Models of Memory

Memory can be classified in various ways, according to different aspects of the material remembered and the processes involved. Memory is by its nature

something that persists over time, so memories can be classified according to how long they are remembered for.

There are two broad categories of memory: declarative and procedural (see figure 3.2). Declarative (or explicit) memory is memory for facts or knowledge that we have conscious access to, whereas procedural (or motor) memory is unconscious memory for learned tasks or skills, or „knowing how‟ memory (Cohen & Squire, 1980; Longstaff, 2005). Within declarative memory, there is a temporal component, so declarative memory can be split up according to how long something is

remembered for, into sensory memory, short-term memory and long-term memory. It can also be split according to the information remembered into episodic memory, or memory for specific events associated with a particular time and place, and semantic memory, memory of facts unrelated to specific events (Tulving, 1972). The focus of this introduction will be declarative memory, as this is the main focus within the current study. Whilst it is beyond the scope of this study, readers are directed to (Gazzaniga et al., 2009) for a more detailed outline of further types of memory. Firstly, temporal classifications of memory and how they interact will be considered.

Sensory memory is over milliseconds or seconds, and does not require specific paying of attention. The auditory „echo‟ that persists for a few seconds can be retrieved, even if attention is not paid, and is known as the sensory memory trace or sensory register. For audition, it is echoic memory whereas for vision it would be iconic memory (Gazzaniga et al., 2009). These sensory traces are thought to decay very quickly, and are considered not accessible to conscious awareness, but can

on (Sperling, 1960; Gazzaniga et al., 2009). Iconic memory is visual information present for only a few hundred milliseconds which then rapidly decays, whereas the echoic trace for auditory information is thought to last up to 10 seconds (Sams et al., 1993; Gazzaniga et al., 2009).

Figure 3.2: Tree of memory

Short term memory is, in contrast to sensory memory, freely available to our conscious awareness, however it has a much more limited capacity. It is only a temporary store, viable for seconds to minutes, and material requires repeated rehearsal to keep it there. Material in short term memory is thought to be lost by decay of information (hence the need for repeated rehearsal) or by disruption from other input, or a combination of both (Gazzaniga et al., 2009). Primacy and recency effects are demonstrated in normal subjects when testing short term memory, for example by repeating back a list of unrelated words (Glanzer & Cunitz, 1966). Primacy is greater recall of material at the start of the list, because these have been rehearsed most and been transferred to long term memory, whereas recency is greater recall of material at the end of the list, because this has had less time to decay so is still in short term memory. This is known as the serial position effect (see

Memory

Declarative Procedural

Short term memory

Long term memory

Working memory Sensory Episodic memory Semantic memory

figure 3.3) (Glanzer & Cunitz, 1966). Primacy is affected by the speed of

presentation, and is eliminated if the material is presented too quickly, whereas recency is eliminated by distraction tasks after the presentation (Gazzaniga et al., 2009).

Figure 3.3: Serial Position Effect (Glanzer & Cunitz, 1966)- reproduced with permission

N.B. Spacing refers to the length of time between presentation of the words in the word list

There are several models of short term memory. The Atkinson and Shiffrin modal model (Atkinson & Shiffrin, 1968) suggests sensory information first enters a sensory register, or sensory memory, and then attentional processes move certain items into the short term memory. Items then move into the long term memory by repeated rehearsal, and at each stage information can be lost by interference, decay or a combination of the two. By this hierarchical model, items are passed from sensory memory to short term memory, and only then to long term memory, but this view is disputed by other experimental evidence, particularly studies of patients with brain damage which have been useful in examining how memory functions. Case studies of patients with very limited short term memory but almost intact abilities to form long term memories suggest that short term memory cannot be the only gateway to long term memory (Shallice & Warrington, 1970;

based on the same neural networks as long term memory, but they are not activated in quite the same way (Zola-Morgan & Squire, 1993; Ranganath & Blumenfeld, 2005).

Another proposed model is the levels of processing model (Craik & Lockhart, 1972). This model proposes that there are other factors influencing what information is passed to long term storage, including that the „deeper‟ an item is processed, the better it is stored in long term memory. This suggests elaborate encoding and relating information to previously acquired knowledge provides better learning than storing information as simple visual or verbal codes.

The working memory model (Baddeley & Hitch, 1974) is a widely accepted model proposed to explain some of the shortcomings of short-term memory models. Working memory is seen as a limited capacity store for both retaining information in the short term, and also processing that information during complicated tasks. It can involve information straight from the sensory input and information retrieved from the long term memory put together to enable processing complex tasks, such as driving (Longstaff, 2005). A multi-component model of working memory has been proposed (Baddeley & Hitch, 1974) and since updated (Baddeley, 2000; Repovs & Baddeley, 2006) to explain the shortcomings of the unitary model proposed by Atkinson and Shiffrin in 1968, initially comprising of a central executive assisted by two storage systems: the phonological loop and the visuospatial sketchpad. This has since been updated and a fourth component introduced to the model: an episodic buffer (Baddeley, 2000; Repovs & Baddeley, 2006). This provides a model of working memory as a limited capacity store providing an interface between perception, long term memory and action.

The phonological loop subsection of this working memory model is a short lived store for information presented acoustically, and coding mechanism for that

information by repeated rehearsal sub-vocally. Subjects with problems in short term phonological memory tend to have lesions in the left temporo-parietal area

(Warrington et al., 1971; Vallar et al., 1997), suggesting this may be the anatomical base for the phonological loop, an idea backed up by functional neuroimaging studies suggesting the phonological loop is based in the left pre-frontal and parietal regions (Cabeza & Nyberg, 2000).

The visuospatial sketchpad is an equivalent store and processing mechanism for visual information. It provides a capacity to hold and manipulate visuospatial information, and it is thought that it might be possible to separate the visuospatial sketchpad into visual and spatial memory components (Repovs & Baddeley, 2006). Lesions particularly in the right hemisphere can lead to deficits in visuospatial working memory (Hanley et al., 1991), and in functional neuroimaging studies, bilateral parietal activation is generally associated with spatial working memory (Smith et al., 1996; Cabeza & Nyberg, 2000).

The central executive mechanism acts as a control system, overseeing the two subsystems and managing and directing attentional processes (Repovs & Baddeley, 2006). This is the least researched component of the original working memory model, as well as the least understood (Repovs & Baddeley, 2006). Executive functioning has been linked to the frontal lobes in neuroimaging studies (Smith & Jonides, 1997).

The idea of an episodic buffer has been added to the working memory model to explain some of the unexplained issues from the previous model. For example, it accounts for the advantage in recall for semantically linked words, provides an explanation for the links between working and long term memory, and explains how the two slave-systems (phonological loop and visuo-spatial sketchpad) interact and information from them binds together (Repovs & Baddeley, 2006). It is

proposed to integrate information from both the working memory systems and long term memory, and represents a separate storage system using a multi-modal code (Repovs & Baddeley, 2006).