There's a standing joke in our family, or rather between my sister and I. It usually comes up at family get-togethers, about the time that we're all trying to decide what to do, where to go, or what to eat. Mom will say something like, "let's have that spaghetti casserole with turnip greens that I made at last summer's get-together, we all enjoyed it so much." Jackie and I will exchange a look that says, "gross, we hated that," and invariably Mom, seeing the look pass between us, will say in utter seriousness, "no, no. I DISTINCTLY REMEMBER. We all loved that casserole."

What continually amuses both Jackie and I, is the degree of confidence with which Mom remembers the same event we do, but so differently. How can people be so certain and yet so mistaken about events in our own histories?

And yet, it happens time and again, and not just in my family. The observation that episodic memory retrieval is vulnerable to distortion has been documented thoroughly. Even the pattern of errors can be predicted with some reliability. (1)

In order to situate "episodic memory," it's useful to know several distinctions in memory research. The most basic one is between brief and enduring memories, called short-term memories (STM) (or working memory, WM), and long-term memory (LTM). Within LTM, there are qualitative distinctions, such as between explicit and implicit, and declarative and procedural -- both of these distinctions have to do with consciousness about the memory. Explicit/declarative memory encompasses facts, figures, and all of conscious memory. Its what we commonly refer to as memory. (2) This type of memory is flexible, fast, and specialized for one-time learning. (3) Procedural/implicit memory is thought to be the most durable memory, and encompasses learned habits, skills and things that you "know" but don't consciously think about. (2)

Declarative memory is further divided into episodic and semantic. Semantic memory refers to the knowledge system, and encompasses memory for facts and bits of disembodied information, whereas episodic is event memory, "memory for context embedded events of one's own past," (4) such as Mom's dinner memory.

The following discussion of memory pertains mainly to one kind of long-term, explicit memory - namely, episodic memory. Assuming that both Mom and I were exposed to and attended to roughly the same incoming information during dinner last summer, is there something about either the memory system or memories themselves that could explain why people's autobiographic recall is not reliable over time?

How we account for memory errors probably has a lot to do with how we envision memory and its processes. Early models, like Penfield's (5) or Atkinson-Shiffron's, (6) likened memory processes to a filing cabinet system, or a tape recorder. Incoming bits of information were thought to be processed through the mental machinery on a sort of conveyor belt: from the sensory register, through the short term store, until they reached their final destination where they were filed away in tact, in some long term storage box, remaining there until needed.

Although these models could account for some types of errors, such as forgetting -- in terms of losing links to information (misfiling), or information getting erased over time (oops, did that end up in the trash can), it could not account for the distortions that episodic memories seemed to go through. Mutations couldn't be just a filing error.

Such distortions suggested that either something was going on inside the storage box, or that retrieval didn't happen the way we envisioned it, or perhaps more radical still, that "a memory" wasn't the entity we envisioned it to be. More recent researchers, like Elizabeth Loftus who studies errors in episodic memory, have suggested that "a memory" of a specific event is not an entity at all. Rather, bits and pieces, component parts, are diversely stored. In this paradigm, every time a recall is initiated, memory representations are constructed anew out of the bits stored throughout the system. In this sense, no entity ever exists in long-term memory, a representation exists only briefly as it is assembled in short term memory. (1)

Markowitsch says, "contrary to some traditional views, there is little doubt that memory is represented in many widely distributed areas of the brain. (4) However the debate is far from over. "It is still unknown whether singular neuronal modules or widespread neuronal networks represent particular pieces of information." (4) In tandem with mounting research on the brain's functional localization, there is also mounting evidence for the notion for the network model of memory.

For the sake of argument, let's just assume the network model is accurate on some level. If so, is there any evidence about memories, how they are made, how they are stored, and how we get at them that would support or inform the network model of memory, and its purported "reconstructive errors"?

Perhaps a good place to start would be to build a basic anatomical model that would include all the parts necessary to support the network theory. If memories are constructed out of bits held diversely in a long-term store, then the brain should come with at least few pieces of equipment: a place where the information is processed, another place where the bits are stored, and a place/mechanism for reassembling the pieces. It would be nice if the reassembling plant were also the same plant that did the DIS-assembly, or at least had instructions from the dispersers.

Kornhuber's biological model does support the idea of functionally and anatomically separate places for storage and encoding. He proposes that after information comes in through the sensory system, it is initially processed in cortical regions (STM), (2) goes to the limbic system, where it interacts with emotional and motivational input for further processing, and is eventually transferred back to various parts of the cortex for long term storage (LTM). (4) Kornhuber's model is supported by a variety of evidence. His model does not account for the recall piece.

Fernandex localizes the LTM memory in the cortex, and the processing in the hippocampus and. Further, he shows that this processing takes time. (7) Time is an important factor in any information-processing model of memory, because it implies that the information does something--maybe gets broken down into the pieces.

Another piece of evidence for different processes localized to corresponding brain structures comes from the study of amnesiacs. A lesioned hippocampus renders a patient unable to lay down new explicit long-term memories, while leaving her old memories in tact, (2) suggesting that the processing of memory occurs in a different place than the storage of memory.

There is some evidence that both encoding and retrieval of declarative memories occur in the medial temporal region of the hippocampus. Does that mean the dispersers and re-assemblers are the same mechanism? Apparently not, as they occur in different parts of the hippocampus, further supporting the idea that they represent different mechanisms. Using fMRI on human subjects, Gabrieli and associates showed that encoding declarative memories activated the posterior medial temporal region while retrieval activated the anterior medial temporal region of the hippocampus. (8) Additionally, Tulving and associates found differential hemisphere activation for encoding and retrieval of episodic information. (4)

The two site/two mechanism evidence for encoding and retrieval might suggest some reasons why episodic memories are subject to change: its sort of like the left hand trying to gather up what the right dispersed. There are bound to be errors.

However, that can't be the whole story, because memory errors increase over time. It is possible that later incoming bits information simply interfere with or clutter the network so that the retrieval compiling mechanism can't find all the pieces to the earlier memory. But that is not really a satisfying explanation, given that incoming memory bits are plunged into a sea of other memory bits every time one is encoded. In that case, why would time make a difference at all-- reconstructed memories should be muddled from the start.

Is there another way to account for increasing errors over time? As it turns out, there is some evidence that at first, a cortical memory is tied to its limbic (hippocampal) processing system, but over time, it consolidates and becomes independent of it. (3) Perhaps the processing system has some sort of master plan or blueprint for where the bits are stored. And maybe as LTM becomes increasingly disjointed from the processing system, it becomes more vulnerable to reconstruction, in that the links to the encoding blueprint are broken.

Perhaps on a cellular level, the trend towards separation of LTM from hippocampal encoding processes can be explained. Long term memory is thought involve lasting modifications to synaptic connectivity in groups of neurons, (3), (9) also called long term potentiation (LTP). These potentiation modifications are accomplished either by changes either in the amount of neurotransmitter released, the number of post-synaptic receptors available to receive it, and/or the rate of recycling it. (5)

While the widely accepted view is that memories consist of "the change in the strength of already existing neural connections and formations of new connections," (3) others have suggested that change in LTP is not sufficient to account for memory. (2) Some have further suggested that the limbic system triggers LTP, which activate consolidation process in the cortex storage areas. These models propose that memory itself is not the change in LTP per se, but resides in "synaptic enlargement, dendritic spine growth, expansion of neuropil and the like." (4)

To relate this back to separation between LTM and encoding, it has been suggested that the limbic (hippocampus) system drives memory consolidation in the cortex by activating LTP changes in the cortex. At some point those modifications stabilize. And maybe as they do, the effect of the limbic system diminishes, and drops out of the picture. When that happens, perhaps any direct link between the original limbic blueprint telling where all the bits are dispersed is cut off from the bits themselves, and then when the memory is up for recall, the limbic system, no longer connected to its bits in the cortex, does not know which parts to pull into the construction, and it ends up making errors.

Mounting evidence suggests that STM, LTM, encoding and retrieval are anatomically and functionally distinct, and these separations may help to account for why we see errors in episodic memory over time. What was not clear from the literature is exactly how the system distributes the bits of information. Does it scatter the bits randomly, does it have some high level organization that makes having a re-constructive blueprint largely unnecessary? If there is not some highly organized system for storing the bits, what it is that pulls the memory back together again? My guess is that if there is some "re-constructive blueprint" for episodic memory, it exists somewhere in the hippocampus. As of yet, we have no word on its existence.

But more puzzling than the question of what operates to disperse and reassemble the information, is a more basic question -- why? Why might the memory be system built this way? What possible advantage is there for taking incoming information and scattering bits of it in widely distributed networks, only to reassemble it later (and possibly without the benefit of some guiding blueprint)? Under such conditions, its not only easy to see why memory is open to distortion, it begs the question, how do we ever come up with an ACCURATE memory. One explanation is that perhaps it's adaptive in other ways. Maybe having information widely distributed in bits makes networking to other memories more accessible, which in turn could facilitate the creation of adaptive knowledge networks with more speed and efficiency.

For instance, perhaps Jonny approaches an unfamiliar dog for the first time and the dog snaps at him. And another time, Jonny approaches a friend's dog that is well trained, and behaves compliantly. Jonny may well remember these events in his episodic memory store, which is a fine thing. However, even if the specific episodes are eventually come apart over time, maybe the various information bits (wild dog behavior, house pet behavior) from those episodes combine with other similar information bits, forming "knowledge" about doginess, Jonny may well be able to use that conglomerate of information the next time he sees a dog. He may, without knowing, be able to efficiently evaluate whether or not to approach the dog, based on some (wild dog/ house pet) schema made out of gathered, but no longer contextualized, bits of information.

Perhaps the punch line in all of this that maybe it matters less that we accurately remember whether we liked Mom's cooking last summer or not, than that the experience was processed and combined with other bits of information, guiding our future decisions making about trying out creative kitchen creations.

WWW Sources

2) The Medial Temporal Lobe, in the Washington University School of Medicine Neuroscience Tutorial.

3) Human Memory , in Neuropsychology

4) The Anatomical Bases of Memory , Chapter 54, by Hans J. Markowitsch

5) Conversations with Neil's Brain , by William H. Calvin

6) Galotti, K M. Cognitive Psychology in and out of the Laboratory, 2nd Ed. Wadsworth Publishing Co., 1999, pg 128. The Atkinson-Shiffron Memory Model

7) Real-Time Tracking of Memory Formation in the Human Rhinal Cortex and Hippocampus , by Guillén Fernández et al, in Science 1999 September 3; 285: 1582-1585

8) Separate Neural Bases of Two Fundamental Memory Processes in the Human Medial Temporal Lobe , by John D. E. Gabrieli et al, in Science 1997 April 11; 276: 264-266

9) Localization of a Short-Term Memory in Drosophila , by T. Zars et al, in Science 2000 April 28; 288: 672-675.

| Course Home Page | Forum | Brain and Behavior | Serendip Home |