Methodology
The first activity required is knowledge acquisition. Different authors present methodologies with varying stages of knowledge acquisition, but fundamentally they all involve: the identification and conceptualisation of requirements and problem characteristics, formalising these into some mediating representation scheme, implementation, and final testing and validation (Graham & Barrett, 1997). Knowledge acquisition can be machine-aided or human-labour oriented.
Johnson and Johnson’s methodology (1987), enhanced by Graham (1990), proposes a three-stage knowledge acquisition process based around semi-structured interviews.
The first phase is to perform a broad, but shallow survey of the domain. This allows the elicitor to become oriented with the domain, so that a more flexible approach can later be taken. This type of horizon broadening is a standard approach in social science research. Once this shallow trawl of the domain has been done, the second phase requires that a more detailed task analysis is performed by the elicitor, focussing on the area of interest. The structure of the interview uses a teachback technique to traverse the domain and validate elicitor understanding with the result that the elicitor progressively refines the model of the expert’s competence. This model is qualitatively drawn up and uses a mediating representation, Systemic Grammar Networks (SGNs) (Bliss, Monk, & Ogborn, 1983). These are a context free, qualitative representation, which can be used as a tool for the systems design, but their use does not imply the final application of any particular knowledge engineering software or methodology. SGNs have been used in many domains including oncology, printed circuit board (PCB) design, and fault diagnosis. The third phase of this approach is to validate the models drawn up from the expert with the wider expert community. The theoretical predictions of the model are presented to the initial community used in the first phase, and then to a further independent population, to check the appropriateness and validity of the model which has been created.
This knowledge acquisition methodology was adopted and tailored to the needs of the project. The first phase, the Broad and Shallow Survey, was achieved by arranging a telephone survey in conjunction the Royal National Institute for the Blind (RNIB) in London, and with a visually impaired student at the University of Ulster, to complete questionnaires specifically tailored to suit visually impaired interviewees. The second phase, a more detailed task analysis, was achieved through the design of semi-structured interviews with the visually impaired student expert at Ulster. Knowledge synthesis and analysis of survey and interview findings led to design criteria rather than the employment of SGNs which were not considered practical for visually impaired experts. Validation (and verification) was achieved by the evaluation of implemented criteria in exemplars at Ulster and York, for teaching computer science and electronics respectively.