Investigating the use of metaphors for knowledge acquisition.
A major phase in the construction of Knowledge-Based Systems (KBSs) is the
Knowledge Acquisition stage. It involves acquiring knowledge from experts,
books, examples and from other relevant sources. Until recently, researchers
assumed a simplistic view of knowledge which required it being 'extracted' or
'mined' from experts. It is now clear that this simplistic view ignored the rich
epistemological, cognitive and perceptual basis of what we refer to as knowledge.
This thesis set about to address this problem. It identifies domain analysis
as an early stage in the knowledge acquisition process and argues that it should
capture the expert's perception of the domain, rather than the knowledge engineer's
view of the expert's perception. It puts forward the thesis that metaphors
are invaluable cognitive devices for perceiving and articulating domains. This idea
has been postulated by several other researchers but never, until now, tested.
This thesis prescribes a novel method for exploiting metaphors for knowledge
acquisition. The method is based on Black's interaction view of metaphors
which has been merged with Kelly's personal construct psychology. We have
implemented a tool, DAKUM, based on this new method and evaluated the usefulness
of metaphors for knowledge acquisition.
We conclude that metaphors are useful for constructing a domain's functions
and structure. However, we note that metaphors are only useful, when
describing Juzzy domains; their use in describing relatively organised domains
could often lead to confusion.
Metaphorically speaking, we conclude that using metaphors is analogous to
using a walking stick. The latter is only useful when it is needed; otherwise its
use is often a hindrance.