How is the three system approach to memory best conceptualized

Abstract

Learning and memory functions are crucial in the interaction of an individual with the environment and involve the interplay of large, distributed brain networks. Recent advances in technologies to explore neurobiological correlates of neuropsychological paradigms have increased our knowledge about human learning and memory. In this chapter we first review and define memory and learning processes from a neuropsychological perspective. Then we provide some illustrations of how noninvasive brain stimulation can play a major role in the investigation of memory functions, as it can be used to identify cause–effect relationships and chronometric properties of neural processes underlying cognitive steps. In clinical medicine, transcranial magnetic stimulation may be used as a diagnostic tool to understand memory and learning deficits in various patient populations. Furthermore, noninvasive brain stimulation is also being applied to enhance cognitive functions, offering exciting translational therapeutic opportunities in neurology and psychiatry.

Encoding/Retrieval Interactions and Their Implications

Processes of learning and memory are typically conceptualized as involving three stages: encoding, storage, and retrieval. Encoding is the initial registration and acquisition of information, storage is the maintenance of information over time in the nervous system (represented as a memory trace), and retrieval is the process whereby stored information is brought back into conscious awareness or otherwise affects ongoing behavior. Strength theory essentially proposes that encoding conditions will produce main effects on performance measured on different memory tests and will never interact with retrieval conditions, because different memory tests simply vary in terms of their sensitivity to memory strength. However, the literature on human memory is replete with examples in which encoding and retrieval conditions interact. The example mentioned previously, that high-frequency words are better recalled than low-frequency words are, whereas low-frequency words are better recognized than high-frequency words are, is one example of an encoding/retrieval interaction.

Larry Jacoby has designed compelling experiments demonstrating that two different measures of memory with great surface similarity can be uncorrelated or even negatively correlated. Participants in Jacoby's experiments were presented with lists of words under various study conditions and were given one of two different memory tests. One group of individuals was given a standard yes/no recognition memory test, in which they were presented with a long list of test words and were asked to determine which words had been previously studied. The other group of individuals was given a test that involved identifying words presented at very fast rates (around 30 msec per word). The proportion of words correctly identified was the dependent measure. Some of the words flashed during the test had been presented on the study list but other test words had not been previously studied. In this speeded word identification test, the improved ability to name the briefly flashed words that had been presented during the study phase is known as priming.

In one of Jacoby's experiments, the independent variable was the level of processing of words during the study phase. Students were presented with one of three questions that oriented them toward either the surface features of the target word (e.g., “Is the word in all capitals?”), the sound of the word (e.g., “Does the word rhyme with chair?”), or the meaning of the word (e.g., “Is the word a type of animal?”) before the presentation of each target word (e.g., BEAR). These three different orienting questions manipulated the level of processing that individuals performed on each word. The effects of levels of processing on performance in the two different memory tests are depicted in Fig. 2. In the recognition memory test, the typical levels of processing effect was observed: individuals were best at recognizing words they had processed at a meaningful level and worst at recognizing words they had processed at only a surface level, whereas processing the sounds of the words produced intermediate recognition performance. In contrast, consider performance on the speeded word identification test, shown in the right panel of Fig. 2. In this test, priming was measured as the difference in performance between naming studied and nonstudied words. Although all of the priming scores were positive, indicating retention of the studied words, all three encoding conditions produced equivalent levels of priming! Levels of processing, which had such a profound effect on recognition memory, had no effect on priming. Although both tests were measuring retention of the same list of items, the two measures were completely uncorrelated in this experiment.

In another experiment, Jacoby demonstrated that measures of recognition memory and speeded word identification could even be negatively correlated. In this experiment, individuals studied a list of words under one of three different encoding conditions: they were either asked to read the target words in a neutral context (XXXX-cold), to read each word paired with its opposite (hot-cold), or to generate each target word given its opposite (hot-????). Thus, in all three conditions, participants said out loud the same list of target words, but the means of having participants produce the words differed dramatically. The effects of these encoding manipulations on performance in the two different memory tests are shown in Fig. 3. Although generating the target words produced the best performance on the recognition test and reading the words in a neutral context produced the worst performance (a finding known as the “generation effect”), the opposite pattern of results was observed in the speeded word identification test: reading the words produced better identification performance than did generating the words! The results of Jacoby's experiment demonstrate that two measures of memory that appear to be very similar on the surface may be negatively correlated with each other under some circumstances.

Interactions between encoding and retrieval conditions demonstrate that measures of retention can reveal positive, zero, or negative correlations with one another. These encoding/retrieval interactions have an important implication for understanding human memory: although the concept of memory is labeled with a single word, it is hardly a single entity. Many different types of memory exist, and there are several different valid measures of memory. Some of the most prominent ways to measure memory are considered in the following discussions.

2.1 Definitions and Concepts

We use the words learning and memory routinely in ordinary discourse but they are also scientific concepts, defined formally by psychologists and neuroscientists. Lay usage of the term ‘learning’ is generally restricted to situations where there is some element of deliberation or intent – such as in learning a language or learning to drive. One would not, for example, ordinarily learn what one had for breakfast. In contrast, memory tends to be used most frequently in reference to the recall of events that, at the time they happen, we do not deliberately memorize – as in remembering what happened last Christmas.

In contrast, formal psychological definitions of these terms do not entail any reference to intent. Learning is generally defined as ‘the act of acquiring information or skill such that knowledge and/or behaviour change’. It may occur in a variety of different ways.

Memory is defined in at least two ways. It is used to refer to a presumed ‘mental storage device’ in which information may be held, as in the concept of a phonological store. Additionally, it is used to refer to a putative ‘capacity of mind’, as in the concept of episodic memory. Psychologists recognize different types of memory, distinguished in relation to the types of information they process (e.g. words vs pictorial information), their capacity or persistence (e.g. short-term vs long-term), and their operating characteristics (e.g. the mental codes in which information is held).

Definitions of learning and memory in neurobiology bring in such factors as the neuroanatomical localization of a putative system, or the physiological and cell-biological mechanisms involved. In stepping along this path, neuroscientists recognize that learning is an act of acquiring information, as psychologists define it, but also assert that it is a process that is thought to engage specific areas of the brain, to depend on specific patterns of neural activity and, importantly, to result in biological changes in brain cells that outlast the learning experience.

Similarly, the term memory is also widely used alongside specific networks in the brain, such as a group of structures or set of neuronal connections that is thought to carry out memory functions. The ‘medial temporal lobe memory system’ is such a concept (Squire, 1992), as is the idea of amygdala-dependent memory (LeDoux, 2000).

In his pioneering book on computational aspects of vision, Marr (1982) distinguished what he referred to as computational, algorithmic and implementation levels of analysis in information processing science. This tripartite distinction has been useful, but it does not map very directly onto the numerous levels of analysis at which individual neuroscientists operate.

Contemporary approaches to learning and memory are concerned with linking these levels of analysis, but this is far from easy, largely because most neuroscientists find themselves at the limit of their understanding when they stray outside the disciplines in which they were trained. Few seem to realize the complexity of developing a ‘general theory of memory’ that would link the many levels at which it can be analysed.

Whereas neuropsychologists are interested in understanding the mapping of psychological process onto neuroanatomical structures and networks, using patients and functional magnetic resonance imaging, fMRI (see Chapter 4 ‘Advanced neuroscience technologies’), their eyes tend to glaze over when attention turns to the chemical pathways that mediate biological changes at the cell level within neurons. Conversely, whereas the ‘autophosphorylation of the alpha sub-unit of calcium-calmodulin kinase within the postsynaptic density of glutamatergic neurons’ is the stuff of coffee-room debate in hard-core neuroscience departments, the relevance of this and other biochemical mechanisms to explicit or implicit memory might not grab the same level of their attention (Fig. 9.1).

We have begun by contrasting the approach adopted by experimental psychologists with that of neurobiologists. One meeting point of these cultures has to do with the fundamental property of memory. This is that memory is of necessity a change in the brain that outlasts the stimuli that trigger it. The change may be entry into an active state of reverberation amongst a network of neurons or a physical change in neurons thought to mediate more lasting memories. This brings us to the concept of a ‘memory trace’ – the physical ‘substrate of storage’ (Hebb, 1949).

Changes at synapses – the connection points between neurons – are currently the favoured locations for storage of long-term memory traces. Much current research focuses on how synapses change in strength (Martin et al., 2000). Indeed, synaptic neurophysiological researchers often describe plasticity as a model of memory, with its neural mechanisms the focus of interest, including the autophosphorylation of αCAMKII. Study of these physiological mechanisms, together with computational modelling, has revealed the possible existence of many different learning rules that could determine whether a trace is stored and how it represents information within various kinds of neural network (Rumelhart and McClelland, 1986; O'Reilly, 1998). Patterns of neural activity serve as memory cues and reactivate traces later. The resulting output is what psychologists and neuroscientists agree as being memory.

Another meeting point for experimental psychologists and neurobiologists is that we all recognize that we can subdivide learning and memory into distinct temporal phases or processes – encoding, storage, consolidation and retrieval. Encoding has to do with the formation of memories – what must happen for a memory to form in the first place. Storage has to do with what lasts in the mind or brain, with different kinds of storage device mediating short-term and long-term memory. Retrieval refers to the process of memory reactivation. The concept of consolidation refers to something that happens to memory traces after they have been stored and that alters their persistence or sensitivity to brain damage (McGaugh, 2000). This ‘something’ is not the same as memory retrieval per se, although one view of consolidation is that it entails repeated acts of retrieval and re-storage that may even happen during sleep.

5.0 A debate: is consciousness needed for episodic learning?

We learn about objects and scenes by paying attention to them. As mentioned in Chapter 8, the most obvious result of selective attention is that we tend to become conscious of the objects of attention, as we can prove by reporting our conscious experiences. Episodic memory is generally defined as memory for specific conscious episodes, like the sight of a coffee cup. However, we have also cited evidence that the hippocampal system can be stimulated by unconscious events, such as a subliminal picture of snakes or of emotional facial expressions.

Whether stimuli need to be conscious to lead to episodic memory is therefore a subject of debate. Because it is difficult to ensure that conscious and unconscious brain stimulation lead to comparable MTL activity, the results of that debate are still unclear.

5.3 Implicit learning in language

The past few years have witnessed the emergence of increasing studies of implicit learning in language. This is perhaps not so surprising in that language acquisition, like implicit learning, involves incidental learning conditions. Further, cogent use of language likewise does not require explicit knowledge of grammar. Recently, several authors have begun to explore this connection empirically. For instance, Saffran and colleagues (1997) showed how incidental exposure to artificial language-like auditory material (e.g., bupadapatubitutibu…) was sufficient to enable both children and adult subjects to segment the continuous sequence of sounds they had heard into the artificial words (e.g., bupada, patubi, etc.) that it contained, as evidenced by their above-chance performance in a subsequent recognition test.

Based on these data, Saffran and colleagues suggested that the word segmentation abilities demonstrated by these subjects were due to the transitional probabilities of successive syllables which are higher within words than between words. Saffran and colleagues interpreted their findings as representing a form of implicit learning. The connection is obvious when one recognizes that language acquisition, like implicit learning (Berry & Dienes, 1993; Cleeremans, 1993) is likely to involve, at least in part, incidental learning of complex information organized at differing levels.

Part of the convergence between language acquisition and implicit learning suggested by Saffran and colleagues can be attributed to the impact of computational modeling on the field of memory research. For instance, connectionist models such as the Simple Recurrent Network have been extensively used with significant success in both the language acquisition and implicit learning domains (Christiansen et al., 1998; Redington & Chater, 1997). In effect, the problems faced in both domains are quite similar: how to best extract structure from a complex stimulus environment characterized by “deep” systematic regularities when learning is incidental rather than intentional. The answer, in both domains, appears to be embodied by distributional approaches.

Figure 9.14 brings out several features of learning and memory. Notice that conscious cognition leads to explicit learning and memory retrieval in this figure. An obvious example is deliberately trying to memorize a technical term in cognitive neuroscience. What may not be so obvious, however, is that implicit learning also happens along with learning of conscious or explicit stimuli.

Figure 9.14. Implicit and explicit (conscious) learning. This version of our functional diagram suggests that there are two ways for information in working memory to lead to long-term memories. In the case of “explicit learning,” a conscious event (an “episode”) is registered in episodic memory (gray boxes at the bottom). However, a great many things we learn are implicit, such as the implicit inferences we make from the two sentences like “The glass broke. It shattered on the kitchen floor.” The complete meaning of those sentences is stored in memory, including the idea that glass is brittle. This is an example of an implicit memory.

Source: Baars.

Thus Figure 9.14 shows both explicit or conscious and implicit or unconscious learning. Episodic memory is the storage of conscious episodes (also called autobiographical memory). Semantic memory, usually viewed as memory for facts, is also conscious, in the strict sense that people can accurately report the facts they believe. This is the standard operational definition of conscious brain events (see Chapter 8). Finally, perceptual memory capacities, such as our ability to “learn to hear” music and art, also involve conscious, explicit kinds of memories.

On the right-hand side of Figure 9.14, we also see the learning of implicit memories. Infants may hear sequences of speech sounds, but they are not explicitly learning the rules and regularities of grammar. Those are apparently learned unconsciously, as we will see later. In general, implicit learning is often evoked by explicit, conscious events, but it often goes far beyond the events given in conscious experience (Banaji & Greenwald, 1995). Overpracticed habits and motor skills are also largely implicit. As we will see, priming effects are often implicit. Contextual phenomena are often implicit, such as the assumptions we make about visual space, the direction of the incoming light in a visual scene, the conceptual assumptions of a conversation, and so on. These are often hard to articulate, implicit, and to some degree are unconscious (Baars, 1988).

Figure 9.8 shows one version of learning with consolidation, in which input into the neocortex and the hippocampal regions (MTL) evoke an active state, with neuronal processes making new synaptic connections. As just mentioned, immediate memory is encoded in improved synaptic connectivity between billions of neurons in the neocortex. Normal sleep, especially the slow wave stage, is important to turn these temporary connectivities into long-lasting memory traces.

Learning and memory serve a critical function in allowing organisms to alter their behavior in the face of changing environments. This chapter considers the nature and mechanisms of emotional learning and memory, particularly the acquisition and expression of memory for aversive (fearful) events. Of particular importance in this regard is work with animals that has taken advantage of invasive techniques to yield extensive information into the biology of emotional learning and memory systems. Hence, extensive coverage is given to the anatomy and physiology of brain systems involved in fear memory based on experimental investigations in animals.

Learning and memory are intensively studied topics in modern brain and cognitive science. Drosophila has been used in the study of visual learning and memory for approximately the past 20 years. In this chapter, we discuss the architecture and function of fruit fly’s visual system, which provides it with the sensory and neural substrate for color, motion, and shape vision. We then discuss visual learning, visual memory, invariant pattern recognition, selective attention, choice behavior, context generalization, spatial learning, habit formation, cross-modal memory transform, and synergism. Finally, we address the issues of collective learning and social decision making.

WIDE RANGE ASSESSMENT OF MEMORY AND LEARNING (WRAML)/WIDE RANGE ASSESSMENT OF MEMORY AND LEARNING—2 (WRAML-2)63,64

The WRAML (ages 5-17) and WRAML-2 (ages 5-90) are designed to test visual and verbal memory. The WRAML-2 contains six core subtests (the WRAML has nine): Story Memory, Verbal Learning, Design Memory, Picture Memory, Finger Windows, and Number/Letter Memory. Verbal Memory Index (Story Memory, Verbal Learning), Visual Memory Index (Design Memory, Picture Memory) and Attention/Concentration (Finger Windows, Number/Letter Memory) summary scores are obtained (M = 100, SD = 15). There are optional Sentence Memory, Sound-Symbol, Verbal Working Memory, and Symbolic Memory subtests. Delayed recall and recognition memory can also be assessed. A General Memory Index is computed from the core subtests. Scores on Memory Screening, consisting of the first four core subtests (taking 20 minutes), correlate highly with those of the General Memory Index (r = 0.91). In contrast to the WRAML, there is no Learning Index in the WRAML-2. The WRAML-2 also allows assessment of primary/recency effects, immediate/delayed recall, rote versus meaningful information, visual/verbal differences, working memory, short-term memory, sustained attention, and recognition versus retrieval memory. This test is useful in evaluation of children with learning disorders, those suspected of having verbal processing problems, and those suspected of having ADHD.

Verbal Learning and Memory

VLM are two of the most extensively examined impaired CF in CBUrs. A growing number of studies suggest performance deficits in acutely intoxicated subjects, in terms of immediate and delayed recalls of words, intrusion, and learning (Solowij & Battisti, 2008; Gonzalez et al., 2012). A review focused on the long-term effect of CBU use on memory suggested that impairments are not dissimilar to those associated with acute intoxication, and have been related to the duration, frequency, dose, and age of onset of CBU (Solowij & Battisti, 2008). A study on CBU and VLM found significant associations between certain components of VLM and frequency of use, cumulative lifetime dose, and duration of regular use (Wagner, Becker, Gouzoulis-Mayfrank, & Daumann, 2010).

Finally, recent CBUrs demonstrated significantly worse performance than nonusers, across cognitive domains of AT/WM, information processing speed, and EF (Thames et al., 2014). There were no statistically significant differences between recent users and past users on neurocognitive performance. Frequency of CBU in the last 4 weeks, and the amount of daily CBU, were negatively associated with neurocognitive performance. Although some of these widespread adverse effects of CBU on neurocognitive functioning appear to attenuate with abstinence, past users’ neurocognitive functioning was consistently lower than nonusers.

Procedural learning and memory involve the acquisition, consolidation, and retrieval of individual representations that are behaviorally expressed in an inflexible manner. Acquisition of stimulus–response and stimulus–affect associations represents prominent forms of procedural learning. Extensive evidence indicates that these forms of procedural learning are mediated by relatively independent neural systems that contain the dorsal striatum and amygdala as primary components, respectively. This article provides a brief description of evidence implicating the dorsal striatum and amygdala in procedural learning and memory, focusing on studies employing brain lesion and pharmacological approaches in lower animals.

7.02.3.1.5 Summary

Learning and memory are highly interrelated and cannot be fully understood independently of each other. Most studies on memory in aging are cross-sectional and there is typically a lack of information necessary for ruling out the influence of the health status of the participants. The system model of memory employed in recent research usually distinguishes between episodic, semantic, primary, and working memory. Memory performance is seen as influenced by numerous properties acting at both encoding and retrieval stages of memory processing. Task properties may generally be recognized in terms of cognitive support (low–high), although there is no evidence of a simple relation between level of support and performance. The relatively few studies on episodic memory performance in very old age portray a gradual decline into the very late stages of the life span.