Serial Learning

Abstract

Serial learning is a type of verbal learning: the acquisition of a set of items in a fixed order, so that each is produced in its correct place, and with paired-associate learning one of the two paradigms on which the experimental study of memory was built. Unlike free recall, it requires reproducing not only the items but their arrangement, making order itself the object of study. Three findings organize the field: errors fall in a bowed serial-position curve, lowest at the start and peaking just past the middle; remote associations bind non-adjacent items, ruling out a simple chain; and order errors are overwhelmingly transpositions to nearby positions. Positional-coding and context-retrieval models reproduce these patterns by representing each item's place in the sequence rather than links to its neighbours.

Keywords: serial learning, serial-position curve, remote association, chaining, positional coding

What Serial Learning Is

Serial learning is the learning of a sequence of items in a fixed order, such that the learner can reproduce the whole series in that order. It is a form of verbal learning, the parent tradition whose other central paradigm is paired-associate learning; where paired-associate learning asks the participant to attach a response to a cue, serial learning asks the participant to attach each item to its place in an ordered chain. The material is typically a list of nonsense syllables, digits, or words, and the measure of learning is how error falls, or how many items are produced correctly, as the list is studied again and again.

What makes the paradigm distinctive is that order is not incidental to the task; it is the task. In free recall the participant may report a studied list in any sequence, and the experimenter scores only which items were produced. In serial learning the same item counts as an error if it appears out of place. This forces two kinds of information apart: item information, that a syllable was on the list at all, and order information, that it occupied the fourth position rather than the sixth. Much of what the paradigm has revealed concerns how order information is coded, and why it is more fragile than item information.

The task is also the laboratory ancestor of a large class of everyday skills. Reciting the alphabet, dialing a number, spelling a word, playing a memorized phrase of music, and producing the words of a sentence in grammatical order are all ordered sequences, and the mechanisms that serial-learning experiments probe, the coding of position and the planning of sequence, are the mechanisms those skills depend on. Table 1 sets out the principal accounts of how order is represented and the evidence that separates them.

Table 1

Accounts of Serial Order

Note. The main theoretical positions on how a sequence is represented, from item-to-item chains to positional and context-based codes. Each is developed in a later section; full citations appear in the References (Ebbinghaus, 1885/2013; Lashley, 1951; Henson, 1998; Logan, 2021).

The Serial Anticipation Method

The classical procedure is the serial anticipation method, run historically on a memory drum, a rotating cylinder that exposes one list item at a time through a window at a fixed rate. The participant's task is to anticipate: as each item appears, they must say the next item before the drum rotates to reveal it. On the first pass nothing can be anticipated, and performance climbs over successive passes as each item comes to cue the one that follows. Learning is indexed by the number of trials, or the time, needed to recite the whole list once without error, a criterion measure, or by the proportion of correct anticipations on each trial, a continuous one.

Ebbinghaus, working on himself a decade before the memory drum, used a related criterion. He read a list aloud at a metronome's pace until he could recite it once perfectly, recording the number of readings required. His savings method then measured retention by relearning: the reduction in readings needed to reach the same criterion a second time, after a delay, expressed how much of the ordered list survived even when unaided recall had failed (Ebbinghaus, 1885/2013). The anticipation method made the same logic trial-by-trial and position-by-position, so that learning could be watched as it accumulated unevenly across the list.

The Bowed Serial-Position Curve

When errors are plotted against an item's ordinal position, serial learning produces a characteristic bowed curve. Errors are fewest at the beginning of the list, a strong primacy region; they rise to a maximum a little past the middle; and they fall again toward the end, a weaker recency region. The position of maximum difficulty is reliably just beyond the midpoint rather than at it, and the primacy advantage is larger than the recency advantage, so the curve is not symmetric. The same positional signature appears whether the measure is errors, trials to master each item, or latency.

The bowed curve of serial learning is a near relative of, but not identical to, the U-shaped serial-position curve of free recall. Murdock's canonical free-recall functions show a pronounced recency arm, the last few items recalled best of all, because at the moment of an immediate test those items are still held in a limited short-term store (Murdock, 1962). Glanzer and Cunitz demonstrated the two-store logic directly: a filled delay between the last item and recall abolishes the recency arm while leaving primacy intact, dissociating a short-term from a long-term contribution (Glanzer & Cunitz, 1966). In serial anticipation the recency advantage is more modest, because the task requires producing each item in place rather than dumping the most accessible items first; what the two curves share is the robust primacy region and the mid-list peak in difficulty.

Position is not the only thing that governs how readily a list item is learned. Von Restorff's isolation effect shows that an item made distinctive from its neighbours, printed in a different colour, or drawn from a different category, is learned and retained far better than a homogeneous item in the same position (von Restorff, 1933). Distinctiveness raises an item above the bowed curve independently of where it sits, evidence that the curve reflects the relational structure of the list, not a fixed property of each slot.

What Is the Functional Stimulus?

If learning a serial list means forming associations, what exactly becomes associated with what? The oldest answer is the chaining hypothesis: each item is the stimulus for the next, so the list is learned as a chain of item-to-item links, A cues B, B cues C, and so on. Chaining is intuitive and fits the anticipation method, in which the visible item does seem to prompt the next response. But it makes a strong prediction, that the only effective cue for any response is the item immediately before it, and that prediction fails.

Ebbinghaus had already found the decisive evidence. After learning a list, he constructed derived lists that preserved the original items but changed their spacing, pairing every second item, or every third, and skipping those between. Relearning a derived list showed savings relative to a wholly new list, and the savings declined smoothly as the derivation skipped more items. Associations, in other words, had formed not only between adjacent items but between items two, three, and more steps apart; these remote associations grow weaker with distance but are unmistakably present. A simple chain of adjacent links cannot produce them.

The alternative is that the effective, or functional, stimulus for each response is not the preceding item but the item's ordinal position in the list, its place counted from the ends. On this view the participant learns, in effect, “syllable JIR goes in position four,” and position is the cue that retrieves it. Encoding-specificity logic supports treating position as a genuine cue: a retrieval probe aids recall to the extent that its relation to the target was encoded at study, and in a fixed list an item's position is encoded on every trial (Tulving & Thomson, 1973). The modern consensus is that serial order rests on position-like and context-like codes rather than on a chain of item-to-item bonds, with remote associations and the pattern of order errors as the primary evidence.

Chaining and the Problem of Serial Order

Lashley gave the critique of chaining its general form in a 1951 essay that reached well beyond the memory drum. Ordered behaviour, he argued, includes far more than memorized lists: speech, typing, handwriting, and skilled music are all sequences produced too fast, and with the wrong kinds of error, for each act to be triggered by the sensory feedback of the one before it (Lashley, 1951). The latencies between successive keystrokes of a fluent typist are shorter than a stimulus-response loop would allow, and the errors are telling. Anticipatory errors, in which an element due later intrudes early, show that later items are already active while earlier ones are being produced; a strict chain, in which an item becomes available only once its predecessor has occurred, cannot generate them.

Lashley concluded that the serial order of behaviour must be imposed by a central plan that holds the whole sequence in a state of graded readiness and selects items in turn, rather than by a peripheral chain of reflexes. This reframing, from associative chain to planned sequence, set the agenda for the positional and hierarchical models that followed, and connects serial learning to the wider study of sequencing in language and action.

Positional Coding and Order Errors

If position rather than the preceding item is the cue, the signature should be visible in the errors. It is. When serial recall breaks down, the dominant error is not an item intrusion from outside the list but a transposition: a list item recalled in the wrong position. Transpositions are not random; they cluster around the item's correct place, so that an item is far more likely to be recalled one position early or late than five positions away. Plotted against displacement, order errors form a sharp locality gradient.

Henson's Start-End Model accounts for this gradient by coding each item's position as its distance from two anchors, the start and the end of the sequence (Henson, 1998). Early items are well specified by a strong start signal, late items by a strong end signal, and middle items, far from both anchors, are the most confusable, which reproduces the bowed difficulty curve. Because neighbouring positions have similar start-and-end codes, a noisy positional signal retrieves an adjacent item, which is exactly the locality gradient of transpositions. The model also predicts the detailed shape of these confusions, including the tendency for an omitted item to be filled in by the next item in the list and for a displaced item to return at its correct later position. The same positional architecture extends beyond verbal lists: using the time taken to make each response as a window on the underlying code, Hurlstone and Hitch showed that the serial order of a purely visual-spatial sequence is represented positionally in the same way, with transposition latencies tracking distance from the start and end markers (Hurlstone & Hitch, 2018).

A second positional account, Page and Norris's Primacy Model, dispenses with explicit start and end markers and stores order instead as a primacy gradient: a pattern of activation strongest for the first item and declining across the list, with items recalled by repeatedly selecting the most active and then suppressing it once produced. The gradient alone reproduces the primacy-dominated serial-position curve and captures the effects of list length and phonological similarity, while noisy competition among the similar activations of neighbouring items yields the same locality gradient of transpositions (Page & Norris, 1998). It and the Start-End Model are the two canonical positional models, differing in whether order is anchored to the list's ends or carried by a single decaying gradient.

Positional coding does not abolish associations between items; it reframes them. Items may still be linked, but the backbone that carries order is a representation of position, or of a slowly changing temporal context, against which items are stored and from which they are retrieved. That shift is what makes remote associations, the bowed curve, and the transposition gradient fall out of a single mechanism rather than three separate ones.

Interference and Transfer

Serial learning is also a workhorse for the study of interference, the main cause of forgetting established by the verbal-learning tradition. Because a list can be scored item by item, the effect of learning one list on memory for another can be measured precisely. Melton and Irwin showed that retroactive interference, the disruption of an earlier list by a later one, grows with the amount of interpolated learning, and that part of the loss reflects an active unlearning of the original list rather than simple competition at recall (Melton & Irwin, 1940).

Underwood then overturned the assumption that laboratory forgetting is mostly retroactive. Re-examining the steep forgetting typically seen over a day, he showed that most of it is proactive: it is caused by the many lists a practised participant had already learned before the target list, not by the time that passed afterward (Underwood, 1957). The transfer paradigm, in which the same items are re-paired or re-ordered across successive lists, became the tool for dissecting these effects, with transfer ranging from strong positive, when structure is preserved, to strong negative, when old order conflicts with new (Postman, 1962). Postman and Underwood later drew these results into a general statement of interference theory and its unresolved problems (Postman & Underwood, 1973).

Meaningfulness and the Rate of Learning

How fast a serial list is mastered depends heavily on the material. Nonsense syllables were introduced precisely to strip away pre-existing meaning, but they are not equally meaningless, and the residual meaning matters. Noble quantified this with meaningfulness (m), the average number of associations an item elicits in a fixed interval, and showed that it predicts the rate of learning across the range from low-association syllables to familiar words: high-m items are learned in fewer trials and occupy the list more securely (Noble, 1952). Meaningfulness interacts with position, flattening the bowed curve for easy material and exaggerating it for hard, so that the serial-position effect is in part a story about how limited coding capacity is spread across a list of competing items.

Worked Example

Consider the savings logic that underlies serial-learning retention measures. A participant learns a list of nonsense syllables by serial anticipation to a criterion of one perfect recitation, which takes 16 trials. Twenty-four hours later, unaided recall of the list is poor, but the participant relearns the same list, in the same order, to the same criterion, and this time reaches it in 10 trials.

The savings score expresses retention as the proportion of original effort spared on relearning:

savings = (original trials − relearning trials) / original trials

Substituting,

savings = (16 − 10) / 16 = 6 / 16 = 0.375 = 37.5%

Even though the participant could recall little of the list unaided after a day, relearning was 37.5% faster than original learning, so better than a third of the ordered list survived in a form that recall alone could not detect. This is the sense in which Ebbinghaus's savings method is more sensitive than a recall test: it registers residual order information below the threshold of explicit reproduction. Had relearning instead taken the full 16 trials, savings would be (16 − 16)/16 = 0%, complete forgetting; had it taken 4 trials, savings would be (16 − 4)/16 = 75%, substantial retention.

Current Directions

The study of serial order is now dominated by computational models that try to reproduce the full set of benchmark results at once. Oberauer and a large consortium set out those benchmarks explicitly, cataloguing the findings, among them the bowed curve, the transposition gradient, and the fill-in and in-fill patterns of order error, that any adequate model of short-term and working memory must reproduce, and so giving the field a shared target rather than a scatter of isolated effects (Oberauer et al., 2018).

Against that target, Logan proposed a unifying account, the context retrieval and updating (CRU) model, in which a single mechanism, encoding a context representation built from the items themselves and updating it as each item is produced, generates serial order across perception, cognition, and action, from whole report to typing to serial recall (Logan, 2021). The proposal drew a sharp reply. Osth and Hurlstone argued that a context made only of the previous items, an item-dependent context, cannot capture several benchmark findings, and that an item-independent positional or temporal signal of the kind Henson's model uses is still required (Osth & Hurlstone, 2023). The exchange marks the live question in the field: whether the order of a sequence is carried by the items that compose it, by an abstract code of position, or by both, and the serial-learning phenomena assembled over a century, remote associations, the bowed curve, and the locality gradient of transpositions, remain the evidence these models are judged against.

Discussion

Serial learning began as the most austere paradigm in psychology, a lone observer reciting nonsense syllables to a metronome, and it has proved to be about something general: how a nervous system imposes order on a set of items and later reproduces it. The century of work converges on a few durable conclusions. Order is coded separately from item identity and is the more fragile of the two. The code is not a chain of adjacent links but something closer to position, or to a gradually changing context, against which items are placed; remote associations and the locality gradient of transpositions are its fingerprints. And the difficulty of a list is distributed in a bowed curve that any serious model must explain.

What has changed is the level of description. The early tradition measured order with trials-to-criterion and savings and theorized in words; the current tradition measures the detailed distribution of errors and theorizes with equations that must reproduce an agreed list of benchmarks. The questions, though, are recognizably Ebbinghaus's and Lashley's: what binds an item to its place, and how is a whole sequence held ready before any of it is produced.

Common Misconceptions

A serial list is learned as a chain, each item cueing the next.

Remote associations, savings on derived lists that skip items, and anticipatory errors in skilled sequences all show that associations extend beyond adjacent items and that later items are active before earlier ones finish (Ebbinghaus, 1885/2013; Lashley, 1951). A simple adjacent chain cannot produce these, which is why positional and context codes replaced it.

The serial-position curve is the same as the U-shaped free-recall curve.

They are relatives, not twins. Serial anticipation yields a bowed curve with strong primacy, a mid-list peak in difficulty just past centre, and only modest recency; immediate free recall yields a pronounced recency arm that a filled delay erases (Murdock, 1962; Glanzer & Cunitz, 1966). The shared feature is primacy, not the shape as a whole.

Order errors are random when memory fails.

They are highly structured. The dominant error is a transposition of a list item to a nearby position, forming a sharp locality gradient; items rarely jump far, and intrusions from outside the list are comparatively rare (Henson, 1998). This structure is the main evidence for positional coding.

Nonsense syllables are all equally meaningless, so material does not matter.

Syllables vary in meaningfulness, and higher-meaning items are learned in fewer trials and held more securely (Noble, 1952). Meaningfulness interacts with position, so the material shapes the serial-position curve itself.

Glossary

Anticipatory error.

An error in which an element due later in a sequence is produced too early; evidence, central to Lashley's critique, that later items are active before earlier ones are completed and so cannot be triggered by a strict chain.

Chaining hypothesis.

The proposal that a serial list is learned as a chain of item-to-item associations, each item serving as the stimulus for the next; contradicted by remote associations and by anticipatory errors in skilled sequences.

Context retrieval and updating (CRU).

Logan's model in which serial order is produced by encoding a context representation built from the items and updating it as each is produced; proposed to unify serial order across perception, cognition, and action.

Derived list.

A re-spacing of an already-learned list that preserves its items but changes their intervals, for example pairing every second item; used by Ebbinghaus to measure remote associations through savings.

Functional (effective) stimulus.

The cue that actually controls a serial response; the central question of whether it is the preceding item, the item's ordinal position, or a compound context determines which model of serial order is correct.

Isolation effect.

The superior learning of a list item made distinctive from its neighbours; also called the von Restorff effect, it raises an item above the serial-position curve independently of where it sits.

Meaningfulness (m).

Noble's continuous measure of a verbal unit, defined by the average number of associations it elicits in a fixed interval; it predicts the rate at which serial lists are learned.

Positional coding.

The representation of an item's order by a code for its position in the sequence, rather than by links to neighbouring items; the dominant modern account of serial order.

Primacy effect.

The superior learning and recall of items at the beginning of a list; the larger and more durable of the two end advantages in serial learning.

Primacy Model.

Page and Norris's model of immediate serial recall in which order is stored as a primacy gradient of activation, strongest for the first item and declining across the list, with items selected strongest-first and suppressed once recalled.

Proactive interference.

The disruption of memory for a list by lists learned earlier; Underwood's demonstration that most laboratory forgetting over a day is proactive rather than caused by the passage of time.

Recency effect.

The superior recall of items at the end of a list; pronounced in immediate free recall, where it reflects a short-term store, and abolished by a filled delay, but more modest in serial anticipation.

Remote association.

An association formed between non-adjacent items of a serial list, revealed by savings on derived lists; its strength declines with the distance between items, which rules out a chain of purely adjacent links.

Retroactive interference.

The disruption of memory for an earlier list by learning a later one; shown by Melton and Irwin to grow with interpolated learning and to involve active unlearning.

Savings method.

Ebbinghaus's measure of retention: the reduction in trials or time needed to relearn a list to criterion compared with learning it originally; sensitive to order information that unaided recall cannot detect.

Serial anticipation method.

A procedure, historically run on a memory drum, in which list items are exposed one at a time at a fixed rate and the participant must produce the next item before it appears; learning is indexed by trials to criterion or by the proportion of correct anticipations.

Serial learning.

The learning of a set of items in a fixed sequential order, such that each item can be produced in its correct place; with paired-associate learning, one of the two central paradigms of verbal learning.

Serial-position curve.

The function relating an item's ordinal position to how readily it is learned or recalled; in serial learning a bowed curve, low at the start, peaking in difficulty just past the middle, and intermediate at the end.

Start-End Model.

Henson's positional-coding model in which each item's position is coded as its distance from the start and the end of the sequence, reproducing the bowed difficulty curve and the locality gradient of transposition errors.

Transposition error.

An order error in which a list item is recalled in the wrong position; transpositions cluster around the item's correct place, forming a locality gradient that is the primary evidence for positional coding.

Trials to criterion.

The number of study trials a participant needs to reach a fixed standard of mastery, typically one errorless recitation of the whole list; the classic measure of serial-learning rate.

Key Researchers

Hermann Ebbinghaus

(1850-1909). Founded the experimental study of serial learning. Working on himself with lists of nonsense syllables learned in order, he derived the serial-position effect, the savings method, and the remote associations of derived lists, the first evidence that serial associations are not confined to adjacent items. See his Wikipedia biography and Wikidata record.

Richard N. A. Henson

(b. 1970). Professor of Cognitive Neuroscience at the MRC Cognition and Brain Sciences Unit, University of Cambridge. His Start-End Model codes serial order by each item's distance from the start and end of the sequence, reproducing the bowed difficulty curve and the locality gradient of transposition errors. ORCID 0000-0002-0712-2639; faculty page; Google Scholar.

Karl S. Lashley

(1890-1958). Posed the problem of serial order in behavior, arguing that the speed and the anticipatory errors of skilled sequences such as speech and typing rule out a stimulus-response chain and require a central plan that holds the whole sequence in graded readiness. His 1951 essay reoriented the field from associative chains toward positional and hierarchical schemes. See his Wikipedia biography and Wikidata record.

Gordon D. Logan

(living). Centennial Professor of Psychology at Vanderbilt University. His context retrieval and updating (CRU) model proposes a single mechanism for serial order across perception, cognition, and action, built on a context representation composed of the items themselves. ORCID 0000-0002-8301-7726; faculty page.

Bennet B. Murdock

(1925-2022). Turned the serial-position curve into a clean modern result, systematically varying list length and presentation rate to produce the canonical functions still printed in textbooks, and built mathematical models of the order information that distinguishes serial from free recall. See his Wikipedia biography and Wikidata record.

Klaus Oberauer

(living). Professor of Cognitive Psychology at the University of Zurich. Led the consortium that set out the agreed benchmark findings, including the serial-order error patterns, that any model of short-term and working memory must reproduce, giving the modelling literature a shared empirical target. ORCID 0000-0003-3902-7318; Wikidata record.

Frequently Asked Questions

What is serial learning?

Serial learning is the learning of a set of items in a fixed order, so that the learner can reproduce the whole series in sequence. It is a form of verbal learning and, with paired-associate learning, one of the two paradigms on which the experimental study of memory was built. What distinguishes it is that order itself must be reproduced, not just the items.

How does serial learning differ from free recall?

In free recall the participant may report a studied list in any order and is scored only on which items appear. In serial learning an item counts as an error if it is produced out of place, so the task isolates order information from item information. The two tasks also yield different serial-position curves: a bowed curve with strong primacy in serial learning, a pronounced recency arm in immediate free recall.

What is the serial-position curve in serial learning?

It is the bowed function relating an item's position to how readily it is learned. Errors are fewest at the start of the list, rise to a maximum just past the middle, and fall again toward the end, with the primacy advantage larger than the recency advantage. The curve is asymmetric, unlike the more symmetric U of immediate free recall.

What are remote associations?

Remote associations are links that form between non-adjacent items of a serial list. Ebbinghaus revealed them by relearning derived lists that skipped items: savings appeared even when items were two or three steps apart, and declined with distance. Their existence shows a serial list is not learned as a chain of adjacent links alone.

Why did the chaining hypothesis fail?

Chaining says each item is the sole cue for the next. It cannot explain remote associations, nor the anticipatory errors of skilled sequences, in which a later element intrudes early, which show that later items are active while earlier ones are still being produced. Lashley's 1951 analysis made this the decisive case against a stimulus-response chain.

What is positional coding?

Positional coding is the idea that order is carried by a representation of each item's position in the sequence rather than by links to its neighbours. Henson's Start-End Model codes position as distance from the start and end markers, which reproduces both the bowed difficulty curve and the tendency for order errors to be transpositions to nearby positions.

Why are order errors usually transpositions to nearby positions?

Because neighbouring positions have similar positional codes. When the code is retrieved noisily, it is more likely to return an adjacent item than a distant one, producing a sharp locality gradient of transpositions. This structured pattern of errors is the main evidence that order rests on a positional code.

Is serial learning still studied today?

Yes. The phenomena are now the benchmarks against which computational models of serial order are tested. Current work debates whether order is carried by the items themselves, as in Logan's CRU model, or also requires an item-independent positional signal, as Osth and Hurlstone argue, a live question framed by the error patterns catalogued by Oberauer and colleagues.

References

Ebbinghaus, H. (2013). Memory: A contribution to experimental psychology. Annals of Neurosciences, 20(4), 155-156. (Original work published 1885) https://doi.org/10.5214/ans.0972.7531.200408

Glanzer, M., & Cunitz, A. R. (1966). Two storage mechanisms in free recall. Journal of Verbal Learning and Verbal Behavior, 5(4), 351-360. https://doi.org/10.1016/S0022-5371(66)80044-0

Henson, R. N. A. (1998). Short-term memory for serial order: The Start-End Model. Cognitive Psychology, 36(2), 73-137. https://doi.org/10.1006/cogp.1998.0685

Hurlstone, M. J., & Hitch, G. J. (2018). How is the serial order of a visual sequence represented? Insights from transposition latencies. Journal of Experimental Psychology: Learning, Memory, and Cognition, 44(2), 167-192. https://doi.org/10.1037/xlm0000440

Lashley, K. S. (1951). The problem of serial order in behavior. In L. A. Jeffress (Ed.), Cerebral mechanisms in behavior: The Hixon Symposium (pp. 112-146). Wiley.

Logan, G. D. (2021). Serial order in perception, cognition, and action. Psychological Review, 128(1), 1-44. https://doi.org/10.1037/rev0000253

Melton, A. W., & Irwin, J. M. (1940). The influence of degree of interpolated learning on retroactive inhibition and the overt transfer of specific responses. American Journal of Psychology, 53(2), 173-203. https://doi.org/10.2307/1417415

Murdock, B. B. (1962). The serial position effect of free recall. Journal of Experimental Psychology, 64(5), 482-488. https://doi.org/10.1037/h0045106

Noble, C. E. (1952). An analysis of meaning. Psychological Review, 59(6), 421-430. https://doi.org/10.1037/h0054087

Oberauer, K., Lewandowsky, S., Awh, E., Brown, G. D. A., Conway, A., Cowan, N., Donkin, C., Farrell, S., Hitch, G. J., Hurlstone, M. J., Ma, W. J., Morey, C. C., Nee, D. E., Schweppe, J., Vergauwe, E., & Ward, G. (2018). Benchmarks for models of short-term and working memory. Psychological Bulletin, 144(9), 885-958. https://doi.org/10.1037/bul0000153

Osth, A. F., & Hurlstone, M. J. (2023). Do item-dependent context representations underlie serial order in cognition? Commentary on Logan (2021). Psychological Review, 130(2), 513-545. https://doi.org/10.1037/rev0000352

Page, M. P. A., & Norris, D. (1998). The primacy model: A new model of immediate serial recall. Psychological Review, 105(4), 761-781. https://doi.org/10.1037/0033-295X.105.4.761-781

Postman, L. (1962). Transfer of training as a function of experimental paradigm and degree of first-list learning. Journal of Verbal Learning and Verbal Behavior, 1(2), 109-118. https://doi.org/10.1016/S0022-5371(62)80007-3

Postman, L., & Underwood, B. J. (1973). Critical issues in interference theory. Memory & Cognition, 1(1), 19-40. https://doi.org/10.3758/BF03198064

Tulving, E., & Thomson, D. M. (1973). Encoding specificity and retrieval processes in episodic memory. Psychological Review, 80(5), 352-373. https://doi.org/10.1037/h0020071

Underwood, B. J. (1957). Interference and forgetting. Psychological Review, 64(1), 49-60. https://doi.org/10.1037/h0044616

von Restorff, H. (1933). Über die Wirkung von Bereichsbildungen im Spurenfeld. Psychologische Forschung, 18(1), 299-342. https://doi.org/10.1007/BF02409636