«A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy (Psychology) in the University of Michigan ...»
AVOIDING THE RECENT PAST:
WHICH STIMULUS DIMENSIONS INFLUENCE PROACTIVE INTERFERENCE?
Kimberly S. Craig
A dissertation submitted in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
in the University of Michigan
Associate Professor Cindy A. Lustig, Chair
Professor John Jonides Professor John E. Laird Professor Patricia A. Reuter-Lorenz © Kimberly S. Craig All Rights Reserved 2013 Dedication For my family.
I would not have made it this far without your inspiration, encouragement, and support.
ii Acknowledgements First, I would like to thank my advisor, Dr. Cindy Lustig, for her advice and encouragement at all stages of this project. In many ways this project would not have been possible without her. I would also like to thank Dr. Marc Berman and Dr. John Jonides for their input on Chapter 2 of this work. They provided an additional insight to this project that shaped this work in important ways.
My committee, as well, has provided extremely valuable feedback throughout the dissertation process; Dr. John Laird, Dr. Cindy Lustig, Dr. John Jonides, and Dr. Patricia Reuter- Lorenz have all contributed their perspectives and advice, helping me to focus my ideas and analyses to better answer empirical questions about the nature of proactive interference. Dr.
David Meyer has also provided a large amount of support and encouragement throughout the dissertation process.
In large part, this research would not have been possible without the many research assistants who have helped with experimental design, data collection, and data analysis over the years. I would like to thank Laura Argintar, Jennifer Chronis, Halley Feldman, Elyssa Guslits, Yona Isaacs, Emily Kurzawa, Karin Lavie, Xu Li, Lina Lu, Matthew Lugo, Jessica Oakley, Eric Peist, Saqib Rana, Page Sloan, Alexander Tereshchenko, Maria Tocco, Dani Weinberg, and Madison Weisberg for all of their help.
Finally my family, who support me unconditionally and always go above and beyond to encourage me in whatever I choose to pursue. You have given me the confidence and passion for iii learning that led me to graduate school and will continue to lead me in a positive direction in years to come.
iv Table of Contents Dedication........
Table 2. Statistical analyses and effect sizes for Experiment 1 and Experiment 2.
......... 28 Table 3. Average median response times (standard error) and accuracies (standard error) for high and low performing groups in Experiment 2, as determined by a comparison
Table 10. Statistical analyses and effect sizes for Experiment 5.
........................55 Table 11. Average median response times (standard error) in milliseconds, and accuracy scores
Table 15. Statistical analyses and effect sizes for the 1-back task.
.......................95 Table 16. Response times (msec) and accuracies (% correct) for negative trials of high- and
Table 20. Statistical analyses and effect sizes for IES in Chapter 3.
.................... 126 Table 21. Average IES by trial type for Chapter 4; color instruction and number instruction
Figure 6. Average median response times (top panels) as well as accuracies (bottom panels) for negative responses in Experiment 4(left; panels a and b) and Experiment 5 (right;
Figure 11. Average median response times (a) and accuracy (b) for negative trials.
......... 94 Figure 12. Inverse efficiency scores for negative trials in Chapter 2, Experiments 1-5.......121
Why Do We Forget?
Imagine that you are waiting tables at a busy restaurant. One of your customers orders the mozzarella sticks but, after a few seconds, decides to get the nacho plate instead. When you return to the kitchen you see mozzarella sticks waiting and begin to pick them up, but then realize your error, set them down, and wait for the nacho plate. The temporary confusion you would experience in this situation is an example of proactive interference: information from the past disrupting current memory performance. This type of interference is a major mechanism of forgetting, both in short-term memory (STM) and long-term memory (LTM) (Underwood, 1957).
Interference is a robust, reliable phenomenon in STM tasks (e.g., Berman, Jonides & Lewis, 2009; Mecklinger, Weber, Gunter & Engle, 2003) but also somewhat mysterious. How can we forget information that we have seen only moments ago? Can we ever escape our recent past, or are we always vulnerable to its influence? The following work will focus on proactive interference within STM, investigating the conditions that may be necessary to create this type of interference. Implications for current theories of proactive interference and current models of STM will be discussed.
Theories of Proactive Interference Two competing theories currently dominate the proactive interference debate. The first suggests that information in STM decays over time (Altmann & Gray, 2002; Brown, 1958). This
gives the item high activation strength, and over time this activation strength decreases.
Interference is created when (now-irrelevant) recent items with relatively high activation strengths compete with target items for attention and current information processing. As time passes and activation strength decreases, the item loses its ability to interfere.
The activation-strength idea is simple and intuitive: The more recent an item is, the stronger its residual activation strength, and thus the more likely it is to cause interference. A
recent paper by Atkins et al. (2011) appeared to find compelling evidence for this idea:
Participants were slower to reject recently-seen information (an index of proactive interference) even when that information belonged to a completely separate category than the currentlyrelevant information (e.g., being slow to reject ‘England’ when the relevant information only consists of fruit names). From the activation-strength perspective, seeing the recent but now incorrect item (‘England’) automatically re-activates its representation and pulls it back into the focus of attention, creating interference. It is clear that according to activation strength theory, proactive interference should be pervasive, occurring whenever a recently-presented item has the chance to compete for the focus of attention.
A second theory argues that it is a high degree of similarity between recent and current items that creates proactive interference (e.g., Crowder, 1976; Johnson, 1933; Keppel, 1968;
McGeogh & McDonald, 1931; Underwood, 1945, 1957). From this perspective, a recent item that is highly dissimilar to the target of current information processing should not interfere.
Support for this theory comes from the release-from-proactive-interference effect, where performance declines in a list-learning paradigm over time for similar items (e.g., from the same category), but then is eliminated when a word appears that is highly dissimilar to others on the
strength theories. Here, an item interferes because it is similar, not because it is recent.
Both the activation-strength and similarity-based competition views are appealing, but both may also be too simple to fully explain interference in STM. However, their strengths and weaknesses are complementary and suggest that both must be taken into account. Returning to our restaurant example, if recent activation was the primary driver of interference, how could a server learn and hold in STM the orders from multiple people at the same table without suffering from catastrophic interference? On the other hand, recency must play some role: The server deals with multiple iterations of a closed set of similar items (i.e., the menu) throughout the day, and it seems unlikely that one patron’s lunchtime order of nachos will interfere with the memory for a dinner patron’s appetizer. To bring these views together, we consider the structure of STM and the nature of its representations.
The Structure of STM There are many models of STM’s structure, but they tend to fall into two classes that differ in whether they treat STM and LTM as distinct, separate entities or as more of a continuum. The most prominent example of the first class is Baddeley’s working memory model (Baddeley& Hitch, 1974; Baddeley, 2000). In the canonical version, STM is comprised of two temporary storage components (the phonological loop and visuospatial sketchpad), and a central executive component that acts as an attentional controller. A more recent modification (Baddeley,
2002) adds an episodic buffer that binds and stores multimodal information. This model has been extremely influential in past decades, both in behavioral (i.e., Burgess & Hitch, 1999) and neuroimaging research (i.e., Paulesu, Frith & Frackowiak, 1993; Wagner, Shannon, Kahn &
called into question (for a review, see Lustig et al, 2009).
More recently, unitary models of memory have gained popularity (e.g., Cowan, 2001;
Oberauer, 2002). In these models, STM is conceptualized as an activated portion of LTM, rather than a separate component. Typically, these models consist of a focus of attention and a region of direct access. Item(s) currently being attended to have the highest activations and can be found in the focus. These items may move into the focus either from the inactive portion of LTM or via a perceptual process, thus allowing new items to enter STM. When an item moves from the focus it enters the region of direct access, a privileged portion of LTM that contains previously or potentially relevant items. Items in the region of direct access retain a relatively high degree of activation, and are easily moved back into the focus.
To fully understand proactive interference, it is important to consider how interference may be created within these models. For instance, in unitary models, items in the region of direct access may retain the residual activation strength important for creating proactive interference for recently-seen items (according to an activation strength account). As noted above, however, the activation strength account may be too simple to fully explain proactive interference; instead, concepts from both activation strength and similarity-based competition accounts may be important to consider in order to thoroughly understand this phenomenon.
A more detailed consideration of what it means to be an “item” in STM may help resolve the previously-described tension between activation-strength and similarity-based competition views of interference. According to feature-based theories (e.g., Nairne, 2002; Oberauer & Lange, 2008), STM does not hold whole items but instead retains activated cues or feature codes which can be used to recall the associated item. In other words, rather than discrete “items”, the
pointers to the codes (e.g., perceptual, lexical, phonologic, semantic, temporal) for such features in LTM. When two items have a high degree of similarity, they share many of these features or cues, leading to decreased distinctiveness.
Importantly, an item’s temporal context can be considered another featural dimension along which it may be more or less similar to other items (see Howard & Kahana, 2002). At first glance, temporal information is seen as distinct and separate from other stimulus dimensions, such as size, shape, or color. This might occur because temporal information about a stimulus is not a part of the item itself, but is instead about the context in which that item was seen. Hasher and Zacks (1979) argue that although dimensions such as temporal and spatial information are based on context, items are always experienced contextually, and contextual information about an item is processed automatically. This automatic processing may allow contextual stimulus dimensions to be treated in the same way as information about an item’s size, shape, or color.
Futher, recent work by Howard and Kahana (2002) suggests that temporal information may be coded neurally based on oscillatory frequencies, allowing for comparisons along the temporal dimension in addition to similarities along other stimulus dimensions. If temporal information is indeed simply another featural dimension, feature-based views could be interpreted by positing that recently-presented items cause interference not because of residual “activation strength” in an energetic sense, but rather because they are similar to (and thus compete with) target items along the temporal dimension of similarity.
The experiments in this dissertation test a refinement of similarity theory: Items will only compete and cause interference if their similarity is along a dimension relevant to the test cue.
Lustig and Hasher (2001) proposed that this task-relevance might explain why implicit (indirect)
Specifically, the dimension of temporal similarity is almost by definition irrelevant to implicit memory tests, but is integral to explicit tests. Within the context of STM, this hypothesis suggests that proactive interference based on recent presentation depends upon whether temporal information is relevant to the task being performed; if not, recent presentation should not create interference.
During the performance of any given task, comparisons of similarity should be made between items only along dimensions of similarity that are task-relevant. When items are similar along these task-relevant dimensions, interference will occur. However, if items are similar along task-irrelevant dimensions, this will not create interference because these dimensions will not be compared.
The present experiments test this hypothesis in part by manipulating the task relevance of several dimensions of similarity and examining the effects of these manipulations on interference in STM.