MULTIPLE MEMORY SYSTEMS

Human memory is commonly divided into memory systems which operate on different characteristic time-scales. An example of this is the coarse division of memory into short-term and long-short-term memory. One influential model of short-short-term memory was formulated by Baddeley & Hitch (1974). The Baddeley-Hitch model is not a simple model of short-term encoding, storage, and retrieval of information but includes components which are thought to support several higher cognitive functions, including reasoning and language (Baddeley, 1986). The Baddeley-Hitch model is therefore a model of working memory. In the original Baddeley-Hitch model, working memory consists of a central executive with two support systems, the phonological loop, for short-term encoding and storage of verbal information, and the spatial sketch pad, for short-term encoding and storage of visuo-spatial information. Recently Baddeley (2000) added another component, the episodic buffer to the working memory model (Figure 4.2).

[Figure 4.2] The extended working memory model of Baddeley and Hitch. The episodic buffer comprises a limited capacity system that provides temporary storage of information held in a multimodal code, capable of binding information from the subsidiary

systems, and from long-term memory, into an episodic representational format. The episodic buffer is supposed to provide an interface to the other slave systems of working memory and to long-term memory, feeding information into and retrieving information from episodic long-term memory.

Baddeley (2000) suggests that the episodic buffer is a capacity limited system that provides temporary storage of information held in a multimodal code, which is capable of binding information from the subsidiary systems as well as from long-term memory into an episodic representational format. The episodic buffer shares some characteristics with the concept of episodic memory (Tulving, 1989) with respect to its principal mode of storing information in episodes and its integrative aspects, but differs in that it is assumed to be a temporary store. In the extraction of information from working memory a key function for the episodic buffer is integration between the different subcomponents of working memory.

The episodic buffer is thought to provide an interface between the components of working memory and long-term memory. In emphasizing its short-term integrative role and its episodic format, one may hypothesize that the episodic buffer is related to the prefrontal cortex (PFC) and the medial temporal lobe (MTL) as well as the interaction between these structures. The transient early role of the MTL system in long term memory formation and sequence encoding in conjunction with the PFC makes these likely candidates (Eichenbaum, 2000; Simons & Spiers, 2003). The functional anatomical correlate of the phonological store is putatively in the left inferior parietal region (Brodmann’s area [BA]

39/40) together with parts of the superior temporal cortex (Becker, MacAndrew, & Fiez, 1999; Paulesu, Frith, & Frackowiak, 1993), and the articulatory rehearsal process involving a left frontal circuit including Broca's region (BA 44) and the pre-motor cortex (BA 6, Smith & Jonides, 1998, 1999).

Human long-term memory is also commonly subdivided in different component memory systems (Tulving & Schacter, 1994) and although the concepts and terminology used to characterize these memory systems has varied, there is a consensus concerning the broad division of human memory into declarative and non-declarative memory (Figure 4.3). Declarative memory supports the capacity to encode, store, and retrieve facts and

events and is contrasted with a heterogeneous collection of non-declarative memory abilities including skills and habits (Knowlton, Mangels, & Squire, 1996), different forms of conditioning (Bechara et al., 1995), and repetition priming (e.g., facilitation of recognition, reproduction or biases in selection of stimuli that have recently been perceived, Schacter, 1994). The knowledge or information acquired by non-declarative memory systems is commonly expressed through performance changes rather than explicit retrieval.

Different forms of non-declarative memory depend on the integrity of specific brain systems; for example the basal ganglia, the amygdala, and the cerebellum (Eichenbaum &

Cohen, 2001).

skills/habits priming conditioning other LTM

declarative

semantic episodic

non-declarative

[Figure 4.3] Taxonomy of human long-term memory (LTM) systems.

Another commonly used distinction is that between explicit and implicit memory. The terms explicit and implicit memory usually refer to forms of memory expression. In this usage, implicit memory denotes the expression of memory without awareness of its

acquisition or use; that is, behavioral expressions of what an individual has learnt without remembering how, when, or where the learning occurred. In contrast, explicit memory commonly refers to the expression of what the individual is aware of and can explicitly report if probed (Tulving, 1995). In the following we will focus on declarative memory, but we will also, briefly, mention some of the non-declarative memory system.

Explicitly retrieved declarative memories are commonly conceptualized as integrated associative structures, which are continuously updated through active re-organization and integration of new information within the context of previous experiences and previously acquired knowledge (Eichenbaum, 2000). Recollection of memories represents a re-construction (re-creation) process which is partly determined by the nature and organization of the stored information as well as previously acquired knowledge. This type of memory, declarative memory, involves the representation of episodic information within the context of general knowledge. It is thought that episodic representations encode sequences of micro-events and micro-features that compose unique, individual experiences, indexed by specific times and places. Semantic (general) knowledge, on the other hand, represents an acquired knowledge base of organized and inter-related factual information, which is independent of the specific episode(s) in which the information was acquired (Eichenbaum, 2000). General world knowledge is not tied to a specific time and place of acquisition. Declarative memory thus represents the capacity to form and retrieve episodic and semantic information. A key feature of declaratively stored information is its flexible accessibility and expressibility that can be used adaptively in novel situations in an elaborate manner (i.e., flexible memory expression, Eichenbaum & Cohen, 2001; Schacter

& Tulving, 1994), for example, to solve new problems and support the inferential expression of associations that are linked across separated experiences; the medial temporal lobe (MTL) memory system might play a role in integrating overlapping experiences into general knowledge in terms of reorganization, abstraction, and re-integration of episodic information (Eichenbaum, 2000).

The declarative memory system has a well-defined neuroanatomic correlate in the MTL memory system (Squire, 1992; Squire et al., 2004; Squire & Zola-Morgan, 1991).

However, it should be noted that trace-conditioning is a form of complex conditioning which depends on the MTL (in particular trace conditioning, see e.g., Takehara, Kawahara,

& Kirino, 2003; Weiss, Bouwmeestev, Power, & Disterhoft, 1999), while simple conditioning seems not to depend on the MTL. In the trace-conditioning paradigm the conditioned- and unconditioned stimulus are separated by a relatively long stimulus-free interval, and it might be the case that trace-conditioning depends on associative and temporal integration capacities of the MTL. The MTL memory system is composed of three principal components: neocortical regions, the parahippocampal region and the hippocampus (Amaral, 1993; Amaral, 1999; Squire & Zola-Morgan, 1991; Suzuki, 1996, cf. Figure 4.4). The neuroanatomic organization complements the findings from studies of amnesia, suggesting that the MTL contribute to declarative memory by altering the nature and persistence as well as organization of stored memory representations in the neocortex (Eichenbaum, 2000). In contrast, the MTL memory system is not essential for non-declarative memory, which include the acquisition of perceptual, cognitive, and motor skills, as well as acquired habits and learned response biases (cf. e.g., Packard & Knowlton, 2002). These forms of memory are expressed implicitly through performance alterations (e.g., changes in error patterns, improved response times or performance scores) on a variety of tasks (cf. e.g., Knowlton et al., 1996; Knowlton & Squire, 1996; Petersson, Forkstam, & Ingvar, 2004; Poletiek, 2002; Salmon & Butters, 1996; Squire, 1994; Squire et al., 1993; Stadler & Frensch, 1998). For example, systems that include the basal ganglia and cerebellum mediate forms of implicit learning and non-declarative (procedural) memory. It now seems clear that the basal ganglia, in particular the dorsal striatum, play a role in learning and memory (Packard & Knowlton, 2002). Moreover, recent evidence suggests that the basal ganglia and the MTL memory systems can be activated simultaneously during learning and that in some learning situations competitive interference exists between these two systems (Poldrack et al., 2001; Poldrack, Prabhakaran, Seger, & Gabrieli, 1999). However, recent FMRI data indicate that the caudate nucleus and the MTL can interact non-competitively and that the caudate nucleus is not only engaged after repeated practice, but also after single-trial learning and thus in parallel with the hippocampus (Voermans et al., 2004).

Another prominent structure of the MTL is the amygdala and some forms of affective learning and memory rely on a system that includes the amygdala as a core structure. This memory system mediates fear conditioning as well as other forms of

emotional memory (Bechara et al., 1995; Cahill, Babinsky, Markowitsch, & McGaugh, 1995). Here emotional memory refers to the formation of affective representations that is not necessarily available for explicit retrieval but can be implicitly expressed in for example attraction- and avoidance behavior, as well as in the modulation of autonomic nervous system responses. However, the amygdala appears to have a broader role in human long-term memory. In particular, the amygdala, which is a part of the anterior MTL, has prominent recurrent connections with the hippocampus and the MTL memory system.

Thus, it seems that the amygdala is well placed anatomically to modulate declarative memory. For example, a time-varying learning rate that changes with the relevance of the information being processed, opens up for the possibility to control learning rate by various relevance or 'print-now' signals. This mechanism can be used to make the memory selective and modulated by relevance (cf., appendix 2.2). Several functional neuroimaging studies have investigated the role of the amygdala in enhancing declarative memory for emotional experiences and suggested a correlation between amygdala activation during encoding and subsequent memory. For example, the degree of activity in the left amygdala during encoding was predictive of subsequent memory (Canli, Zhao, Brewer, Gabrieli, & Cahill, 2000). Furthermore, it has also been suggested that the amygdala may play a role in modulating the strength and consolidation of memories in other memory systems (Cahill et al., 1995, cf. chapter 2 and appendix 2.2).