Modeling the age-related associative deficit in self-initiated recall

We use Howard & Kahana’s (2002) Temporal Context Model (TCM) to help explain the well-known age-related deficit in episodic recall. TCM is a distributed memory model that postulates two basic forms of information: item and context. Throughout both study and recall, TCM associates the presently active item with a distributed representation of context that evolves over time. The model postulates that retrieval of contextual states that were associated with the currently active item adds to the current state of context. Using this retrieved context as a cue activates nearby items, producing the well-documented forward-biased associative tendencies seen in recall. Likewise, using time-of-test context as a cue activates recently experienced items, producing the recency effect in recall. Fitting TCM to data on episodic recall by young and older adults, we show that associative deficits in older adults can be understood in terms of a failure to inhibit noise associated with contextual retrieval. Not all forms of memory are equally impaired over the normal course of human aging. Whereas older adults invariably show significant impairment on episodic memory tasks, such as recognition or recall, their performance on implicit memory tasks, such as perceptual priming, is only slightly impaired (La Voie & Light, 1992; Verhaeghen, Marcoen, & Goossens, 1993). Among episodic memory tasks, older adults have the greatest difficulty with recall tasks, especially those that provide minimal contextual support, such as associative recall of unrelated words (NavehBenjamin, 2000), or free recall of randomly arranged word lists (Ceci & Tabor, 1981; Craik, Byrd, & Swanson, 1987; Laurence, 1967; Kahana, Howard, Zaromb, & Wingfield, 2002; Kahana & Wingfield, 2000). This paper uses mathematical modeling techniques to help clarify the age-related deficit in episodic recall. We focus on the age-related deficit in free recall due to recent advances in our understanding of the component processes underlying performance of this task (Kahana, 1996; Howard This research was funded by National Institutes of Health grants AG15852 and MH55687. We are also grateful for support from the W. M. Keck foundation. Correspondence concerning this article should be addressed to Michael J. Kahana, Volen National Center for Complex Systems, MS 013, Brandeis University, Waltham, MA 02454-9110. Marc Howard is now at Boston University, Department of Psychology, 64 Cummington Street, Boston, MA 02215-2407. Electronic mail may be sent to kahana@brandeis.edu. TEMPORAL CONTEXT MODEL 2 & Kahana, 1999, 2001; Kahana et al., 2002). Section 1 of this paper describes how the retrieval dynamics in free recall, as measured by output order effects, can be used to separate associative and recency-sensitive retrieval processes. Section 2 describes the Temporal Context Model (TCM) of free recall, and shows how it accounts for recency and contiguity effects in free recall. Section 3 demonstrates that TCM can account for the intact recency effects and impaired contiguity effects observed in free recall with older participants. Section 4 discusses the implications of this modeling work for theories of cognitive aging. 1. Recency, association, and the decomposition of free recall in normal aging In the free recall task, participants are asked to recall as many items as they can from a recently presented list, in any order. Going beyond a simple measure of recall accuracy, one can plot the probability of recalling items as a function of their serial position within the list. The characteristic feature of the serial position curve, which has fueled much of the theory on retrieval from episodic memory, is the presence of a small primacy and a large recency effect. The primacy effect, a response preference for the first few items of a list, is much smaller than the recency effect and depends critically on rehearsal (Brodie & Murdock, 1977; Rundus, 1971; Tan & Ward, 2000). The recency effect was, until recently, seen as evidence for the operations of a phonological buffer, capable of rehearsing and maintaining a small number of items in the absence of interfering activity (Atkinson & Shiffrin, 1968; Raaijmakers & Shiffrin, 1981). However, the finding that the recency effect persists over long time scales (Bjork & Whitten, 1974; Glenberg et al., 1980; Howard & Kahana, 1999) has led a growing number of researchers to ascribe the recency effect to positional distinctiveness (Nairne, Neath, Serra, & Byun, 1997; Neath & Crowder, 1990) or temporal context (Howard & Kahana, 2002). The serial position curve, which has played a key role in almost all theories of human memory function (Murdock, 1960; Anderson, 1976; Metcalfe & Murdock, 1981; Atkinson & Shiffrin, 1968; Raaijmakers & Shiffrin, 1980, 1981; Anderson, Bothell, Lebiere, & Matessa, 1998), reflects the end product of a complex and dynamic process. A weakness in traditional serial position analyses is in the way they have focused on the probability of recall as a function of presentation order while disregarding the order of recall. As such, they have discarded information about sequential dependencies in retrieval. Such information, we argue, is crucial for understanding the process of self-initiated memory retrieval. Figure 1a shows serial position curves for young and older participants performing an immediate free recall task (as reported in Experiment 1 of Kahana et al., 2002). Figure 1d shows serial position curves for delayed free recall (Experiment 2 of Kahana et al., 2002). The age-related deficit in free recall is generally seen at all serial positions, though it becomes somewhat less pronounced towards the end of the serial position curve (Parkinson, Lindholm, & Inman, 1982; Foos, Sabol, Corral, & Mobley, 1987; Rissenberg & Glanzer, 1987; Poitrenaud, Moy, Girousse, Wolmark, & Piette, 1989; Capitani, Della Sala, Logie, & Spinnler, 1992; Kahana et al., 2002). As expected, the recency effect seen in immediate free recall is largely attenuated by a 16 s arithmetic distractor task given to participants at the end of the list in the delayed experiment. Howard and Kahana (1999) proposed a decomposition of free recall performance into two measures: 1) a measure of how participants initiate recall, and 2) a measure of the way participants make transitions in recall from one item to the next. Combined with a stopping rule, these two measures summarize the dynamics of retrieval, and can predict the form of the serial position curve. TEMPORAL CONTEXT MODEL 3 2 4 6 8 10 Serial Position 0.2 0.4 0.6 0.8 1 P ro ba bi lit y of R ec al l a.