no storage of its own (Baddeley & Hitch, 1974).

What makes the loop tractable, where so much of cognitive theory is not, is that it makes sharply testable predictions. If storage is by sound, then similar-sounding items should be confusable; if rehearsal is a real-time articulatory process, then longer words that take longer to say should crowd out capacity; if that rehearsal can be blocked, its blocking should remove exactly the effects that depend on it. Each of these predictions has been confirmed, and the fact that they hold together — that the same small architecture explains four distinct effects and their interactions — is what lifted the loop from a convenient label to a mechanism (Baddeley, 1992). The model has also proved durable: it survived the major revision that added the episodic buffer in 2000, which binds phonological, visual, and long-term information into unified chunks but left the loop's account of pure verbal memory intact (Baddeley, 2000).

The Two Subcomponents

The loop is not a single device but a pairing of a store and a process, and the distinction is the heart of the model. The phonological store is a passive buffer that holds acoustic or speech-based traces in a form that decays within roughly two seconds unless it is refreshed. The articulatory rehearsal process is the refreshing mechanism: a form of inner speech that reads the contents of the store and feeds them back into it, resetting the decay clock, in the same way that a person keeps a number alive by muttering it. Rehearsal has a second duty — it is the gateway through which visually presented material is converted into a phonological code and entered into the store in the first place (Baddeley, Lewis, & Vallar, 1984). Table 1 sets out the division of labour and the evidence that isolates each part.

Table 1

The Two Components of the Phonological Loop

Note. The store and the rehearsal process are dissociated by their differing responses to the four experimental manipulations: a manipulation that targets the store leaves rehearsal-dependent effects intact, and vice versa (Baddeley, 2003).

The Signature Effects

The first and defining property of the store is that it codes by sound. Reuben Conrad demonstrated this before the loop was named, in the phonological similarity effect: lists of letters that sound alike (B, C, D, G, P, T, V) are harder to recall in the correct order than lists that sound distinct (F, K, Q, R, W, X, Y), and, crucially, this holds even when the letters are presented visually, so that the confusion cannot be in the eye (Conrad, 1964). Because errors take the form of one similar-sounding item substituting for another, the finding shows that the store represents items by their acoustic features, so that overlapping features breed confusion. The demonstration below maps that confusability directly, showing which letters crowd together in the sound space the store uses.

The second effect exposes the rehearsal process and its timing. The word-length effect is that immediate memory span is smaller for lists of long words than for lists of short ones: a person can repeat back more one-syllable words (wit, sum, harm) than five-syllable words (university, opportunity, refrigerator). Baddeley, Thomson, and Buchanan traced this to articulation time rather than syllable count as such, finding that memory span corresponds closely to the number of items a person can pronounce in about two seconds — as though the loop were a tape of fixed duration that can hold only as many words as fit within it when spoken (Baddeley, Thomson, & Buchanan, 1975). Span is therefore not a fixed number of slots but a fixed span of time, and it varies with how quickly the material can be rehearsed. The demonstration below drives that relationship, computing predicted span from articulation rate and the duration of the loop.

Suppression and Irrelevant Speech

The remaining two effects are experimental interventions that switch parts of the loop off, and their value is diagnostic: because each effect depends on a particular component, silencing that component removes exactly the effects that rely on it and spares the rest. Articulatory suppression — requiring the participant to repeat an irrelevant sound such as the, the, the aloud while trying to remember a list — occupies the rehearsal process and so blocks it. Suppression abolishes the word-length effect, because that effect is a product of rehearsal; and for visually presented lists it also abolishes the phonological similarity effect, because the visual items can no longer be recoded into sound and never reach the phonological store (Baddeley, Lewis, & Vallar, 1984). For spoken lists the similarity effect survives suppression, since heard words enter the store directly without needing rehearsal to recode them — a double dissociation that pins each effect to its component.

The irrelevant speech effect, first reported by Salame and Baddeley, is that unattended background speech impairs recall of a visually presented list even when the participant is told to ignore it and the speech is in an unknown language or is meaningless (Salame & Baddeley, 1982). The interference is with the store, not with rehearsal: irrelevant sound gains obligatory access to the phonological store and corrupts the traces already held there, whereas non-speech noise of comparable loudness does little. Because the effect targets the store and suppression targets rehearsal, the two manipulations combine informatively, and their full pattern across modalities is the loop's most compact fingerprint. The demonstration below lays out that pattern, letting the reader toggle presentation modality and each manipulation to see which effect remains.

Neural Basis and Neuropsychology

The strongest evidence that the phonological store is a distinct system, and not merely a way of talking about verbal memory, comes from patients in whom it is selectively damaged. Giuseppe Vallar and Baddeley studied patient PV, a woman with a dense, specific impairment of auditory-verbal short-term memory — her digit span was reduced to about two items — in the presence of normal long-term learning, normal speech, and normal intelligence (Vallar & Baddeley, 1984). PV showed no phonological similarity effect and no word-length effect for auditory material, precisely the profile expected if the phonological store itself were damaged, and her deficit dissociated cleanly from the visuospatial and executive components. Such patients demonstrate that the loop can fail on its own, which a single undifferentiated short-term memory could not do.

Neuroimaging then separated the store from the rehearsal process anatomically. In an early positron-emission study, Paulesu, Frith, and Frackowiak contrasted tasks that engaged the store with tasks that added rehearsal, and localised the phonological store to the left inferior parietal cortex (around the supramarginal gyrus) and the articulatory rehearsal process to the left inferior frontal region associated with speech production, Broca's area (Paulesu, Frith, & Frackowiak, 1993). The two-part psychological model thus mapped onto two separable brain regions. Later work complicated this tidy localisation: reviewing the evidence, Buchsbaum and D'Esposito argued that no single region is a dedicated phonological store, and that verbal short-term memory instead emerges from the same sensorimotor circuits of the posterior superior temporal and inferior parietal cortex that support speech perception and production (Buchsbaum & D'Esposito, 2008). The relationship between holding speech in mind and processing it in the first place has itself become a research question, with the boundary between phonological short-term memory and the speech system increasingly seen as porous (Jacquemot & Scott, 2006).

The Phonological Loop as a Language-Learning Device

If the loop merely held telephone numbers, it would be hard to see why evolution would have equipped the brain with it. Baddeley, Gathercole, and Papagno proposed a more fundamental function: the phonological loop is a device for learning language, specifically for acquiring the sound patterns of new words (Baddeley, Gathercole, & Papagno, 1998). Learning a new word means forming a durable long-term representation of an unfamiliar sequence of sounds, and doing that requires holding the novel sequence in a phonological form long enough for a long-term trace to be laid down — exactly what the loop provides. The prediction is that the capacity of phonological memory should predict the rate of vocabulary acquisition, and it does, in two converging lines of evidence.

In children, Gathercole and Baddeley followed a cohort longitudinally and found that performance on nonword repetition — the ability to hear and immediately repeat back an invented word such as blonterstaping — measured at age four predicted vocabulary size a year later, over and above nonverbal ability, implicating phonological memory as a cause rather than a correlate of vocabulary growth (Gathercole & Baddeley, 1989). In adults, Papagno, Valentine, and Baddeley showed the same logic at work in foreign-language learning: articulatory suppression, which disables the loop, selectively impaired the learning of foreign vocabulary while leaving the learning of paired words in the native language largely untouched, because native pairs can lean on existing semantic representations whereas a foreign word form has none (Papagno, Valentine, & Baddeley, 1991). The loop, on this account, is the bottleneck through which the sound of every new word must pass on its way into the mental lexicon.

Debates and Alternatives

The phonological loop is a successful model, but it is not the only account of verbal short-term memory, and its central mechanism — a dedicated rehearsal process running a decaying store — is contested. The most developed alternative is the time-based resource-sharing model of Pierre Barrouillet and Valerie Camos, which holds that verbal memory is maintained not by a special articulatory loop but by general attention, which must be switched rapidly between refreshing the memory traces and performing any concurrent processing (Barrouillet, Bernardin, & Camos, 2004). On this view forgetting reflects temporal decay during the moments when attention is diverted, and the key variable is the cognitive load of a task — the proportion of time it captures attention — rather than articulation. A partial reconciliation now distinguishes two maintenance mechanisms that can operate in parallel: articulatory rehearsal, which is domain-specific and verbal, and attentional refreshing, which is domain-general, so that the loop is one route to maintenance rather than the only one (Camos, 2017).

A further refinement, less a challenge than a correction, concerns what the store holds. A purely phonological buffer should treat words and nonwords alike, yet immediate span is reliably larger for real words than for matched nonwords — the lexicality effect. Hulme, Maughan, and Brown traced this to long-term memory: familiar words carry stable lexical representations that can be used to reconstruct a partly decayed trace at recall, a process called redintegration, whereas a nonword has no such support (Hulme, Maughan, & Brown, 1991). Schweickert formalised the idea in a multinomial processing-tree model in which each item is either recalled intact from the phonological trace or, if that trace has degraded, redintegrated with some probability from long-term knowledge, so that lexicality and similarity act at separable stages (Schweickert, 1993). Redintegration shows that even the classic loop is never purely phonological in practice: long-term knowledge leaks into the short-term store at the moment of retrieval, which is also why the loop can serve as a bridge into the lexicon during word learning.

A second challenge comes from language itself. Acheson and MacDonald argued that the serial ordering of verbal material in short-term memory draws on the same mechanisms that order words during speech production, so that what looks like a dedicated memory rehearsal loop is really the language production system pressed into temporary service (Acheson & MacDonald, 2009). This emergent-process view, developed further by Schwering and MacDonald, treats verbal working memory not as a separate storage buffer but as the sustained activation of the very networks that comprehend and produce language (Schwering & MacDonald, 2020). Against these reductions, Dennis Norris has defended the necessity of a distinct short-term store, arguing on computational grounds that models collapsing short-term memory into activated long-term memory cannot account for the ability to hold and reproduce a novel sequence, which requires representations dedicated to order rather than content (Norris, 2017). The multicomponent model's own proponents have meanwhile restated it in the light of these debates, tracing its descent from the earlier modal model and defending the loop as a component that earns its place (Baddeley, Hitch, & Allen, 2019).

Worked Example

The word-length effect is worth working through numerically, because the arithmetic is exactly what the articulation-loop demonstration computes and it makes the model's central claim concrete: memory span is a fixed duration of speech, not a fixed count of items. The governing relation from Baddeley, Thomson, and Buchanan is that span equals the number of items a person can articulate within the life of the loop, so that predicted span is the product of articulation rate and loop duration, S = R × d (Baddeley, Thomson, & Buchanan, 1975).

Take the loop duration d to be 1.8 seconds, within the one-and-a-half to two-second range the classic estimates give. Suppose a reader can pronounce short one-syllable words at a rate R of 3.0 words per second. Predicted span is S = 3.0 × 1.8 = 5.4 words. Now suppose the list is made of long words that can be pronounced at only 1.5 words per second. Predicted span falls to S = 1.5 × 1.8 = 2.7 words. The span has almost halved without a single slot being removed, purely because each word now consumes more of the fixed two-second window. The model makes a further, stronger prediction that the data bear out: the ratio of the two spans should equal the ratio of the two articulation rates, here 5.4 / 2.7 = 2.0, matching 3.0 / 1.5 = 2.0. Because span tracks speaking rate so closely, the same equation predicts the large differences in digit span across languages, since digits that are quicker to say in one language than another yield a proportionally longer span. Moving the demonstration's rate and duration controls recomputes S by this same relation.

Discussion

The phonological loop occupies an unusual place in cognitive psychology: it is a hypothetical mechanism specified tightly enough to have been tested, localised, and challenged on its own terms, and it has largely survived. Its strength is the convergence of independent evidence — four behavioural effects that interlock, patients whose selective deficits match the model's predictions, imaging that separates its two components, and a developmental function in vocabulary learning that gives it a reason to exist (Baddeley, 2003; Baddeley, Gathercole, & Papagno, 1998). Few constructs in the field are constrained by so many kinds of data at once, and that is why the two-part loop remains the textbook account decades after it was proposed (Baddeley, Hitch, & Allen, 2020).

Its limits are equally instructive. The clean localisation of a dedicated store has blurred, the necessity of a special rehearsal loop is disputed by attentional and language-based accounts, and the boundary between holding speech and processing it has proved harder to draw than the original model assumed (Buchsbaum & D'Esposito, 2008; Barrouillet, Bernardin, & Camos, 2004). What is not in dispute is the body of phenomena the loop was built to explain: the similarity, length, suppression, and irrelevant-speech effects are robust, and any successor theory must account for them just as the loop does. The model's likely fate is not refutation but absorption — a reframing in which its store and rehearsal process are understood as states of the language and attention systems rather than as separate boxes, while the predictions that made it valuable are preserved (Norris, 2017).

Current Directions

The most active current questions concern whether verbal short-term memory is a distinct faculty at all, and whether it can be improved. The emergent-process program continues to argue that verbal working memory is the sustained engagement of language comprehension and production rather than a separate buffer, and recent statements of that position marshal computational and neural evidence for treating maintenance as reactivation of linguistic knowledge (Schwering & MacDonald, 2020). The reconciliation of rehearsal with attentional refreshing has likewise sharpened into specific proposals about when each mechanism dominates, framed within the time-based resource-sharing account (Camos, 2017). Against the dissolution of the store into language, the computational case for a dedicated order-sensitive short-term system has been restated with new force (Norris, 2017).

A practically important strand concerns training. The popularity of commercial working-memory training rests on the hope that expanding phonological memory would carry over to vocabulary, reading, or reasoning, but the evidence is discouraging: Norris, Hall, and Gathercole found that intensive practice on verbal short-term memory tasks produced gains that were highly specific to the trained materials and did not transfer to untrained short-term memory or to related abilities (Norris, Hall, & Gathercole, 2019). The finding fits the model's logic — if span is set by the physical rate of articulation and the decay constant of the store, there is little a training regime can move — and it reframes the loop's capacity as a stable individual characteristic rather than a trainable skill, with consequences for education and clinical practice that are still being worked out.

Commonly Confused With

Working Memory

The phonological loop is a part; working memory is the whole. The loop is one of the two subsidiary stores in Baddeley and Hitch's model, dedicated to speech-based material, and it sits alongside the visuospatial sketchpad under the control of the central executive. Calling the loop working memory mistakes a single specialised buffer for the entire multicomponent system; the loop cannot manipulate information or coordinate tasks, which are the executive's jobs. The confusion is natural because verbal span tasks are the ones most often used to measure working memory, but the task taps the part, not the whole.

Common Misconceptions

The word-length effect happens because long words have more syllables to store.

The controlling variable is time to articulate, not the number of syllables or letters. Words matched for syllable count but differing in spoken duration produce the effect, and span corresponds to what can be said in about two seconds (Baddeley, Thomson, & Buchanan, 1975). The loop is a fixed span of time, so slow-to-say words crowd it regardless of how they are counted.

The phonological store only holds things that were heard.

Printed words and nameable pictures also reach the store, recoded into a sound-based form by the articulatory rehearsal process, which is why visually presented lists still show the phonological similarity effect (Conrad, 1964). That recoding route is exactly what articulatory suppression cuts off, removing the similarity effect for visual but not for spoken material (Baddeley, Lewis, & Vallar, 1984).

Training the phonological loop expands its capacity and boosts learning.

Intensive practice on verbal short-term memory tasks yields gains specific to the trained material that do not transfer to untrained memory or to broader abilities (Norris, Hall, & Gathercole, 2019). Because span is set by articulation rate and trace decay, loop capacity behaves as a stable trait rather than a muscle that training strengthens.

Glossary

Acoustic code.

A representation of material by its speech-sound properties rather than its visual form or meaning; the code in which the phonological store holds its contents.

Articulatory rehearsal.

The subvocal inner-speech process that refreshes decaying traces in the phonological store and recodes visual input into a phonological form.

Articulatory suppression.

Repeating an irrelevant sound aloud to occupy the rehearsal process, used experimentally to block rehearsal and recoding; it abolishes the word-length effect.

Central executive.

The attentional control component of working memory that coordinates the phonological loop and visuospatial sketchpad and allocates resources, holding no information itself.

Cognitive load.

In the time-based resource-sharing model, the proportion of time a concurrent task captures attention, and so the rate at which unattended memory traces decay.

Episodic buffer.

The component added to the model in 2000 that binds phonological, visual, and long-term information into integrated multidimensional chunks under executive control.

Irrelevant speech effect.

The impairment of verbal recall by unattended background speech, which gains obligatory access to the phonological store and corrupts the traces held there.

Nonword repetition.

Hearing and immediately repeating an invented word; a measure of phonological memory that predicts vocabulary acquisition in children.

Phonological loop.

The speech-based subsystem of working memory, comprising a phonological store and an articulatory rehearsal process, that holds verbal material over seconds.

Phonological similarity effect.

The poorer ordered recall of similar-sounding than dissimilar-sounding items, evidence that the store codes material by sound; also called the acoustic confusion effect.

Phonological store.

The passive buffer of the loop that holds sound-coded traces which decay within about two seconds unless refreshed by rehearsal.

Redintegration.

The reconstruction of a partly decayed short-term trace using long-term knowledge of words and sound patterns, which aids recall of familiar over unfamiliar material.

Time-based resource-sharing.

The model on which verbal maintenance depends on general attention switching between refreshing traces and processing, with forgetting driven by temporal decay under load.

Word-length effect.

The smaller memory span for long than for short words, arising because span equals what can be articulated within the fixed duration of the rehearsal loop.

Working memory.

The limited-capacity system that holds and manipulates information over seconds, of which the phonological loop is the speech-based storage component.

Key Researchers

Alan Baddeley (b. 1934). Emeritus Professor of Psychology at the University of York; with Graham Hitch he proposed the multicomponent model of working memory and, with colleagues, established the two-part structure and language-learning function of the phonological loop. Faculty Page - ORCID - Google Scholar - Wikipedia

Pierre Barrouillet (b. 1954). Professor of Developmental Cognitive Psychology at the University of Geneva; with Valerie Camos he developed the time-based resource-sharing model, the leading attentional alternative to the articulatory account of verbal maintenance. Faculty Page - ORCID - Google Scholar

Reuben Conrad (1916-2020). Researcher at the MRC Applied Psychology Unit, Cambridge; his discovery of acoustic confusions in immediate memory showed that the short-term store codes material by sound, the finding on which the phonological store rests. BPS Obituary

Susan E. Gathercole (b. 1958). Emeritus Professor at the University of Cambridge and former Director of the MRC Cognition and Brain Sciences Unit; she established the developmental link between phonological short-term memory and vocabulary acquisition. Faculty Page - ORCID - Google Scholar - Wikipedia

Graham J. Hitch. Emeritus Professor of Psychology at the University of York; with Alan Baddeley he proposed the 1974 multicomponent model of working memory within which the phonological loop is defined. Faculty Page - Google Scholar - Wikipedia

Dennis Norris. Programme Leader at the MRC Cognition and Brain Sciences Unit, Cambridge; he builds computational models of verbal short-term memory and has defended the necessity of a distinct short-term store against accounts that reduce it to activated long-term memory. Faculty Page - ORCID

Costanza Papagno. Professor of Neuropsychology at the University of Trento; she showed that phonological short-term memory predicts foreign-language vocabulary learning and studied the loop in neurological patients. Faculty Page - ORCID - Google Scholar

Giuseppe Vallar (b. 1951). Emeritus Professor of Psychology at the University of Milano-Bicocca; with Baddeley he described the patient evidence that fractionated working memory and isolated a selective phonological short-term store. Faculty Page - ORCID - Google Scholar

Frequently Asked Questions

What is the phonological loop?
The phonological loop is the speech-based subsystem of working memory, made up of a phonological store that holds sound-coded traces for about two seconds and an articulatory rehearsal process that refreshes them by inner speech (Baddeley, 2003).

What are the two parts of the phonological loop?
It has a passive phonological store, which holds acoustic traces that decay quickly, and an active articulatory rehearsal process, which refreshes those traces and recodes visually presented words into a sound-based form (Baddeley, Lewis, & Vallar, 1984).

What is the word-length effect?
Memory span is smaller for lists of long words than short ones because span corresponds to the number of items a person can articulate in about two seconds, so slower-to-say words fill the fixed loop sooner (Baddeley, Thomson, & Buchanan, 1975).

What is articulatory suppression?
Articulatory suppression is repeating an irrelevant sound aloud while trying to remember a list, which occupies the rehearsal process and abolishes the word-length effect and, for visual lists, the phonological similarity effect (Baddeley, Lewis, & Vallar, 1984).

How does the phonological loop help language learning?
The loop holds the unfamiliar sound sequence of a new word long enough for a long-term representation to form, so phonological memory predicts vocabulary growth in children and foreign-word learning in adults (Baddeley, Gathercole, & Papagno, 1998).

Where in the brain is the phonological loop?
Neuroimaging localised the phonological store to the left inferior parietal cortex and articulatory rehearsal to the left inferior frontal speech region, though later work argues the store is not a single dedicated area (Paulesu, Frith, & Frackowiak, 1993).

Is the phonological loop the same as working memory?
No; the loop is only the speech-based storage component of working memory, which also includes the visuospatial sketchpad, the central executive, and the episodic buffer (Baddeley, 2000).

Can the phonological loop be trained to hold more?
Practice on verbal short-term memory tasks improves the trained materials but does not transfer to untrained memory or to broader abilities, suggesting loop capacity is a stable trait set by articulation rate and trace decay (Norris, Hall, & Gathercole, 2019).

References