Long-Term Working Memory

What makes this theory different?

Unlike other theories of expert processing, long-term working memory (LT-WM) is able to account for two main phenomena in the expertise literature. Firstly, expert memory for domain specific material seems to be quite robust. If given an interference task, in which after the task (e.g. memorizing a chess position), an intervening task or simply a long delay is imposed prior to recall, experts seem to show only minimal degradation (6-8%) in what was to be recalled (Charness, 1976). Secondly, experts are able to memorize information faster than “chunking” theory allows and larger than the standard STM capacity of 7 or so chunks (Ericsson & Kintsch, 1995).

Is this "skilled memory"?

Long-term working memory is built on the skilled memory literature (which looked mainly at digit span) and their conclusions which are that a) elaborate cues or associations (relations) to prior knowledge are used to encode information, b) skill level and practice determines how long it takes for information to be encoded or retrieved, and finally c) that experts make use of retrieval structures (Chase and Ericsson, 1982).

What are retrieval structures?

LTWM Encoding Model
Long-term Working Memory Methods of Encoding

According the LT-WM theory, information can be encoded either by using retrieval structures (for information that is rapidly presented or by elaboration of patterns and schemas which take longer. Both access information which has been encoded in LTM.
(Gobet, 1998)

Retrieval structures are essentially mechanisms that help to facilitate retrieval of information that is stored in long-term memory. General encoding schemes are used to make elaborate cues and associations for a particular memory to increase likelihood of recall (e.g. method of loci). Furthermore, retrieval structures need to be activated prior to use and is limited to only including one type of material in a rapid presentation task. Chase and Ericsson (1982) described a case study of subject SF whose digit span was able to increase from the standard 7 +/- 2 (Miller, 1956) to around 80 digits. This feat was accomplished through a learned strategy which SF employed when he realized that running times could help facilitate memory of “random” digits. (He was a runner and thus familiar with running times.) This hierarchal organization was a type of retrieval structure.

Looking at the figure to the right, you can see the two encoding methods employed by LT-WM. One way is through the retrieval structure discussed earlier which is used during shorter presentation times (when schemas cannot be sufficiently elaborated) and the second way is through knowledge-based, elaborated structures (schemas and patterns) that is formed through associating a stimulus with another stimulus or with the context under which it is being encoded. Which encoding method that is used is determined by the domain task that is being done. Information is essentially coded into long-term memory accessible through retrieval cues in short term memory (STM)

Defining LT-WM

So from skill memory theory, LT-WM arose, described as a

"cognitive processes occur as a sequence of stable states representing end products of processing, and that acquired memory skills allow these end products to be stored in LTM" (Ericsson & Kintsch, 1995)

Ericsson and Kintsch were able to apply this theory across numerous domains including medical diagnosis and mental calculation. Therefore unlike the other theories, LT-WM’s greatest strength is that it is domain general although, as Gobet (1998) argues, makes it slightly vague and under specified.

More on LT-WM:

Valid XHTML 1.0! Valid CSS! Level Double-A conformance icon, 
          W3C-WAI Web Content Accessibility Guidelines 1.0 Bobby 508 Compliant! © Copyright Victor Long, Chandra Singh, and David Snitkof 2005
All Rights Reserved