Sensory, working and long-term memory

In 1960 George Sperling flashed 12 letters for a split second. People could name only about four, yet a tone played after the letters vanished showed that about nine were still there.

Someone reads you a phone number, you repeat it under your breath on the way to the keypad, and a minute after the call it is gone. Your own birthday, on the other hand, has stayed put for decades. Psychologists explain the difference by treating memory as a set of stages that hold information for very different lengths of time. Every stage has three jobs: encoding (getting information in), storage (keeping it) and retrieval (getting it back out).

Sensory memory: a picture that fades in under a second

Richard Atkinson and Richard Shiffrin proposed in 1968 that information first lands in sensory memory, a very brief store for sights, sounds and other raw input. Sperling had already measured the visual version. When he showed a grid of 12 letters for a fraction of a second and asked people to write down all of them (whole report), they averaged slightly over four. Then he changed the task. A high, middle or low tone sounded just after the letters disappeared, and people reported only the row the tone pointed to. They got 76 percent of the cued letters right. Because nobody knew which row would be called, the whole grid must have been available for a moment, about 9.1 of 12 letters. Delay the tone by a fraction of a second and that advantage faded. The image was there; it simply vanished faster than anyone could read it out.

Working memory: a small desk

What you attend to moves into short-term memory. OpenStax gives its duration as about 15 to 30 seconds without rehearsal. In 1959 Lloyd and Margaret Peterson asked people to remember three-letter strings such as CLS while counting backwards, which stopped them from repeating the letters. About 80 percent were recalled after 3 seconds and only about 10 percent after 18 seconds. The Petersons blamed decay, but Keppel and Underwood (1962) looked at the very first trials and found almost no forgetting there. Letters from earlier trials were getting in the way, an effect called proactive interference. Both decay and interference matter.

How much fits on the desk? George Miller reviewed the evidence in 1956 and called seven, plus or minus two, a magic number. Later work, summarised by Nelson Cowan, puts the limit nearer four items, plus or minus one, once people cannot group items into chunks. Alan Baddeley and Graham Hitch (1974) added that short-term memory is a workspace with parts: a phonological loop for sounds and words, a visuospatial sketchpad for images and layout, and a central executive that directs attention. Baddeley later added an episodic buffer that links them with long-term memory. That is why it is called working memory.

Long-term memory: more than one kind

Long-term memory has no known capacity limit. It splits into explicit memory, which you can put into words, and implicit memory, which shows up in what you do. Endel Tulving (1972) divided explicit memory into episodic memory (your last birthday, with a where and a when) and semantic memory (the capital of a country). Implicit memory includes procedural skills such as riding a bicycle. The patient known as H. M. showed that these are separate. After surgery removed the hippocampus on both sides to control his seizures, he could not form new episodic or semantic memories, yet he got faster at a puzzle each day he practised it, without remembering having seen it before.

The cards above use the same boxes as every card set on this site. A card you mark as known moves up a box and comes back later; a card you miss drops to box 0. The spaced repetition unit explains why that schedule works.