Recall

Abstract

Recall is the retrieval of stored information from memory without the item being present to prompt recognition, the effortful reconstruction that produces a name, a fact, or an event on demand. It is measured in free, cued, and serial forms, each exposing a different facet of how retrieval cues contact stored traces. Classic findings (the serial position curve, encoding specificity, and the availability-versus-accessibility distinction) show that recall failure is often a failure of access rather than of storage. The act of recall also changes memory: retrieval practice strengthens what is retrieved and can distort it. Interactive demonstrations trace cued recall, the serial position curve, and the testing effect.

Keywords: recall, retrieval, encoding specificity, serial position, testing effect

Recall is the process of bringing a stored representation back to mind in the absence of the original stimulus: producing a witness's account, a vocabulary word, or the items on a shopping list left at home. It is distinguished from recognition, which requires only that a presented item be judged as previously encountered. Where recognition supplies the target and asks whether it has been seen before, recall supplies only a cue, or nothing at all, and asks the memory system to regenerate the target itself. Because it demands active reconstruction rather than a match against something present, recall is the more searching test of memory, and its successes and failures have furnished much of what is known about how human retrieval works (Tulving & Pearlstone, 1966).

What Recall Is

Recall and recognition are the two canonical ways of measuring retrieval, and they differ in what the test provides. In a recognition test the target is present and the task is a decision about its prior occurrence; in a recall test the target is absent and must be produced. This difference makes recall the harder measure: an item can be recognised when it cannot be recalled, because recognition can succeed on a partial match that is insufficient to regenerate the whole. The dissociation is not merely one of difficulty. Retrieval depends on how information was encoded, and manipulations that deepen encoding, such as attending to a word's meaning rather than its surface form, improve later recall far more than shallow processing does (Craik & Lockhart, 1972). The dependence on encoding can be dramatic: rating words for their relevance to a survival scenario yields some of the highest recall rates on record, exceeding well-established deep-encoding tasks and suggesting that memory is tuned to information with fitness value (Nairne et al., 2007).

A classical account of this asymmetry is the generate-recognize theory, which casts recall as a two-stage process: the system first generates candidate items by searching memory and then subjects each to a recognition check before producing it, whereas a recognition test supplies the candidate and so requires only the second stage (Anderson & Bower, 1972). On this view recall is the harder measure because it carries the extra burden of generation, and an item that cannot be generated cannot be recalled even when it would have been recognised. The theory is a useful first approximation, but, as the next section shows, it cannot be the whole story.

The central lesson of the recall literature is that a memory can be present in storage yet unreachable at the moment of test. Tulving and Pearlstone drew this distinction sharply as one between availability, whether a trace exists in memory, and accessibility, whether it can be retrieved under the current conditions (Tulving & Pearlstone, 1966). Their participants who failed to recall a word freely often produced it at once when given its category name, proving the word had been available all along. Much of what looks like forgetting is therefore a retrieval failure, and the study of recall becomes largely the study of what makes a stored trace accessible.

Varieties of Recall

Recall is not a single task but a family of them, distinguished by the cue supplied and the order demanded. In free recall the person reproduces the items of a studied list in any order; in cued recall each target is paired with a prompt, such as its category or a paired associate, given at test; in serial recall the items must be reproduced in their original order. The three differ systematically in difficulty and in what they reveal, as Table 1 sets out. Free recall leaves the retrieval cue unspecified, so the person must generate cues internally, which is why providing category names converts a free-recall failure into a cued-recall success (Tulving & Pearlstone, 1966). Serial recall imposes the further burden of order, taxing the representation of sequence as well as of item.

The time over which recall is attempted also matters. When rehearsal is prevented, memory for a small set of items decays within seconds: Peterson and Peterson found that recall of a consonant trigram fell from near-perfect to below half in eighteen seconds of distractor-filled delay, evidence for rapid loss from a short-term store when retrieval is not refreshed (Peterson & Peterson, 1959). Recall over longer intervals draws instead on durable representations whose accessibility depends less on time than on the quality of the cue.

Encoding Specificity and Retrieval Cues

If accessibility is what recall lacks, then the retrieval cue is what supplies it, and the governing principle is encoding specificity. Tulving and Thomson stated it as follows: a cue is effective to the degree that the information it carries was encoded together with the target at study (Tulving & Thomson, 1973). A cue that was not part of the original encoding, however strongly it is associated with the target in general knowledge, may fail to reach it; a cue that was encoded with the target, however weak its independent association, can succeed. Retrieval, on this view, is not a property of the cue or of the trace alone but of the match between the conditions of encoding and the conditions of retrieval.

The decisive evidence for this relational view was a result the generate-recognize account cannot accommodate: the recognition failure of recallable words. Watkins and Tulving found that a word a person later recalled to its studied cue was frequently not recognised when presented alone, so recall succeeded exactly where recognition failed (Watkins & Tulving, 1975). A model that makes recognition a stage within recall cannot explain a word being recalled yet unrecognised; encoding specificity can, because the cue present at recall reinstated the encoded trace while the isolated recognition probe, encoded differently, did not contact it. Retrieval success is governed by the encoding-retrieval match rather than by the intrinsic strength of the item.

The principle extends to the background context in which learning occurs. Godden and Baddeley had divers learn word lists on land or underwater and tested them in the same or the other environment, finding that recall was reliably better when the test environment matched the study environment (Godden & Baddeley, 1975). The context was never part of the words' meaning, yet, having been encoded alongside them, it functioned as a retrieval cue whose presence or absence moved recall. Encoding specificity thereby unifies a wide range of findings, from category cues to environmental context, under a single relational rule: what matters is not the cue in isolation but its correspondence to the encoded trace.

The Serial Position Curve

When people freely recall a list, the probability of recalling an item depends strongly on where it appeared. Plotting recall against input position yields the serial position curve, a U-shape with two limbs: a primacy advantage for the first few items and a larger recency advantage for the last few, with a depressed middle (Murdock, 1962). The two limbs have different origins. Primacy is attributed to rehearsal: early items enter an uncrowded memory and receive more rehearsal, which transfers them more effectively to long-term storage. Recency is attributed to the continued availability of the last items in a short-term store at the moment recall begins.

The strongest evidence for two contributions comes from dissociating them. Glanzer and Cunitz showed that a filled delay between study and recall selectively abolishes the recency effect while leaving primacy intact, exactly as a short-term store that empties during the delay would predict, whereas variables such as presentation rate affect primacy without touching recency (Glanzer & Cunitz, 1966). The double dissociation, one manipulation removing recency and another moving primacy, was among the most influential early arguments that memory comprises distinct short-term and long-term components rather than a single store.

Retrieval Practice and Its Consequences

Recall is not a passive readout that leaves the trace unchanged; the act of retrieving an item modifies its later accessibility. The clearest case is the testing effect: retrieving information from memory produces better long-term retention than an equivalent period of restudying it. Roediger and Karpicke had students study prose and then either restudy it or take a recall test, and found that although restudy produced better performance on an immediate test, retrieval practice produced markedly better retention after a delay of a week (Roediger & Karpicke, 2006). The benefit is not a quirk of one paradigm: retrieval practice enhances retention across materials, ages, and test formats, an effect robust enough to carry direct educational recommendations (Roediger & Butler, 2011).

Why retrieval should help more than restudy is itself informative. Successful recall exercises the retrieval route to a trace, strengthening the cues and pathways that reach it, so that the same route is easier to travel later; a demanding but successful retrieval leaves a larger benefit than an easy one (Karpicke & Roediger, 2008). This is one case of a broader principle Bjork named desirable difficulties: conditions that slow acquisition and depress immediate performance — retrieval rather than restudy, spaced rather than massed practice, varied rather than constant conditions — frequently improve long-term retention, so the fluency displayed during study is a treacherous index of the learning that will actually endure (Soderstrom & Bjork, 2015). Retrieval also shapes neighbouring memories: recalling one item can suppress related competitors, so that the very act of remembering causes forgetting of what was not retrieved (Anderson et al., 1994). Retrieval, in short, is an act of learning in its own right, one that redistributes accessibility across the whole set of related traces rather than simply reporting the state of one.

Reconstruction and Error

Because recall regenerates rather than replays, it is open to systematic error. What is produced is assembled from the trace, the cue, and the person's general knowledge and expectations, and this reconstruction can introduce content that was never studied. The most controlled demonstration is the false recall of unpresented words: after studying a list of associates (bed, rest, tired, dream), people frequently recall the missing theme word (sleep) with confidence equal to that for words actually seen (Roediger & McDermott, 1995). The intrusion is not a lapse of attention but a predictable product of retrieval: the theme was activated at encoding and is regenerated at test as though it had occurred.

Reconstruction is the price of a memory system built for gist and inference rather than verbatim record. The same processes that let recall fill gaps sensibly, inferring what must have been there, also let it insert what was merely expected. This is why the confidence accompanying a recalled detail is a poor guide to its accuracy, and why the study of recall bears directly on applied questions such as the reliability of eyewitness testimony, where a reconstructed memory can be both wrong and firmly held.

Worked Example

The availability-versus-accessibility distinction can be made quantitative with a simple cued-recall model in the spirit of Tulving and Pearlstone (Tulving & Pearlstone, 1966). Suppose a person studies 48 words organised into 12 categories of 4 words each, and that, once a category is brought to mind, they recover on average 3 of its 4 words. The number recalled then depends entirely on how many category cues are available at test.

- Free recall. With no cues provided, the person must self-generate category labels and, say, retrieves 6 of the 12. Recall = 6 categories x 3 words = 18 of 48 words, or 37.5%. - Cued recall. With all 12 category names provided at test, every category is entered. Recall = 12 categories x 3 words = 36 of 48 words, or 75%. - The accessibility gap. The difference, 36 - 18 = 18 words, is the set that was available in storage the whole time but inaccessible under free recall for want of a cue. It doubles measured recall without any change to what is stored.

The model makes the central claim concrete: the words recovered by the cues were not newly learned at test but merely made reachable. It also predicts the boundary condition. If the person could already self-generate all 12 category cues, the two scores would coincide and cueing would add nothing, which is why the availability-accessibility gap is largest exactly when self-cued retrieval is weakest.

Discussion

Across a century of study, recall has proved to be less a matter of whether a memory survives than of whether it can be reached, and by what. The availability-accessibility distinction reframed forgetting as frequently a retrieval failure, and encoding specificity supplied the rule that governs access: retrieval succeeds to the degree that the conditions of test reinstate the conditions of encoding (Tulving & Thomson, 1973). The serial position curve then showed that even a single free-recall test decomposes into contributions from distinct stores, making recall a probe of memory's architecture as well as its contents (Glanzer & Cunitz, 1966). More recently the emphasis has shifted from recall as a measurement to recall as an intervention, an act that strengthens, suppresses, and sometimes distorts the very memories it queries (Roediger & Butler, 2011). What remains contested is the mechanism that links these facts: whether the benefits of retrieval, the costs of retrieval-induced forgetting, and the errors of reconstruction all flow from one process that reshapes accessibility, or from several that happen to share the retrieval stage.

Current Directions

Three lines of work now define the active front. The first concerns the cognitive mechanism of the testing effect: large meta-analyses confirm that quizzing reliably raises later retention across classroom settings and specify the conditions, feedback, spacing, and test format, under which the benefit is largest (Yang et al., 2021). The second examines how retrieval interacts with new learning and consolidation: retrieval practice does not only stabilise the tested memory but potentiates the encoding of information studied afterward, and may act as a fast route to the consolidation ordinarily attributed to sleep (Chan et al., 2018; Antony et al., 2017). The third is neurocognitive, mapping the systems that support retrieval-based learning and asking why an act of recall changes a memory trace more than restudy does, with reviews now drawing the behavioural testing effect together with its candidate hippocampal and cortical mechanisms (van den Broek et al., 2016; Karpicke, 2017). The through-line is a move from cataloguing when retrieval helps to explaining, mechanistically, how the act of recall rewrites what is stored.

Commonly Confused With

Recognition

Recognition presents the target and asks only whether it is old or new; recall withholds the target and requires the memory system to regenerate it. An item can be recognised but not recalled, because a partial match sufficient for a yes-or-no decision may be too weak to reconstruct the whole. The two are distinct retrieval measures, not degrees of the same one, which is why they can dissociate under the same encoding conditions.

Common Misconceptions

If something cannot be recalled, the memory is gone.

Recall failure is often a failure of access, not of storage: an item that cannot be recalled freely is frequently produced at once when a cue is supplied, proving it was available all along (Tulving & Pearlstone, 1966). Availability and accessibility are different things.

Testing just measures learning; it does not produce it.

Retrieving information is itself a learning event. A recall test produces better long-term retention than restudying for the same time, the testing effect, so retrieval changes memory rather than merely reading it out (Roediger & Karpicke, 2006).

A confidently recalled memory is an accurate one.

Recall reconstructs rather than replays, and reconstruction can insert content that was never studied. People falsely recall an unpresented theme word with confidence equal to that for studied words, so confidence is a poor guide to accuracy (Roediger & McDermott, 1995).

Glossary

Accessibility.

Whether a stored memory can be retrieved under the current conditions; contrasts with availability, whether the trace exists at all.

Availability.

Whether a memory trace is present in storage, independent of whether it can be retrieved on a given test.

Cued recall.

A recall test in which each target is prompted by an associated cue, such as its category name or a paired word, given at test.

Desirable difficulties.

Bjork's term for learning conditions that slow acquisition and depress immediate performance yet improve long-term retention, such as retrieval practice, spacing, and varied practice.

Encoding specificity.

The principle that a retrieval cue is effective to the degree that it was encoded together with the target at study.

Free recall.

A recall test in which studied items may be reproduced in any order, leaving the person to generate retrieval cues internally.

Generate-recognize theory.

A two-stage account of recall in which the system first generates candidate items and then applies a recognition check; challenged by the recognition failure of recallable words.

Primacy effect.

The superior recall of items early in a list, attributed to their receiving more rehearsal and firmer long-term storage.

Recency effect.

The superior recall of items at the end of a list, attributed to their remaining in a short-term store when recall begins; abolished by a filled delay.

Recognition.

A retrieval test in which the target is presented and judged as previously encountered or not, requiring a match rather than regeneration.

Retrieval cue.

Any information present at test that helps regenerate a target, effective in proportion to its overlap with the encoded trace.

Retrieval practice.

Studying by actively retrieving information from memory rather than restudying it; the manipulation that produces the testing effect.

Retrieval-induced forgetting.

The impaired later recall of items related to a retrieved target, caused by the suppression of competitors during retrieval.

Serial position curve.

The U-shaped function relating recall probability to an item's list position, combining primacy and recency.

Serial recall.

A recall test in which items must be reproduced in their original order, taxing memory for sequence as well as for item.

Testing effect.

The finding that retrieving information from memory produces better long-term retention than restudying it for the same time.

Key Researchers

Alan D. Baddeley (b. 1934). Emeritus Professor of Psychology at the University of York; with Godden he demonstrated context-dependent recall in the underwater diving experiment, showing that retrieval is best when the test context reinstates the encoding context. ORCID - Faculty Page - Wikipedia - Wikidata

Robert A. Bjork (b. 1939). Distinguished Research Professor at the University of California, Los Angeles; he developed the desirable-difficulties framework and, with the Andersons, the retrieval-induced-forgetting account of how the act of recall reshapes the accessibility of related memories. ORCID - Faculty Page - Wikipedia - Wikidata

Fergus I. M. Craik (b. 1935). Senior Scientist at the Rotman Research Institute, Baycrest, and University Professor Emeritus at the University of Toronto; with Lockhart he framed the levels-of-processing account, showing that deeper encoding yields more durable recall. ORCID - Faculty Page - Wikipedia - Wikidata

Bennet B. Murdock (1925-2022). Cognitive psychologist at the University of Toronto; he charted the serial position curve of free recall, the U-shaped primacy-and-recency signature that became a foundational datum for models of memory. Faculty Page - Wikipedia - Wikidata

Henry L. Roediger (b. 1947). James S. McDonnell Distinguished University Professor at Washington University in St. Louis; he established the modern testing-effect literature and, with McDermott, the false-recall paradigm that made memory reconstruction experimentally tractable. ORCID - Faculty Page - Wikipedia - Wikidata

Endel Tulving (1927-2023). Cognitive psychologist at the University of Toronto; he formulated the encoding-specificity principle and the availability-versus-accessibility distinction that anchor the modern theory of retrieval. Faculty Page - Wikipedia - Wikidata

Frequently Asked Questions

What is the difference between recall and recognition?
Recall requires the memory system to regenerate an absent target from a cue or from nothing, whereas recognition presents the target and requires only a judgment that it was previously encountered; recall is the more demanding measure, and an item can be recognised without being recalled (Tulving & Pearlstone, 1966).

What are the main types of recall?
The principal forms are free recall, in which items may be produced in any order; cued recall, in which each target is prompted by an associated cue at test; and serial recall, in which items must be reproduced in their original order (Tulving & Pearlstone, 1966).

What is encoding specificity?
Encoding specificity is the principle that a retrieval cue aids recall to the extent that it was encoded together with the target at study, so retrieval success depends on the match between the conditions of encoding and the conditions of test (Tulving & Thomson, 1973).

Why do we remember the first and last items of a list best?
Free recall yields a serial position curve because early items receive more rehearsal and firmer long-term storage (primacy) while the final items remain in a short-term store when recall begins (recency); a filled delay abolishes recency but spares primacy (Murdock, 1962; Glanzer & Cunitz, 1966).

Does self-testing actually improve memory?
Yes. Retrieving information from memory produces better long-term retention than restudying it for an equal time, an effect known as the testing effect, so self-testing is a learning activity and not merely a measurement (Roediger & Karpicke, 2006).

Can the act of recalling something change the memory of it?
It can. Successful retrieval strengthens the tested memory, but it can also suppress related competitors, causing retrieval-induced forgetting of items that were not recalled (Anderson et al., 1994).

Why is recall sometimes inaccurate?
Because recall reconstructs a memory rather than replaying it, drawing on the trace together with expectations and general knowledge; this can insert content that was never studied, as when people confidently recall an unpresented theme word (Roediger & McDermott, 1995).

If I cannot recall something, is it lost forever?
Not necessarily. Much recall failure reflects inaccessibility rather than loss: an item that cannot be recalled freely is often produced immediately once an effective cue is provided, showing it remained available in storage (Tulving & Pearlstone, 1966).

References