Theories of memory generally consider both the architecture of the memory system and the processes operating within that structure. Architecture refers to the way in which the memory system is organised, and process refers to the activities occurring within the memory system. Architecture or structure and process are both important, but some theorists emphasise only one in their theoretical formulations.
Learning and memory involve a series of stages. Processes occurring during the presentation of the learning material are known as “encoding”. This is the first stage. As a result of encoding, some information is stored within the memory system. Thus, storage is the second stage. The third, and final, stage is retrieval, which involves recovering or extracting stored information from the memory system. We have emphasised the distinctions between architecture and process and among encoding, storage and retrieval. However, we cannot have architecture without process, or retrieval without previous encoding and storage. It is only when processes operate on the essentially passive structures of the memory system that it becomes active and of use.

Architecture of memory
The multi-store model is based on common features of different theories. Three types of memory store were proposed:
1.
Sensory store: modality specific, brief
Most information in the environment is not attended to. The iconic store (or visual store) has limited capacity and duration. Sperling’s (1960) study found that information decayed in about 0.5 seconds. The echoic store is a transient auditory store holding relatively unprocessed input.
Ioannides et al. (2003) found evidence for the echoic memory lasting 5 seconds in the left hemisphere and 2 seconds in the right hemisphere.
2.
Short-term store: limited capacity and fragile storage
Short-term memory is very limited; for example, digit and letter spans typically have a value of 7 plus or minus 2. Miller argued that the capacity of short-term memory should be assessed by number of chunks – integrated pieces of units of information. Span is less for larger chunks.
Information is lost from short-term memory through decay and interference. Berman et al. (2009) claim that new information enters short-term memory and disrupts the information already there.
INTERACTIVE EXERCISE: Capacity of short-term memory
INTERACTIVE EXERCISE: Duration of short-term memory
3.
Long-term store: unlimited capacity over long periods
Double dissociation indicates that two tasks probably involve different processing mechanisms and provides strong evidence for the distinction between short-term and long-term memory store.
Patients with amnesia have impairments of long-term memory but typically have no problem with short-term memory. They typically have damage to the medial temporal lobe. Patients may also have impaired short-term memory but intact long-term memory (see, e.g., Shallice & Warrington,
1974). These patients typically have damage to the parietal and temporal lobes.
Some information is attended to and processed by the short-term store. Some information is transferred from short- to long-term store. Transfer to long-term storage depends on rehearsal.

While the multi-store model has proven useful in explaining memory, it is oversimplified and some of the assumptions are not correct: The short-term memory is not a gateway to long-term memory (they interact); the model does not account for implicit learning; items that are attended to can be accessed more readily than other items in short-term memory and the model exaggerates the role of rehearsal.
Jonides et al. (2008 ) have argued that the multi-store approach should be replaced with a unitary-store model. It is proposed that amnesic patients have problems with relational memory – forming novel relations or “binding” between items and their context. This is a role played by the hippocampus. Hannula et al. (2006 ) showed that amnesic patients were able to form strong long-term relational memories, and this could account for their good performance on short-term memory tasks.
There is evidence that activation of long-term memory plays a role in short-term memory. However, the unitary-store approach has limitations: It is oversimplified in assuming that short-term memory is only activated long-term memory. Hippocampal involvement is greater in long-term memory than short-term memory and neuroimaging studies do not support the unitary-store model.
According to the multi-store theory, there are separate sensory, short-term and long-term stores. There is strong evidence to support the notion of various qualitatively different memory stores, but this approach provides a very oversimplified view. For example, multi-store theorists assumed that there are unitary short-term and long-term stores, but the reality is more complex. Alternatively, unitary-store models assume that short-term memory consists of temporary activations of long-term memory. The difficulties that amnesic patients have are explained by a problem with forming novel relations. There is some evidence for unitary-store models, but these are still oversimplifications of the structure of memory.
WEBLINK: An article on techniques for memory retention

Working memory
Baddeley and Hitch (1974) and Baddeley (1986) replaced the concept of the short-term store with that of working memory. Baddeley (2001 ) described four components of working memory:

Central executive: o modality-free; o limited capacity; o resembles attention; o deals with cognitively demanding tasks.

Phonological loop: o holds information in speech-based form; o limited capacity.

Visuo-spatial sketchpad: o storage and manipulation of spatial and visual information; o limited capacity.

Episodic buffer
Assumptions from this theory:

If two tasks use the same component, they cannot be performed successfully together.

If two tasks use separate components, it should be possible to perform them as well together as separately.

In a series of dual-task studies, Robbins et al. (1996) showed that selecting chess moves involved the central executive and visuo-spatial sketchpad but not the phonological loop.
Phonological loop
The phonological similarity effect is good evidence for the role of the phonological loop in short-term memory tasks. Larsen et al. (2000 ) found that serial recall was 25% worse if a list of words was phonologically similar.
RESEARCH ACTIVITY: Phonemic similarity
INTERACTIVE EXERCISE: Encoding in short-term memory
The word-length effect ( Baddeley et al., 1975 ) is the finding that memory span is better for shorter words than for longer words, which have a longer articulatory duration.This suggests that capacity of the phonological loop is determined by temporal duration.
According to Baddeley (1986, 1990), the phonological loop consists of:
 a passive phonological store concerned with speech perception;
 an articulatory process linked to speech production.
Baddeley suggests auditory presentation of words has direct access to the phonological store, but visual presentation only has indirect access via subvocal articulation. There is support from research with patients with brain damage. Patients with damage to left inferior parietal cortex have a damaged phonological store but intact articulatory control. Patients with damage to left inferior frontal cortex have intact phonological store but damaged articulatory control. Baddeley et al. (1998 ) suggest that the phonological loop may be more important in learning new words than familiar ones. The capacity of the phonological loop predicts vocabulary size.
Visuo-spatial sketchpad
The visuo-spatial sketchpad is used in the temporary storage and manipulation of spatial and visual information. Logie (1995) argues for two components of visuo-spatial working memory:

The visual cache o stores form and colour information.

The inner scribe o processes spatial and movement information; o rehearses and transfers information from the visual cache to the central executive.
There is experimental support for the notion of separate visual and spatial systems. Klauer and Zhao (2004 ) found a spatial interference task disrupted performance more on a spatial main task than a visual main task.
Smith and Jonides (1997 ) found differences in brain activation during different visual and spatial tasks. The ventral prefrontal cortex is more activated during visual working-memory tasks and the dorsal prefrontal cortex is more activated during spatial working-memory tasks. Many tasks require both components to be used in combination. It is unclear how information from both components is combined and integrated.
Central executive
The central executive is the most important and versatile component of working memory. It is thought to involve the prefrontal cortex. rTMS to the dorsolateral prefrontal cortex disrupts performance on many complex cognitive tasks. Individuals with dysexecutive syndrome have problems with planning, organising, monitoring and initiating behaviour. They typically have damage within the frontal lobes.

The conceptualisation of the central executive has changed over time. Originally, the central executive was viewed as being unitary. More recent proposals highlight several executive processes such as:

Baddeley (1996 ): o Switching of retrieval plans. o Timesharing in dual-task studies. o Selective attention. o Temporary activation of long-term memory.

Miyake et al. (2000 ): o Inhibition function – the ability to deliberately inhibit dominant, automatic or prepotent responses when necessary, for example in the Stroop effect. o Shifting function – shifting back and forth between multiple tasks, operations or mental sets. o Updating function – updating and monitoring of working memory representations.
INTERACTIVE EXERCISE: Stroop – automatic processing
Latent variable analysis on various executive function tasks identifies three separable functions (Miyake et al., 2000). Each executive process activates a different region in the prefrontal cortex. However, all tasks activate the intraparietal sulcus. Executive processes are needed in dual-task situations. Patients with
Alzheimer’s disease are more sensitive than controls to dual-task demands, but cope equally well with single tasks. This suggests Alzheimer’s patients have damage to executive processes involved in dual-task processing.
Based on studies with frontal lobe-damaged patients, Stuss and Alexander (2007 ) disagreed with the notion of a single dysexecutive syndrome arising from frontal lobe damage. Instead, they proposed three executive processes associated with different parts of the frontal lobe:

Task setting (left lateral frontal): setting a stimulus–response relationship.

Monitoring (right lateral frontal): checking the task and behavioural adjustment.

Energisation (superior medial frontal): initiation and sustaining of a response.
These processes are not independent and are used across a wide range of tasks.
The central executive consists of various related but separable executive processes. Accumulating evidence points to inhibition, updating, shifting and dual-task coordination as four major executive processes. The notion of a dysexecutive syndrome is misleading as it suggests a single pattern of impairment. Further research is needed to clarify differences in the executive processes suggested by imaging on healthy individuals and those with frontal lobe damage.
Episodic buffer
Main characteristics ( Baddeley, 2000 ):

Integrates and stores information from various sources into coherent complex structures or episodes.

Acts as an intermediary between the subsystems, combining them into multi-dimensional representations.

Has close links to the central executive.
The episodic buffer was proposed because components in the original model were too separate in their functioning and did not allow for combination and integration of information (e.g., Chincotta et al., 1999 ).

WEBLINK: Iowa Gambling Task demonstration
CASE STUDY: The episodic buffer
Baddeley and Wilson (2002 ) argued that high levels of immediate prose recall depends on the capacity of the episodic buffer and its efficiency. As predicted by this model, patients with impaired long-term memory still have good immediate prose recall, and this recall is better if the amnesic patient has good executive functioning, compared to those with more severe executive deficits.
The episodic buffer is a valuable addition to the working memory model and accounts for our ability to integrate different kinds of information in working memory. Some progress has been made to identifying brain areas associated with the episodic buffer, for example the hippocampus. The use of the episodic buffer is sometimes associated with the central executive. However, it is difficult to design tasks to distinguish clearly between the episodic buffer and other components.
INTERACTIVE EXERCISE: Working memory
There are several advantages of the working memory system over Atkinson and Shiffrin’s short-term store.
Working memory:
 involves both active processing and transient storage of information;
 explains the partial short-term memory deficits sometimes caused by brain damage;
 incorporates verbal rehearsal as an optional process.
However:

It is difficult to identify the number and nature of executive processes associated with the central executive.

The understanding of the main executive processes associated with the central executive is not clear.

More research is needed on the interactions between the four components of the working memory system.
Baddeley replaced the unitary short-term store with a working memory system consisting of three components: an attention-like central executive, a phonological loop holding speech-based information and a visuo-spatial sketchpad specialised for spatial and visual coding. Later, a fourth component – the episodic buffer – was added, allowing information from various sources to be integrated and combined. Some progress has been made in identifying the brain regions involved in various components of the working memory system, particularly from studies with brain-damaged patients. Although originally proposed to be unitary, the central executive is now thought to involve various separable executive processes such as inhibition, updating, shifting, updating and dual-task coordination. It is also thought that the episodic buffer is often used in conjunction with the central executive. However, it remains a challenge to identify the number and nature of executive processes, and to isolate their contributions experimentally.

Working memory capacity
Working memory capacity refers to how much information an individual can process and store at the same time. Working memory capacity correlates with fluid intelligence (Unsworth, 2010). Participants with high working memory capacity have superior executive and/or attentional processes. This is because there are

greater individual differences with these functions than those of the visuo-spatial sketchpad and the phonological loop. It has been measured using:
 the reading span task;
WEBLINK: Reading span task
 the operation span.
WEBLINK: Video of the operation span task
Engle and Kane (2004) indicate that there are two factors involved in working memory: the maintenance of task goals and the resolution of response conflict. Working memory capacity correlates with brain activity, such that, under distraction, those with high working memory capacity show reduced ERP responses than those with low working memory capacity (Yurgil & Golob, 2013). This work suggests working memory capacity is related to fluid intelligence and the ability to respond to novel stimuli and situations.

Levels of processing
Craik and Lockhart (1972 ) proposed a framework detailing various levels of stimulus processing from shallow or physical analysis to deep or semantic analysis. The key theoretical assumptions were:

The level or depth of processing of a stimulus has a large effect on its memorability.

Deeper analysis produces more elaborate, longer-lasting and stronger memory traces.
They argued that maintenance rehearsal, simply repeating previous analyses, does not enhance long-term memory.
Many studies support the main assumptions of the levels-of-processing approach. Craik and Tulving (1975 ) showed depth of processing had major effects on memory performance – performance was three times better with deep than shallow processing. There is more brain activation during semantic than perceptual processing. Craik and Tulving (1975 ) argued that elaboration (the amount of processing of a particular kind) is important. Eysenck and Eysenck (1980) tested this and found words in a non-semantic, distinctive condition were better recognised than non-semantic, non-distinctive words, and almost as well as semantic words. Level-of-processing effects are greater in explicit memory tasks than in implicit memory tasks
(Challis et al., 1996).
Craik and Lockhart (1972) argued correctly that processes during learning have a major impact on subsequent long-term memory. Learning and remembering are by-products of perception, attention and comprehension. This approach has led to the identification of elaboration and distinctiveness as important factors in learning and memory. However, the approach possesses several limitations: the importance of the retrieval environment on performance was underestimated; the relative importance of depth, elaboration and distinctiveness of processing is unclear; many of the terms have not been defined with any precision; and it does not explain why deep processing is so effective.
INTERACTIVE EXERCISE: Levels of processing
Craik and Lockhart (1972) focused on learning processes in their levels-of-processing theory. They (and their followers) identified depth of processing (i.e., the extent to which meaning is processed), elaboration of processing and distinctiveness of processing as key determinants of long-term memory. Insufficient

attention was paid to the relationship between the processes at learning and those at the time of test. Other problems are that the theory is not explanatory, that it is hard to assess the depth of processing and that shallow processing can lead to very good long-term memory.

Learning through retrieval
Practice in retrieving to-be-remembered information during the learning period can enhance long-term memory more than simply engaging in study and restudy. This strong effect is known as the testing effect
(Karpicke, 2012). Dunlosky et al. (2013) have discussed ten techniques of the use of restructuring information in order to improve learning. Pyc and Rawson (2010, 2012) have suggested that, during learning through testing, people find mediators or links between the to-be-remembered items. However,
Peterson and Mulligan (2013) have demonstrated that learning through testing can limit learning by preventing people from detecting underlying structures.
Research on this topic has been primarily conducted through paired-associate learning and therefore may not generalise to other forms of learning. The factors that underlie mediator effectiveness are not well understood. The theory also doesn’t predict results indicating that testing can be less effective than restudying.
WEBLINK: Learning through testing
The studies here show that, in order to improve learning, it can be effective to test yourself during the learning process.

Implicit learning
Implicit learning is learning without conscious awareness of having learned – the process through which we become sensitive to certain regularities in the environment (Cleeremans & Jiménez, 2002):
 in the absence of intention to learn about these regularities;
 in the absence of awareness that one is learning;
 in such a way that the resulting knowledge is difficult to express.
There are similarities between implicit learning and implicit memory but also differences pertaining to the type of knowledge acquired and stimuli used in measurement. Reber (1993) proposed five characteristics of implicit memory and learning that differ from explicit memory and learning:
 age independence

IQ independence
 robustness
 low variability
 commonality of process.
Shanks and St John (1994) propose two criteria to show implicit learning:
1.
Information criterion – the information provided on the awareness test must be the information that is responsible for the improved performance.
2.
Sensitivity criterion – the test of awareness must be sensitive to all relevant knowledge.
There are several experimental tests developed to test implicit memory:
 process-dissociation procedure (Joordens et al., 2010);

WEBLINK: Process-dissociation
 serial reaction time task (Shanks, 2010);
 dual-task performance (Jiménez & Vázquez, 2011).
It is thought the striatum is associated with implicit learning, while explicit learning involves the anterior cingulate and dorsolateral prefrontal cortex. Functional neuroimaging studies have provided limited support for this but have failed to demonstrate clear differences in patterns of activity for implicit and explicit learning tasks. Destrebecqz et al. (2005 ) used inclusion and exclusion conditions of the process dissociation procedure. They found activation in the striatum was associated with implicit learning, while the mesial prefrontal cortex and anterior cingulate were associated with explicit learning.
Failure to discover clear differences in patterns of brain activation could occur because the tasks used are not pure measures of the two forms of learning.
Amnesic patients typically perform poorly on tests of explicit memory but well on tests of implicit memory
(Van Tilborg et al., 2011). However, their performance on implicit learning tasks is variable. Patients with
Parkinson’s disease have damage to the striatum. They show impaired explicit learning, but performance on implicit learning tasks was variable (Foerde & Shohamy, 2011) and poor performance was more likely if they were given feedback.
Much experimental evidence suggests an important distinction between implicit and explicit learning.
There is accumulating evidence that explicit learning is associated with the prefrontal cortex and anterior cingulate, while implicit learning is associated with the striatum. Amnesic patients generally show evidence of intact implicit learning despite their impaired explicit learning. However, there are limitations of research on implicit learning. It is hard to devise tests of awareness that can detect all the task-relevant knowledge of which participants have conscious awareness. Some explicit learning is involved in purported implicit learning tasks. There are individual differences in the extent to which learners are consciously aware of what they are learning.
It has been claimed that implicit learning differs from explicit learning in terms of the presence or absence of consciously accessible knowledge. Much evidence supports the distinction between implicit and explicit learning, and amnesic patients often show intact implicit learning but impaired explicit learning. In addition, there is some evidence from neuroimaging studies that different brain areas are involved in explicit and implicit learning. However, it has proved difficult to show that claimed demonstrations of implicit learning satisfy the information and sensitivity criteria, and most tasks are likely to involve aspects of both implicit and explicit learning.

Forgetting from long-term memory
Forgetting was first studied in detail by Hermann Ebbinghaus (1885/1913). His basic measure of forgetting was the savings method – the reduction in number of trials for relearning compared with original learning.
Findings suggest the forgetting function is approximately logarithmic (Rubin & Wenzel, 1996). Forgetting is fastest shortly after learning and the rate then decreases with time.
Forgetting through decay serves the function of not allowing many trivial memories to be stored (Hardt et al., 2013).

Interference theory assumes the ability to remember can be disrupted by what we have previously learned
(proactive interference) or by future learning (retroactive interference). Both proactive and retroactive interference are maximal when two different responses are associated with the same stimulus.
Proactive interference can be caused by competition between the correct and incorrect response(s). Jacoby et al. (2001 ) argued that proactive interference might occur for two reasons:
1.
problems in retrieving the correct response (discriminability);
2.
strength of the incorrect response learned initially (bias or habit).
Research shows the latter to be the case.
Bergström et al. (2012) discovered that proactive interference involves automatic and controlled retrieval processes. Rememberers’ memory search is generally too broad, including material previously learned but currently irrelevant. Therefore, making participants aware of proactive interference reduces its effect
(Wahlheim & Jacoby, 2011).
Retroactive interference explains why people forget some of their first language when acquiring a second one (Misra et al., 2012). Retroactive interference is greatest when new learning resembles previous learning. Lustig et al. (2004 ; see Eysenck & Keane, 2010, p. 237) identified two explanations for retroactive interference in paired-associate learning:
1.
problems with active searching for the correct response (controlled processes);
2.
high accessibility of the incorrect response (automatic processes).
Dewar et al. (2007 ) found retroactive interference can occur in two ways:
1.
expenditure of mental effort during the retention interval;
2.
learning of material similar to the original material.
There is extensive evidence for both proactive and retroactive interference. However, the theories lack details. Research is limited in several ways. Interference theory does not explain why the rate of forgetting decreases over time.
Freud (1915, 1943) argued that very threatening or anxiety-provoking material is often unable to gain access to conscious awareness – he called this repression. Repression is sometimes an active and intentional process. Non-experimental evidence in which adults recover repressed memories of childhood abuse exists but there is controversy, as some argue that these recovered “memories” are false.
WEBLINK: Jim Hopper’s website on recovered memories
Freud failed to consider the cognitive processes involved in the repression of traumatic memories.
Motivated forgetting can be adaptive for forgetting upsetting or outdated memories. Directed forgetting is a phenomenon involving impaired long-term memory caused by an instruction to forget some information presented for learning. The forget instruction has its effects during learning. Participants may selectively rehearse “remember” items at the expense of “forget” items. Direct forgetting in recall also involves retrieval inhibition or interference.
Endel Tulving (e.g., 1979) argued that forgetting typically occurs when there is a poor match or fit between the information in the memory trace and that available at retrieval. This led him to propose the encoding specificity principle . Context is important in determining forgetting: information about current mood state

is stored in the memory trace; there is less forgetting when the mood state at learning and retrieval is the same (mood-state-dependent memory); linguistic context is also important for memory; and the effects of context are often stronger in recall than recognition memory. Godden and Baddeley (1975) found participants recalled more words when learning and recall took place in the same environment (on land or underwater). Park and Rugg (2008 ) found that, for items that were successfully recognised, pictures received more “picture-like” processing and words were processed in a “word-like” way. However, Rugg et al. (2008) reported that the similarity in patterns of brain activation at learning and retrieval was not great.
The overlap between information stored in the memory trace and that available at retrieval plays an important role in whether retrieval occurs. Recent neuroimaging evidence supports both the encoding specificity principle and transfer-appropriate processing. There is also a valuable emphasis on contextual information in retrieval. However, the theory is only directly applicable to simple memory tasks. It is based on the assumption that retrieval occurs automatically. The measurement of informational overlap is not typically done, leading to circular reasoning. It is not clear when matches between information in the environment and stored memory traces lead to conscious recollection and when they do not (Nairne, 2015).
Context influences are greater on recall than recognition.
Wixted (2004 ) argued that the secret of forgetting may lie in consolidation theory . Consolidation is a process lasting for several hours, or possibly even days, which fixes information in long-term memory.
It is assumed the hippocampus plays a vital role in the consolidation of memories and that recently formed memories are still being consolidated and are vulnerable to interference and forgetting.
Consolidation involves two major phases:

The first phase occurs over a period of hours and centres on the hippocampus.

The second phase takes place over days to years and involves interactions between the hippocampal region, entorhinal region and the neocortex.
Consolidation theory is relevant to Jost’s law and Ribot’s law (effects of brain injury on memory are greater on newly formed memories). Predictions of consolidation theory fit with the typical forgetting curve. In patients with retrograde amnesia, hippocampal damage may have a permanently adverse effect on consolidation processes, thus resulting in the most recently formed memories being most impaired ( Manns et al., 2003 ). People who drink excessive alcohol sometimes suffer from “blackout”. This is probably due to a failure in consolidation. Interestingly, memories formed immediately before drinking alcohol are often better remembered, and this may be due to alcohol preventing the formation of new memories that would interfere (Moulton et al., 2005).
When a memory trace is reactivated a new consolidation process ( reconsolidation ) occurs. This makes the memory trace fragile again. Misinformation provided during this reconsolidation process can make the memory vulnerable to disruption and therefore reduce their accuracy (Chan & LaPaglia, 2013).
While consolidation theory is very useful in explaining the rate of forgetting (for example), the theory ignores other factors in the forgetting process. Consolidation theory does not explain why proactive and retroactive interference are greatest when two different responses are associated with the same stimulus.
The theory may also be an oversimplification.
WEBLINK: Consolidation theory

Freud argued for the importance of repression, in which threatening material in long-term memory cannot gain access to consciousness. There is controversial evidence of recovered memories in adults who claim to have suffered childhood abuse. Strong effects of proactive and retroactive interference have been shown in the laboratory. However, insufficient attention has been paid to active strategies that individuals use to minimise interference effects. Much forgetting is probably cue dependent, and the cues can be either external or internal (e.g., in mood-state-dependent memory). However, it is not clear that forgetting over time can be explained in cue-dependent terms. It may well be that forgetting over time depends mostly on failures of consolidation. Consolidation theory provides an explanation for the form of the forgetting curve and for reduced forgetting rates when learning is followed by sleep or alcohol.
Additional references
Chincotta, D., Underwood, G., Abd Ghani, K., Papadopoulou, E. & Wresinksi, M. (1999). Memory span for Arabic numerals and digit words: Evidence for a limited-capacity visuo-spatial storage system.
Quarterly Journal of Experimental Psychology , 2A: 325–51.
Freud, S. (1943). A general introduction to psychoanalysis . New York: Garden City.
Rugg, M.D., Johnson, J.D., Park, H. & Uncapher, M.R. (2008). Encoding–retrieval overlap in human episodic memory: A functional neuroimaging perspective. In W. Sossin, J.C. Lacaille, V.
Castellucci, & S. Belleville (eds), Essence of memory . Oxford: Elsevier.