Email Calvin || Email Bickerton || Book's Table of Contents || Calvin Home Page || September 1999


William H. Calvin and Derek Bickerton, Lingua ex Machina: Reconciling Darwin and Chomsky with the human brain (MIT Press, 2000), chapter 6.  See also

copyright ©2000 by William H. Calvin and Derek Bickerton

The handheld version of the book is available from or direct from MIT Press.

Webbed Reprint Collection
This 'tree' is really a pyramidal neuron of cerebral cortex.  The axon exiting at bottom goes long distances, eventually splitting up into 10,000 small branchlets to make synapses with other brain cells.
William H. Calvin

University of Washington
Seattle WA 98195-1800 USA


How Are Memories Stored?

Computers have three levels of memory, as I was breathlessly explaining this morning on the pre-breakfast hike around Bellagio (hiking uphill enforces brevity B every writer should try this as a warmup). First, there is the type-ahead keyboard buffer, that stores your keystrokes if you=re a faster typist than I am.

Eventually the letters are fetched and moved into RAM, the fast-access temporary memory that the programs utilize. It has one disadvantage (besides cost), which is the fact that everything in it is lost when the power is turned off. It=s sometimes called volatile memory because the data Aevaporates.

‘We hear and apprehend only what we already half know.'
Henry David Thoreau

From RAM, you can move the string of letters to the hard disk, where they survive even if the power fails. All three types of computer memory have the bad habit of filling up, mostly because they use pigeonholes, a physical slot for each item of information, each with its own address that provides the means of finding what=s been stored.

Superficially, human memory looks as if it has similar functional subdivisions. There=s an immediate memory, a sensory buffer (cells tend to fire for a short while, even after a sensory stimulus is over). There is a volatile short-term memory. This is sometimes called Aworking memory,@ as it=s what you need to hang onto a phone number long enough to dial it; it=s also what you need to repeat back a sentence, as when swearing an oath. It involves lingering traces that assist in recreating the firing patterns anytime in the next few minutes.

Finally there is a process (called Aconsolidation@ and requiring days to accomplish) by which some short-term memory items are made into more durable long-term memories, ones that can survive the disruptive events (concussions, coma, seizures) that would fog or erase short-term memories that hadn=t been consolidated yet.

But long-term memory doesn=t fill up, nor does it seem to have an addressing scheme that we can discover, probably because it doesn=t use pigeonholes. When you learn someone=s name for the first time, you don=t store it in an empty slot, as in a hard disk. You appear to store it redundantly in a number of places, overlapping it with all the previous memories stored in those places. It=s a distributed memory system. Think of the storage method as being like your favorite washboarded road, the one that tries to shake your car apart at a certain speed B but one that also has resonances for trucks overlain on the washboarding that affects cars.

In the first half of the twentieth century, such facts about memory were established and, in 1949, Donald Hebb created our modern formulation of the relation between short- and long-term memories.

Impressed by the fragile nature of short-term memories, and the length of the consolidation period before some became more permanent long-term memories, he said that concepts were implemented by a characteristic firing pattern in a small group of cortical neurons, which he dubbed a cell-assembly. To recall someone=s name, you needed to recreate that musical spatiotemporal firing pattern. But, because of the way that a long-term memory survived episodes such as coma and seizures, it had to be a spatial-only pattern like those washboarding ruts, something that didn=t require ongoing neural activity such as impulses firing away.

It=s all very much like the dominant technological metaphor of Hebb=s time, the phonograph. The long-lasting grooves in the record, when a needle was dragged through them, served to recreate a spatiotemporal pattern called music or speech. And the spatial-only patterns in the grooves had been created by a spatiotemporal pattern during recording, the same one (within limits) as was later recreated.

The brain=s storage method is not two-dimensional, like the record groove. Three dimensions are available, and so is redundancy (as we currently see for computer communications protocols, such as error-correcting codes). On the other hand, there are potential confusions, such as Arecording over@ previous material B and yet we somehow retrieve the desired item amidst the mix. A modern formulation tends to use terms such as Achaotic attractor@ rather than resonance, helping to emphasize the way variant patterns are conformed to a standard. I tend, following the lead of Hebb=s contemporary J. W. S. Pringle, to emphasize the role of a plainchant chorus, utilizing those redundant Agrooves,@ helping to produce a standard version from all the potential variability caused by overwritings.

Confusing things further is that different areas of the brain are important for short-term memory than for long-term memory. It=s pretty clear that neocortical areas are where most of the language-accessible long-term memories are kept, but they=ll never be consolidated there unless the hippocampus is functioning well during the short-term memory period following input. People with damage to the hippocampus (and adjacent cortical areas in the midline face of the temporal lobes, common in Alzheimer=s dementia) may be able to retrieve old long-term memories but they aren=t forming new ones very well because consolidation doesn=t work right. Yesterday may be lost to them, even though their youth is not.

This is a problem totally unlike the Aemeritus professor problem@ of knowing so much that it takes a long time to sort through it, and thus complaining about Amemory problems.@ Emeritus professors (actually, the problems start in the forties for many people, professorial or not) nearly always come up with the right name eventually, proving that it was there all along. This suggests the problem is simply one of lengthening access times, where you cannot keep up with the windows of opportunity afforded by social repartee.

Brain imaging techniques utilizing regional blood flow changes are capable of seeing what areas are working harder during certain memory tasks. When a subject is given a working memory task, one analogous to the telephone operator giving you a new number that you have to remember long enough to dial, the frontal lobe areas just in front of the motor strip seem to work harder, as do the areas in the back end of the sylvian fissure. Both Broca=s area and Wernicke=s area, in the old formulation of language areas, are involved in working memory of this type.

Stimulation mapping of the exposed cortical surface of epileptic patients undergoing neurosurgery reveals an even more detailed picture of short-term memory. The patient watches a series of slides, a new one popping up every six seconds. The first one reads, AThis is a@ and then shows a drawing of a common object. So the patient says, AThis is an apple.@ The second slide will be a distraction of some sort, such as showing a two-digit number from which the patient counts backwards by threes. Then the third slide comes up, merely saying, ARECALL.@ The patient is supposed to say, AApple@ (or whatever the earlier object was). During some of the slides, stimulation occurs as the neurosurgeon moves around, testing different cortical sites, checking accuracy of retrieval. Stimulation of some sites in the temporal lobe during the first or second slide causes errors during recall attempts (a period when there isn=t stimulation), even when they didn=t interfere with naming or with the distraction task. Frontal lobe stimulation sites have effects mostly when applied during the retrieval attempt itself.

The effective sites are collectively known as Ashort-term post-distractional memory sites@ because of the suggestion that in this test, the first slide=s name was either stored there or that the site had important connections to the retrieval process, allowing the electrical buzz to interfere with the recall attempt. These sites were usually located somewhat outside the perisylvian core of sequencing sites, forming something of a periphery around it. The sites affecting reading (Agrammar sites@) were often in between the sequencing core and the memory periphery.

There is a great deal of individual variation in the brain=s language organization, some of it correlated with verbal IQ. Most dramatically, language mapping varies by sex, the male brain having many more naming sites at the back end of the sylvian fissure, the female more naming sites in the frontal lobe. The female arrangement seems more resistant to aphasia by strokes; four out of five aphasics are men. Even with the same amount of cortical damage from the stroke, the woman has much less functional impairment. As age-related mortality has long suggested, the female body plan seems to be the more secure one, less liable to serious trouble.

None of this explains how the neurons accomplish these functions B or how they differ in different areas B but I hope that the foregoing explains why brain researchers expect to find the mind in the brain. I attempted a short course on cortical neurons and their synapses in the sixth chapter of Conversations with Neil's Brain. Then, in the first few chapters of The Cerebral Code, I addressed spatiotemporal patterning much more explicitly and geared up to tackle the problem of a Darwinian process (anything that mimics the full-scale copying competition that bootstraps quality, I call a ADarwin Machine@). A Darwin Machine in the cerebral cortex might operate on the time scale of thought and action, exactly what language needs. Much of the two chapters that follow are my attempt to briefly describe the basic principles and how they apply to convergent and divergent thinking.

Back in 1996, as I finished correcting the page proof of Cerebral Code, I realized that the long-distance common code problem (the spatiotemporal firing pattern characterizing a phrase such as AThe cat on the mat@ temporarily needs to be the same in the temporal lobe as it is in the frontal lobe), which I had found a technical solution for, also provided a powerful mechanism to facilitate the nested embedding of structured language. At the last minute, I added a few extra pages commenting on the subject in the final chapter of Code, but here I can be much more explicit about embedding and on-the-fly associations (though at the expense of being rather too brief on the underlying cellular neurophysiology). Together with the related Darwinian process that helps shape new ideas in the brain, it gives a glimpse of how higher intellectual functions might arise from lowly cells and circuits.


Notes and References for this chapter

Copyright ©2000 by William H. Calvin and Derek Bickerton

The handheld version of the book is available from or direct from MIT Press.

 Email Calvin  

 Email Bickerton  

  Book's Table of Contents  

  Calvin Home Page