In 1968 Atkinson and Shiffrin proposed a model of human memory which posited two distinct memory stores: short-term memory, and long-term memory. Later a third memory store (actually the first in sequence) was added: sensory memory. In this paper I describe this augmented three-store model and indicate some of the characteristics of each memory system.
Information enters the human information processing system via a variety of channels associated with the different senses. Perceptual systems operate on this information to create perceptions. But because of a limited processing ability at the higher levels, most incoming information cannot be immediately dealt with. Instead, we attend only to certain information. However, information not immediately attended to is held briefly in a very temporary "buffer" memory, making it possible to attend to some of it a bit later -- as when you can still hear someone asking you a question even though you weren't really listening when they asked it. This buffer memory is called sensory memory.
Sensory memory is really many sensory memory systems, one associated with each sense. For example, there is a sensory memory for vision, called iconic memory, and one for audition (hearing), called echoic memory. Here are some characteristics of these two sensory memory systems:
- Iconic Memory (vision)
- Capacity: Essentially that of the visual system (Sperling)
- Duration: About 0.5 to 1.0 seconds (Sperling)
- Processing: None additional beyond raw perceptual processing
- Echoic Memory (hearing}
- Capacity: ????
- Duration: About 4 to 5 seconds
- Processing: None additional beyond raw perceptual processing
Short-term Memory (STM) or "Working Memory"
Information that is attended to arrives in another temporary store called short-term or working memory. The more recent term "working memory" is intended to convey the idea that information here is available for further processing. In general information in working memory is information you are conscious of and can work with. Here are some properties of STM:
- Capacity: About 7 plus or minus 2 "chunks" of information (Miller, 1956)
- Duration: About 18 to 20 seconds (Peterson & Peterson, 1959)
- Processing: To hold information in STM, it is often encoded verbally, although other strategies may also be used such as visualization. These strategies make it possible to "rehearse" the information.
The low capacity of STM was first noted by George Miller in a famous paper intitled The Magical Number Seven, Plus or Minus Two. Miller concluded that about seven (plus or minus two) "chunks" of information could reside in STM simultaneously. Miller defined a "chunk" as an independent item of information -- one whose recall did not aid in the further recall of the other items. Random letters such as "GJK" would each be considered a chunk, but letters that form a recognizable larger whole, such as "CAR" would not. (In this case the word "car" is a single chunk.)
Information that enters STM fades away, or decays as soon as it is no longer attended to. (The duration of 18-20 seconds assumes that the information is not being actively rehearsed.) Information that is being actively attended to is represented by a pattern of neural activity in the brain may become represented more permanently by guiding changes in neural connectivity in the brain, a process referred to as storage. But information that is not more permanently stored is simply lost shortly after attention is directed elsewhere.
Because STM presents severe limits on the amount of information that can be held in mind simultaneously and on the duration for which it lasts once attention is withdrawn from it, STM has been described as the bottleneck of the human information processing system.
Long-term Memory (LTM)
Long-term memory is the relatively permanent memory store in which you hold information even when you are no longer attending to it. Information held in LTM is not represented as patterns of neural activity (as in STM), but rather as changes in brain wiring -- in the "conductivity" of existing synapses, and in the formation new synapses and destruction of old ones. Storing information in LTM is equivalent to a computer writing information out to its hard drive, or to a tape recorder writing patterns of magnetization onto tape to record music. The recording process is called storage and the "playback" process, retrieval.
Here are some properties of LTM:
Capacity is unlimited in the sense that nobody seems to run out of the capacity to store new information, even if they live beyond 100 years. If they did, then either they would stop learning entirely or new learning could only take place by first erasing something already stored in LTM. This does not appear to happen -- when storage/retrieval capability is lost it is due to deterioration of brain systems rather than to systems exceeding their holding capacity.
- Capacity: Virtually unlimited
- Duration: Up to a lifetime
- Processing: Information is organized according to meaning and is associatively linked
It is difficult to determine how long memories can exist in LTM. If you cannot remember something you once knew, is it because it has been lost from the system, or because you have developed a problem locating it for retieval? Permanent losses do occur as a result of brain damage, and it is possible that some memories simply decay away if they are not accessed for a very long time. We do know with certainty, however, that some failures of retrieval are due to temporary blockages and not to the loss of the information in memory. You may be unable to remember someone's name at present, for example, but later it comes to you. Obviously, it was there in memory all the while.
A common idea is that everything we have ever experienced has created a long-term memory, but this is unlikely to be so. Much of what we experience is never attended to, or not attended to beyond a few brief moments, and probably does not result in activation of the storage process.
Evidence that STM and LTM are Physically Distinct Systems
I have asserted that memories in STM consist of patterns of neural firing that are being sustained in the brain, and that memories in LTM consist of altered patterns of brain wiring. The latter changes supposedly make it possible for certain inputs to regenerate something like the original patterns of neural firing that were present when the long-term memories were being laid down. One especially convincing piece of evidence for this distinction is found in the case of "H. M.," a victim of severe epilepsy who received experimental surgery in an attempt to control the symptoms. The surgery involved destruction of the hippocampus in both cerebral hemispheres. Following the surgery, H. M. could no longer store new information in LTM (with certain exceptions to be discussed elsewhere), although he could still retrieve information that had been stored there in the years before the surgery. His STM was unaffected. He had become like a tape recorder that could still play back old tapes but could no longer record new ones. The fact that his LTM was severely affected without damaging his STM shows that these are two functionally distinct systems.
There is also evidence that LTM involves a rather slow process requiring many hours to complete, called consolidation. If an individual is rendered unconscious (as in an accident), memories just beginning to consolidate at that time may not form. Of course, STM involves no such consolidation process.