Knowledge Structures and Theories of Memory

Advocates of hypertext suggest that the ideas relevant to a subject can be represented best by a simple associative network, or web, with multiple inter-relationships specified. Electronic storage and manipulation makes it feasible to have a very large amount of material included and the hypertext format 'frees' the content from being structured with a specific ordering. According to Beeman et al. (1987),




"a hypertext should function like an adventure game, permitting you to wander through a world of facts and ideas, jumping from one to another or linking them when you wish to do so."

This vision transfers the responsibility for structuring the information away from the author and vests it jointly, if not equally, with the reader.

Streitz (1990) advocates flexible hypertext formats over conventional paper based formats for representing knowledge. He suggests that hypertexts can actually represent the author's 'internal knowledge structure' which can then be directly apprehended by the reader without the need for an intermediate stage, i.e., the traditional paper document with its apparently linear structure. The idea is also proposed by Baird and Percival (1989) who claim that




"information can be put into the hypertext in the same structure as it is in the author's head."




For Jonassen (1990) hypertext is eminently appropriate because "hypertext mimics the associative networks of human memory", a notion derived directly from Vannevar Bush's seminal paper in 1945 and subsequently repeated (often uncritically) by many commentators. Bush recognised the constraints of traditional alphabetic indexing and claimed that:




"The human mind does not work that way. It operates by association. With one item in its grasp, it snaps instantly to the next that it is suggested by the association of thoughts, in accordance with some intricate web of trails carried by the cells of the brain."





Thus, a model of the way the brain stores information is proposed as an appropriate model for structuring and presenting information. While Bush accepted that an artificial duplication of this mental process was unrealistic, he nevertheless suggested that it should be possible to at least emulate it. As we mentioned in Chapter 1, Bush proposed an information storage and retrieval system, called the memex, which would support 'associative indexing',





"the basic idea of which is a provision whereby any item may be caused at will to select immediately and automatically another."




The perspective outlined by Bush is congruent with contemporary theories of memory which have arisen with the emergence of the computer in the 1960s, and this results in a curious paradox. Those commentators who suggest that associative networks are an appropriate model for the structure of electronic hypertexts are in fact alluding to a psychological perspective - man as an information processing system - which was itself strongly influenced by models of computer architecture. However, even a brief review of the concept of association within human memory indicates that the parallel is facile.


  [Top]

Associationism

Associationism has been a cornerstone of our view of memory and conceptual understanding for a very long time. It was a key element for the British empirical philosophers of the 16th century (Locke, Hume, Berkeley, James Mill.; and John Stuart Mill.S.;) and was also central to much of the work of the learning theorists who dominated experimental psychology in the first half of this century. The central concept is very simple: when two elements are experienced as occurring together in time (contiguously) they will become associated in memory. Recent models of semantic memory have had a strong associative component and have even been described as being "neo-associationist" (Anderson and Bower, 1973). In these models, long-term memory is depicted as a 'network' of countless associations between verbal and conceptual entities, and memory retrieval becomes a search through this maze of pathways.

However, such theories differ radically from the original associationist views in terms of complexity. A variety of new features have been incorporated and these include logical relations between elements, hierarchical structures, and a new stress on context and strategy (see, for example, Collins and Quillian, 1969; Anderson and Bower, 1973; Collins and Loftus, 1975; Rumelhart and Ortony, 1977; Norman, Gentner and Stevens, 1976).

Of primary importance has been the incorporation of subordinate and superordinate relations between semantic categories (Collins and Quillian, 1969), since this led to the inclusion of logical inference in models of memory. Figure 11 depicts a simple but plausible semantic network centred on the concept bitter in which length of line is negatively correlated with strength of association, i.e., the shorter the line, the stronger the association. A number of limitations are clearly illustrated by even a cursory examination.



i.	The associations indicated are arbitrary, and many other linkages between the terms represented could easily be justified
ii.	The concepts are at different levels of generality, e.g., the inclusion of the concepts aperitif, wine and sherry
iii.	The concept depicted is relatively simple and easily defined but many others are much more ambiguous and likely to entail almost unlimited association, at which point graphical representation acts to reduce clarity rather than aid understanding.
iv.	Finally, while there may be areas of significant agreement between individuals, ultimately semantic nets are unique to the individual.

Figure 11: A simple associative semantic net for the concept bitter. [After Collins and Loftus, 1975]

  [Top]

A more sophisticated attempt at representing human memory is shown in Figure 12 where, although concepts are still linked by association, the linkages comprise specific types. The graphic representation for even this simple factual episode is reasonably complex and it seems doubtful whether the approach is suitable for representing anything more involved. The Thoth-II system (Collier, 1987) attempts to do this by employing conceptual objects and typed relations, but the author accepts that there are substantial disadvantages to the approach, particularly with regard to user navigation. It is unclear at present how this structuring can be presented to the user in a way that is meaningful, let alone helpful.







Figure 12: Semantic network for a simple historical episode. [After Rumelhart, Lindsay and Norman, 1972]

  [Top]

 

Our theoretical understanding has come a long way from the simple notions of the Associationists, but even the relatively sophisticated contemporary models of semantic memory and conceptual understanding are acknowledged to be crude representations. The claim that a simple, non-hierarchical associative net, or web, is an ideal or natural model for hypertext because it mimics human memory must be seen as inadequate.