dc.description.abstract |
It has been reported widely that changing the environmental context between learning and the reinstatement of a memory has a deleterious effect on measures of recall (see Smith, 1988; Roediger & Guynn, 1996, for reviews). Measures of recognition memory have also been reported to be affected, but, as made clear by Smith, positive findings have been fickle and infrequent. Over the years, a range of hypotheses have been generated in an attempt to understand why the reinstatement or removal of the learning context does not easily affect recognition memory. Part 1 of this thesis, examines two of these hypotheses, Murnane and Phelps's (1993) d' hypothesis and Murnane and Phelps's (1994) new/different context hypothesis. The experimental procedure used throughout this thesis was based on the continuous recognition design (Shepard & Teghtsoonian, 1961). Experiment 1 and 2 tested recognition memory for words while manipulating aspects of the environmental context (background color, font and word location) and lag (number of words between learning and testing). These experiments demonstrated a strong context-dependent decrement in recognition memory with changes to both a new context, and to a different but equally experienced context. This context-dependent effect was shown with measures of d as well as with the false alarm and hit rate. Both the different context manipulation and the use of the measure d, were predicted by Murnane and Phelps two hypotheses to lead to, at the most, a limit context-dependent effect. This did not appear to be the case. Experiments 1 and 2 showed a robust reliable context dependent effect with both the new and different context manipulation. The context-dependent effect was shown to increase with lag. At the shortest lag (1 intervening item) there was no effect of the context change with the different context condition, while the new context condition showed an unexpected reliable improvement in performance. The problems with these hypotheses, led to a careful examination of four of the most actively discussed formal models of memory [Gillund & Shiffrin's (1984) SAM; Hintzman's (1984, 1988) MINERVA2; Humphreys, Brain & Pike's (1989) Matrix Model; and Murdock's (1982,1983) TODAM] as well as two of the more recently developed related models [Shiffrin & Steyvers (1997) REM; Murdock's (1997) TODAM 2]. Part 2 of this thesis examines the ability of these models to describe recognition memory performance with changes in environmental context, and with changes in number of items in memory. Each model was described and its proposed method (or a possible method) of processing contextual information was discussed. With the exception of Murdock's (1997) TODAM 2, a numerical example of each model's method was given, and a simulation of recognition memory, based on the procedure used in Experiment 1 and 2, was presented. All models examined were able to predict a decrement in memory performance with a change to both a different and a new context. None of the models was able to predict an initial improvement in performance followed by a later decrement in performance with a change to a different context, although one model was able to do this with a change to a new context (TODAM). The method by which the TODAM model was able to do this was termed the "opposing effect hypothesis". It suggests that there are two independent sources of information in the recognition decision - one fast decaying source (item-specific information) that is aided by the context change, and one slow decaying source (associative information) that is impaired by the context change. The combination of functions such as these would explain the data obtained in Experiment 1 and 2 very well. Part 3 of the thesis is in two sections. The first examines the performance patterns of individual participants; are the patterns of performance produced by individuals consistent with the opposing effect hypothesis (Experiment 3). In Experiment 3, two individuals produced recognition memory decay functions that showed a stable context-dependent effect size (parallel decay functions), while another two individuals showed an initial improvement in performance with a later decrement in performance as lag increased (crossed functions). These performance patterns could be expected according to the opposing effect hypothesis. The second section of Part 3 examines the effect of increasing the attention given to associative stimuli (Experiment 4), relative to that in Experiment 1. According to the opposing effect hypothesis this should reduce the interaction between lag and the size of the context-dependent effect (i.e., to produce more parallel functions), and increase the level of same context performance. The manipulation did lead to a large reduction in this interaction, but the same context performance was not reduced. This provided mixed support for the hypothesis. A process dissociation style reanalysis of Experiment 4, was carried out in Part 4 of the thesis. It revealed that the associative information used for recall led to a context-dependent decrement in performance. In contrast, the recognition component of Experiment 4 produced both an improvement (at the shortest lags) and a decrement in performance (largest lags). This is also consistent with the opposing effect hlpothesis. While there appears to be evidence for context dependent recognition leading to both an improvement and a decrement in performance, and some support has be found for two opposing context-dependent effects, further work is clearly required to understand the nature of the processes involved. |
en |