CHAPTER TWO Memory 2.1 Memory as a Psychological Construct 2.1.1 Introduction Memory is perceived as the outcome of several cognitive functions, and refers to the ability to encode, retain, and retrieve information at different points in time (Vanderploeg, 2000). Moreover, memory is not a unitary system but is fractioned into different components (Stracciari, Ghidoni, Guarino, Poletti & Pazzaglia, 1994). Memory is broadly differentiated into shortterm store and long-term store (Neath & Suprenant, 2003). Within these major divisions, memory is differentiated into types and aspects. Recall and recognition are types of memory, and are used to assess the manner in which information is encoded and retrieved. Furthermore, verbal memory and visual memory describe the memory stores for auditory-verbal, and visual-spatial and visual-verbal information. 2.1.2 Short-term Memory Short-term store is the temporary store for newly registered information, has a limited capacity system, and lasts for less than one minute, unless it is lengthened by rehearsal (Lezak, 2004). Information passes through short-term store to enter longterm store, from where long-term memories are retrieved 8 (Vanderploeg, 2000). Immediate, working, primary, and secondary memory are aspects of short-term store, and relate to encoding processes (Matlin, 2001). Primary memory refers to the aspect of the short-term store that retains information that utilizes current attention and current thoughts. The information in this store does not have to be consciously recalled in order to be utilized. While short-term memory highlights the role of time in remembering, primary memory underscores the role of attention, conscious processing and memory capacity (Neath & Suprenant, 2003). By contrast, secondary memory is an aspect of the short-term store that retains information that is currently being processed with the knowledge that this information has been previously experienced or thought of. Working memory is an aspect of short-term memory that describes a workspace or memory buffer, which retains information that is currently being processed (Baddeley, 1992). Working memory is used interchangeably with immediate memory. This aspect of memory refers to the sensory processing of information in working memory (Matlin, 2001). 2.1.3 Short-term Memory Span 2.1.3.1 Auditory-verbal short-term memory span Span refers to the measurement capacity of memory, and is tested with digit span or word span tasks. The capacity of short-term memory is not determined by the physical properties of a 9 stimulus. Instead, it depends on one’s ability to chunk or recode a stimulus into higher-order units (Matlin, 2001). The digit span of short-term memory is limited to seven (plus or minus two) random digits whereas the word span retains four to five random words. This phenomenon is referred to as the span of apprehension (Baddeley & Hitch, 1992). Letters load the short-term capacity to the same extent as digits. In addition, random words load the short-term capacity to a greater extent than a meaningful sentence (Parkin, 1993). While the increased length of sentence tasks improves performance, the increased length of a list span tasks reduces performance (Matlin, 2001). This occurs because list span depends on the auditory familiarity of the stimulus while sentence span relies on intact comprehension (Matlin, 2001). Although, the short-term store is limited regarding the amount and duration of information it can store, rehearsal boosts the short-term span (Matlin, 2001). Span impairment is characterized by the difficulty in retaining information (Matlin, 2001). 2.1.3.2 Visual short-term memory Visual span for symbolic or meaningful stimuli can be tested by briefly exposing the stimulus. This prevents recoding stimuli 10 into a spoken form. Under these conditions, four to five items can be retrieved in normal individuals (Matlin, 2001). The retention of single stimulus items is intact in individuals who exhibit impaired visual retrieval (Kinsbourne & Warrington, 1962) [cited in Matlin, 2001]. Moreover, when items are presented in pairs, retrieval is possible for only one of the items. Errors of omission and substitution are made, and recall of the second item is not possible. The impairment does not result from perceptual or attention deficits, but is specific to symbolic stimuli that cannot be verbally encoded (Kinsbourne & Warrington, 1962) [cited in Matlin, 2001]. Furthermore, auditoryverbal short-term span can be normal in the presence of an impaired visual-verbal span (Parkin, 1993). Visual-verbal span tasks tend to dissociate from tasks that require the retention of the spatial organization of a stimulus. Corsi & Milner (1971) [cited in Neath & Suprenant, 2003] developed a test whereby nine identical blocks are distributed such that no definable pattern in the arrangement emerges. In the presentation of the block, the examiner taps a sequence of blocks at a rate of one block per second. The subject is required to reproduce the sequence immediately afterward. Selective impairment on this task reflects a span of three items rather than the normal five to six item span. If the identical arrangement is presented with numbers visible on the stimuli, 11 performance is within normal levels (De Renzi, Luchelli, Muggia, & Spinnler, 1995). 2.1.4 Serial position curve The serial position curve reflects the free recall of the serial presentation of items as a function of each item’s position in an item list (Parkin, 1993). The curve can be divided into three portions, namely, recency, which refers to the enhanced recall of the last few items on the list; primacy, which refers to the enhanced recall of the first few items on the list; and finally, the asymptote which describes the low performance in the middle of the curve (Parkin, 1993). A popular notion is that recency reflects the effortless output of the short-term store whereas primacy occurs via the long-term store (Matlin, 2002). The transfer of information from the short-store to the long-term store critically depends on consciously recycling information (Matlin, 2001). Rundus (1971) [cited in Parkin, 1993] asserted that the different parts of the curve have different properties that reflect the operation of different stores. The serial position curve is the joint product of the short-term and long-term stores. In addition, words in the primacy and asymptote portions of the serial position curve reflect a positive relationship between rehearsal frequency and improved recall. Moreover, variations in word frequency, presentation rate, interitem relatedness, list length, and mental status significantly 12 affect the primacy and asymptote portions but not the recency segment. Lezak (2004) asserts that amnesic patients exhibit normal recency, and impaired primacy. Recent items are easily recalled since they are strongly represented, while the primacy effect results from items early in the list receiving more rehearsal (Parkin, 1993). However, there is no correlation between rehearsal frequency and the recency effect. Recent items do not have sufficient time to be incorporated into the rehearsal process, thus recall depends entirely on the short-term store. Furthermore, the size of the recency effect cannot be measured in word units, and depends on the nature of the target information and the duration of a distraction task (Glanzer & Razell, 1974) [cited in Parkin, 1993]. The negative recency effect (Craik, 1970) [cited in Parkin, 1993] formed another arm of the short-term/long-term dichotomy. Craik (1970) administered a series of free recall word trials; each of which generated a typical serial position curve. On completion of the experiment, a surprise recall test of all possible words was administered. Words that reflected recency effects in the immediate recall trials, later had a significantly lower level of recall, even when compared to words formerly in the asymptote. Craik attributed this effect to the failure of rehearsing the recent items since the second test was unexpected. 13 These items had a lower chance of being transferred to the longterm store and were, therefore, difficult to recall at a later time. The constant ratio rule states that the possibility that a single theory may account for both types of recency effects. 2.1.5 Level of Processing Model The level of processing model (Craik & Lockhart, 1972) [cited in Matlin, 2001] conceptualizes memory in terms of the encoding processes undertaken during learning, and the pattern of subsequent retention (Matlin, 2001). (Craik & Lockhart, 1972) [cited in Parkin, 1993] proposed that the retention of a stimulus is a positive function of the depth to which the stimulus is processed during learning, hence, the depth effect. The depth effect is corroborated by established research on incidental learning (Neath & Suprenant, 2003). This concept refers to the relative ease of retention that occurs when one is not asked to retain information during a task. Information is encoded along a continuum of encoding dimensions, while processing occurs at three different levels (Craik & Lockhart, 1972). Orthographical processing produces the poorest level of memory, followed by phonological processing. By contrast, semantic processing produces the highest level of retention, and performance is as good as that found in intentional learning. These results are attributed to processing that occurs via a central processor (Craik and Lockhart, 1972). 14 Possessing a limited but flexible capacity, the central processor processes new information. In addition, retention depends on the way in which the central processor is deployed during learning, with deeper levels of processing yielding better retention. A stimulus first undergoes a minimal core encoding, which includes a degree of semantic analysis (Parkin, 1993). This is followed by conscious and directed processing appropriate to the task demands. The notion of a continuum between the shallow and deeper levels of processing was abandoned, and replaced by separable domains of processing (Sutherland, 1972) [cited in Matlin, 2001]. The revised theory reflects the original idea that it is not possible to gradually pass from one level to the next because each level contains qualitatively different dimensions of information processing, which cannot merge in any way (Matlin, 2001). Stimuli that evoke a “yes” response rather than a “no” response, are retained better (Parkin, 1993). This phenomenon is termed the congruency effect. The effect is pronounced in the semantic domain, and ensues since “yes” responses are more likely to allow coherent associations between the stimulus and the retrieval cues (Parkin, 1993; Matlin, 2001). In addition, longer processing time does not reflect deeper processing or greater retention (Craik & Lockhart, 1972). Thus, memory can be affected by qualitative differences in the way information is encoded (Matlin, 2001). 15 The Level of Processing model was criticized for proposing that semantic processing typically produces better retention than nonsemantic processing, since the former involves deeper processing than the latter (Matlin, 2001). However, the enhanced retention following semantic processing depends on the type of retention test used (Craik and Lockhart, 1972). Thus, semantic processing does not have an absolute advantage over non-semantic processing. This phenomenon is referred to as transfer-appropriate processing, which holds that the most appropriate learning strategy is the one that most closely addresses the information required at testing. 2.1.6 Working Memory The commonly received notion was that a single, indivisible structure, primary memory or the short-term store underlies the memory capacity essential for normal conscious mental activity. Memory is organized and assumes an orderly transition from shortterm to long-term storage (Matlin, 2001). Memory span is accepted as the definitive measure of short-term storage, therefore performance on the digit span is related to long-term storage. Refuting this unitary view of short-term store, Baddeley and Hitch (1977; 1992) [cited in Matlin, 2001] suggest that short-term storage does not rely on a single structure or process. In fact, they assert that short-term storage is an integrated, multifaceted system, comprising various components 16 that execute varying tasks. It can then be assumed that different deficits within the short-term store emerge as a consequence of a deficit within a specific subcomponent of the short-term store (Matlin, 2001). The working memory model (Baddeley & Hitch, 1977; 1992) emphasizes the ways in which the memory system is adapted to meet the needs of real-life conscious mental activity. The basic premise of the model is that the short-term store comprises various sub-systems. In addition, the digit span does not reflect the short-term store capacity nor indicate that this store is a single structure (Baddeley & Hitch, 1977; 1992). Using these two assumptions as a point of departure, it was hypothesized that it would be difficult to retain a digit sequence while simultaneously performing a word-learning task that required the same short-term store capacity. Since the digit span reflects the maximum short-term storage capacity, a marked impairment of retention was expected since short-term storage capacity would be fully utilized. However, this did not occur. Thus the system responsible for memory span differs from the system supporting other conscious mental activity (Parkin, 1993). The task of retaining short digit sequences might be executed by a system dissimilar from the systems involved in reasoning and in learning word lists (Baddeley & Hitch, 1977; 1992). 17 The memory span for short words is higher than that for long words, hence the word length effect (Baddeley & Hitch, 1977; 1992). The word length effect led to the conclusion that the system underlying memory span is speech-based, and supported the notion of the articulatory loop (Parkin, 2001). In addition, the articulatory loop accounts for digit retention ability (Baddeley & Hitch, 1977; 1992). The concept draws on the idea that memory span is adversely affected if similar-sounding items are recalled, since the mechanism underlying memory span utilizes a sound-based code (Matlin, 2001). Articulatory suppression causes the word length effect to disappear, whereby short and long words are recalled to the same degree (Parkin, 1993). When this occurs, the memory for short and long words depends on the same processes, and is, therefore, recalled to the same degree (Parkin, 1993). 2.1.7 The Working Memory Model Baddeley (1977; 1992) identified three components of the working memory model viz. the phonological loop, visuo-spatial scratch pad, and central executive. The phonological loop and visuospatial scratch pad hold auditory, verbal, and visuo-spatial information within a limited attentional capacity. Specifically, the visuo-spatial scratch pad provides a workspace within which a visual image can be stored and manipulated in order to execute a task concerning a visual stimulus. The central executive organizes and decides what to do with this information. 18 Daneman and Carpenter (1980) [cited in Parkin, 1993] termed the fixed capacity of the central executive, working memory span. The function of the central executive is explained through the dysexecutive syndrome (Baddeley, 1990) [cited in Matlin, 2001]. This is an umbrella term of deficits resulting from frontal lobe damage that reflect impairment in the executive control of memory. The frontal lobes are critical for planning and executing response strategies, and Norman & Shallice (1991) [cited in Parkin, 1993] assert that schemata control most of these actions. This term refers to the established responses required for particular situations; once triggered, they allow a sequence of actions to be executed automatically. When additional conscious control is required, the supervisory attentional system is activated. This system is operative where planning, problem solving, and decision-making are required (Matlin, 2001). Recall is a reconstructive process, whereby a hypothesis is devised in order to initiate a memory search (Matlin, 2001). Recall can be likened to a form of problem solving which depends on the intact functioning of the supervisory attentional system. Recognition, however, sidesteps the problem-solving aspect described above. As a result, impairment of the supervisory attention system may be disrupted to a lesser extent in recognition (Matlin, 2001). 19 The dysexecutive syndrome may be characterized by a form of memory deficit resulting from damage to the supervisory attentional system (Neath & Suprenant, 2003). Thus, on an aspect of memory requiring a form of executive strategy, a deficit appears; when a memory task is not strategy-dependent, deficits are not evident. The pattern of memory loss may be attributed to the inability to plan or to execute retrieval processes (Matlin, 2001). Consequently, performance on recall tests is poorer since these tasks require the strongest executive demand. The articulatory loop comprises two systems, namely a phonological store, responsible for holding speech-based information, and an articulatory control process that recycles limited amounts of this information (Baddeley & Hitch, 1977; 1992) [cited in Matlin, 2001]. The distinction between the two processes is necessary since it accounts for the fact that, under articulatory suppression, individuals can still make judgements concerning the phonology of an item (Matlin, 2001). Under articulatory suppression, a distinction can be made between words and non-words that sound like words (Besner, 1981) [cited in Parkin, 1993]. Thus articulatory suppression prevents the transfer of information from the phonological store to the articulatory control process (Baddeley, 1992). Initially, Baddeley coined the term inner voice to describe the direct relationship between the articulatory control process 20 and the mechanism governing overt speech (Matlin, 2001). The articulatory control process is not linked to the mechanisms of overt speech but rather reflects a central form of motor programming that precedes speech production. When reading silently, Baddeley’s inner voice is operative. Huey (1908) commented that: “The carrying range of inner speech is considerably larger than vision…. The initial sub-vocalization seems to help hold the word in consciousness until enough others are given to combine with it in touching off the unitary utterance of a sentence which they form….”(cited in Parkin, 1993, p. 128). When vocalization is suppressed, recall is better for easy rather than for difficult stimuli (Hardyk & Petrinovich, 1970) [cited in Parkin, 1993]. This indicates that the articulatory loop executes a more critical function for complex reading stimuli. The articulatory loop provides a useful memory system for encoding sequences of phonemes in the correct order, and establishes a means of holding the intonation contour (Parkin, 1993). The intonation contour occurs when different aspects of phonology enable different meanings of the same sentence to be understood. Under articulatory suppression, visual memory declines when sentences are simultaneously presented. Levy (1977) [cited in Parkin, 1993] confirmed that articulatory coding is specifically involved. 21 2.1.8 The Multi-store Model of Memory William James (1890) termed the memory system supporting consciousness, primary memory and that which supports the permanent record of the past as secondary memory. This view was later refined as the multi-store model of memory (Atkinson & Shiffrin, 1968) [cited in Parkin, 1993]. Accordingly, memory is perceived as the flow of information between three inter-related stores (or repositories), namely the sensory store, the shortterm store and the long-term store. The sensory store receives new information, is transient, and contains information relating to the pattern of sensory stimulation. The sensory storage of visual information is known as iconic memory (Sperling, 1960) [cited in Parkin, 1993] and is implicated in the early stages of visual analysis. Owing to the assortment of sounds that are recognized, the aspect of memory that retains auditory stimuli over a short period of time was later identified (Matlin, 2001). Thus, the auditory equivalent to iconic memory is echoic memory (Parkin, 1993). Information in the sensory store is transferred into the short-term store (Matlin, 2001). Immediate memory is the intermediate memory store between sensory store and short-term store (Parkin, 1993). Short-term store activities are represented as various control processes, which respond to specific information requirements (Matlin, 2001). Furthermore, these 22 processes also determine the contents of the short-term store, in that the information being processed at a given time can be displaced by new information (relevant to the task being undertaken at that time). In addition, rehearsal describes the tendency to repeat information when trying to remember it, and is important for transferring information from the short-term store to the long-term store (Atkinson & Shiffrin, 1968). Words have three distinct encoding dimensions, namely orthographic, which refers to the pattern of letters comprising the word; phonological, referring to the sound of the word; and semantic, which refers to the meaning of the word (Baddeley & Hitch, 1949) [cited in Neath & Suprenant, 2003]. The phonological confusability effect describes the difficulty in recalling stimuli that are phonologically confusing (Conrad, 1964) [cited in Parkin, 1993]. This effect also occurs during the immediate recall of similar-sounding words compared to that of unrelated words (Baddeley, 1949) [cited in Parkin, 1993]. This effect, however, is not evident when semantically related words are used (Baddeley, 1949). In delayed recall, the phonological confusability is ineffectual, but semantically confusable words are recalled more poorly than unrelated words (Baddeley, 1949). These findings spearheaded the notion that the short-and-longterm stores could be distinguished in terms of encoding differences. The short-term store utilizes phonological encoding as well as semantic encoding. 23 2.1.9 Long-term Store 2.1.9.1 Permanent store model Tulving (1972) [cited in Parkin, 1993] refuted the shortterm/long-term dichotomy. Instead, he proposed a tripartite system, in which a distinction is drawn between memories associated with personal recollection from those that are not. Tulving (1972) noted that the long-term store consists of three functionally distinct yet interactive components, namely episodic, semantic, and procedural memory. Episodic memory is the memory that enables one to be aware of having experienced an event before. It refers to an individual’s autobiographical record of past experiences, and represents temporally dated episodes that can later be retrieved. Semantic memory refers to knowledge concerning facts, concepts, and vocabulary; the context for which is explicitly known, and available for retrieval. By contrast, procedural memory refers to the memory that is reserved for skills. Moreover, the terms episodic and semantic can be replaced by a single term namely, declarative memory (Tulving et al., 1973) [cited in Parkin, 1993]. This aspect of memory refers to any memory that is consciously accessed. Tulving later supplemented his definition of semantic memory to include the memory “that allows the individual to construct mental models of the world . . . . It makes possible 24 the cognitive representation of objects, situations, facts, and events” (Tulving, 1985) [cited in Parkin, 1993, p. 41]. The revision accounts for how one can possess a memory of an event without having specific information about why one knows that the event occurred in the past. Tulving (1985) termed this form of semantic memory experience noetic (knowing), in contrast with that of an episodic memory experience, termed autonoetic (selfknowing). Memories that are retrieved without conscious recollection are termed anoetic. A memory is derived from an experience or a learning event (Neath & Suprenant, 2002). Retaining an episodic memory does not require the learning event to be retained. In semantic memory, learning may only occur within the context of an episodic memory that confirms why the semantic stimulus means what it does (Tulving, 1985). Over time, the semantic stimulus becomes assimilated into semantic memory, and is represented independently of the events that gave rise to it (Parkin, 1993). Since the memory representation and learning event do have to be linked, a great saving in storage can be achieved (Parkin, 1993). The multi-store model is founded on the idea that the record of an experience resides separately from the knowledge gained from the experience. Santee (1981) [cited in Parkin, 1993] asserted that experimental evidence supporting this distinction is problematic, given the interdependence that exists between the 25 episodic and semantic memory systems. Accordingly, evidence from amnesic patients constitutes the primary support for the two separate systems. An amnesic patient is likely to exhibit intact semantic and procedural memory, impaired episodic memory, and normal conversational skills (Neath & Suprenant, 2002). In addition to the inability to learn new information, most amnesic patients suffer a parallel loss of memories acquired before the injury or illness (Lezak , 2004). This form of deficit, termed retrograde amnesia, is more pronounced for memories acquired close to the onset of amnesia. 2.1.9.2 Explicit and implicit memory Advancing an alternate theory, Schacter (1987) [cited in Parkin, 1993] proposed that the long-term store could be understood by explicating the ways in which it responds to explicit and implicit memory tests. Explicit memory refers to the memory for a learning event that requires conscious processing, and explicit reference to the event. A test of explicit memory directly assesses the recollection of a previous learning event (Lezak, 2004). Explicit memory can be tested using three different procedures (Parkin, 1993). A free recall test requires an item to be recalled without cuing. Cued recall tests require recalling the target information in the presence of a specific cue. In a recognition test, a stimulus is presented and a decision is made whether it is the target item or not. Recognition can be tested by eliciting either a “yes” or “no” response whereby each item is 26 judged individually, or by a forced choice procedure whereby one item from an array must be selected as the target. Thus, the cued recall test is a form of a recognition testing. Implicit memory refers to the memory for a learning event without making specific reference to that event. The memory tested under implicit conditions forms automatically. The idea that a memory from an experience can be expressed indirectly is not a new concept. Descartes (1649) noted that frightening childhood experiences could remain imprinted on a child’s memory “without any memory remaining afterward” (cited in Parkin, 1993, p. 49). Leibniz (1916) stressed the role of unconscious memories in everyday life, and argued that one could have “remaining effects of former expressions without remembering them” (cited in Parkin, 1993, p. 49). Implicit memory tasks thus test memory indirectly. 2.1.9.3 Priming in long-term memory Priming refers to the learning that facilitates performance through prior exposure to words and other material (Parkin, 1993). Cues that elicit priming do not necessarily have to share any physical properties with the target item. Priming effects are conceptually, perceptually, and semantically driven (Squire, 1987) [cited in Neath & Suprenant, 2002]. 27 Procedural memory is spared in amnesia since the acquired information is embedded in procedures (Lezak, 2004). An alternate account is that priming reflects impairment in the way that preexisting cognitive operations are executed (Crowder, 1985) [cited in Neath & Suprenant, 2002]. Skills and priming are expressed through a memory system that prevents explicit access to the contents of the knowledge base (Parkin, 1993). Memory, is thus expressed only in performance, and prevents an individual’s accumulative experience from being reflected in verbal or nonverbal tests that require judgment or familiarity (Lezak, 2004). In amnesia, declarative memory describes the learning and memory that is impaired in amnesia as this information is explicit (Lezak, 2004). Thus, the recall and recognition impairment, and the preservation of skill learning and priming in amnesiacs, support the hypothesis of two distinct memory systems (Parkin, 1993). 2.1.10 Forgetting Storage failure refers to the inability of the memory system to produce a permanent memory trace of a given event. Storage failure may occur either because the transfer from the short-term store to the long-term store was not initiated or because the permanent memory trace was lost. By contrast, retrieval failure arises from the inability of the memory system to locate an existing memory trace. 28 2.1.10.1 Memory strength theory According to the model, a memory representation is a record of a stored word in memory. A memory representation incorporates information about the context within which the word was learnt, other test items, and the target item itself. Retrieval, thus, depends on context cues, association cues, and item cues. Moreover, retrieval improves with increased presentation time, since the strength of the association and item cues are increased. An increase in the retention interval decreases retrieval because the context cues have changed greatly during the interval. Each learning event possesses a signal strength. A range of factors determines strength values, and includes: rehearsal frequency, processing depth, target familiarity, and the similarity between the context cues during encoding and retrieval. The model proposes that frequently rehearsed items possess greater item strength than items rehearsed less often and are, therefore, more likely to be retrieved. Within working memory, information is activated by the interaction of a cue in working memory with the associated information in long-term memory. Recognition and recall differ in the manner in which the target item in long-term memory is identified. During recognition, the target item and the context form a compound that 29 elicits a representation in memory. The familiarity of the representation is computed to critical value. If familiarity with the item is higher than the critical value, the item is recognized as old, otherwise, it is recognized as new. During recall, the context initiates the retrieval process by triggering search cycles until a word is recalled. A starter word is selected, followed by a selection of additional words, and a decision is made whether the selected word fits the correct context. The most active word in the retrieval structure is recalled first, which tends to be the most recent item (Mixted, Ghadisha & Vora, 1997) [cited in Neath & Suprenant, 2002]. However, if activation is relatively low, retrieval cannot occur and the search stops after several failed attempts. 2.1.10.2 The feature model The feature model conceptualizes immediate memory in terms of performance on immediate serial recall tasks. The order of information is encoded multi-dimensionally and stored with cues in primary memory (Neath, 1999) [cited in Neath & Suprenant, 2002]. Each item is encoded in a manner that allows the items to possess a constant position along its relevant cue dimension and cannot occupy a new position upon encoding a new item (Estes, 1972; 1997) [cited in Neath & Suprenant, 2002]. Recall begins by determining which cue most likely occupies the first position along the relevant cue dimension; this is 30 necessary for sampling cues from secondary memory. A sampled item must be recovered prior to output, and recall begins with the cue in primary memory that has the highest probability of being the cue for the second item. In addition, the similarity between primary memory and secondary memory cues increase the probability of sampling. Incorrect recall results from errors in the sampling procedure where the correct cue for the target item is chosen but subsequent sampling does not occur in chronological order. 2.1.10.3 Theory of distributed associative memory Murdock’s theory of distributed associative memory (1982; 1995) [cited in Neath & Suprenant, 2002] explains the memory for serial list recall and recognition. The model proposes that individual memories are distributed within one memory system. Furthermore, the memory system stores two kinds of information, namely item information that enables recall, and associative information between the items that allow them to be recalled in the correct order. Serial recall is driven by cues, the first of which is the context cue. The instruction to recall the target item is the cue that initiates retrieval. Accordingly, recall occurs through the process of correlation termed chaining, and continues until an item cannot be recalled. When a link in the chain is missing, recall stops. By contrast, recognition occurs through familiarity. If the item is recognized, it is perceived as old; 31 if it is not recognized it is perceived as new. Furthermore, a memory comparison stage occurs whereby the presented item enters a decision cycle, and a decision is made whether the item is old or new. The strength of the model is that it includes both item and associative information. When an item cannot be recalled, the associative information cues for the next item, explaining how recall still proceeds in spite of failing to recall other items on the list. One shortcoming however, is that recognition is also explained through the concept of a competitor set. This concept includes a set of items that an individual thinks is likely to have occurred on the trial. The target item is compared to the item in the competitor set, and whichever item is most similar is perceived as the target item. If recognition is successful (that is if the match is correct), that item in the competitor set is removed. This occurs since individuals rarely repeat an item. The retrieved item is then convolved back into the memory system, explaining how an individual remembers which item has been recalled. The problem is that by the time the last item is ready to be recalled, it is the only item in the competitor set and, therefore, will always be matched with the last item since there is no other possibility. 32 2.1.10.4 Generation-recognition models Generation-recognition models of memory (Watkins & Gardiner, 1979) [cited in Neath & Suprenant, 2002] conceptualize retrieval as a reconstructive, search-guided process. At the outset of retrieval, the memory system generates possible candidates as the target stimulus. These candidates are then subjected to a recognition process, which if successful, results in that candidate being retrieved as a memory. Each known word is represented by a node. When a word is presented, a tag or occurrence marker is established, indicating that the word is part of the target list. At recall, various candidates are generated, and the corresponding nodes are examined for tags, which if detected, result in recognition and, hence, retrieval. During recognition, access to the node occurs automatically, and recognition is dependent on the detection of a marker. High-frequency words are better recalled than low-frequency words since they are more likely to be generated as candidates for recognition. The poor recognition of high-frequency words arises because these words are associated with several markers, making it more difficult to determine whether the target word was presented in a specific list. 33 2.1.10.5 Encoding specificity principle According to the generation-recognition models, recall involves successful generation and recognition, whereas recognition involves successful recognition only. Given this, an item that is recalled must be capable of being recognized since recognition is involved in recall. However, the failure to recognize recallable words does occur; this is termed recognition failure (Tulving et al., 1972) [cited in Neath & Suprenant, 2002]. The encoding specificity principle (Tulving & Thompson, 1973) [cited in Neath & Suprenant, 2002] explains how this phenomenon occurs. Retrieval critically depends on the degree of overlap between the features encoded in the memory trace and those in the retrieval environment. Recall and recognition are different manifestations of a single retrieval system, and are independent of one another since they possess different featureoverlap with the encoded target. A strong associative cue possesses a high degree of featural overlap. Recognition provides the most featural overlap with a target trace, resulting in recognition superiority over recall. However, the memory system can be manipulated such that cued recall for an item can be better than recognition for the same item. 34