Reading:
Johnson, L. and Johnson, N.E., 1987, Knowledge Elicitation Involving Teachback Interviewing in Kidd, A.L., (Ed.), 1987, Knowledge Acquisition for Expert Systems: A Practical Handbook Fay, B, 1996 "Competence" (from Contemporary Philosophy fo Social Science) |
This week we will consider how to acquire knowledge using mediating representations. Specifically using Systemic Grammar Networks and Teachback Interviewing. Unlike the repertory grid technique, this is non-psychological and does not presuppose a theory of mind, but has a 'Socratic' theory of knowledge based on a three-way dialogue between the investigator, subject and the representations themselves.
The technique is based on the iteration and refinement of representations until both parties agree that they faithfully describe the domain under analysis. Elicitation is therefore understood to go through mediating and intermediate representational stages.
Any representational scheme must contain enought inherent structure to adequately model the phenomena, and have built-in constraints to exclude absurdities. The following properties are desirable:
Any representational scheme must also be transparent to others. The following properties are desirable:
Examples: Concept Maps, Systemic Grammar Networks, Repertory Grids, Structured English, Decision and Rule Trees.
The intermediate representation would usually contain an integrated view of different knowledge types: concepts and concept definitions, causal relationships, functional dependencies, causal models, heuristics, lists of previous cases, etc.
The key point is that intermediate representations are the union or integration of the various mediating representations. The overall structure of the domain being modeled will be apparent from the intermediate representation, but not from the mediating representations which are more territorial/local.
See also KADS, the Knowledge Level and Ontologies.
Systemic Grammar Networks (SGNs) were, as the name suggests, developed in linguistics to represent the functional structure of language. They have a basic network structure of links and nodes which lend themselves to representing qualitative data emanating from interviews and transcripts. The use of SGNs in knowledge acquisition is described by Johnson (1985) and Johnson and Johnson (1987), who describe it as a technique directed at `capturing the expert's conceptual structure, not just his procedural skills'.
The objective is to construct a competence or knowledge-level model, as a machine independent, declarative intermediate representation
The practical use of SGNs is best considered as a stylized interview procedure, termed teachback interviewing by Johnson and Johnson (1987). As mediating representations SGNs function as a conversational tool that acts as a vehicle to convey meaning, both within the analytic encounter, and beyond, as a documentary device.
The term mediating representation conveys the sense of synthesis and coming to understand through a representation which is constitutive of knowledge analysis, and this is the key to the use of the SGN
The figure below shows the basic operators followed by a simple example.
The basic idea is that the SGN should be constructed as swiftly as possible by the investigator from interview material or other relevant texts. It should then be refined, firstly against the data and then by the subject or expert critiquing or modifying it.
The trick is to get started quickly by picking out key terms from the text (head terms in the anthropologists' parlance), and use these as nodes by trying to graphically model them using more abstract terms.
SGNs are properly considered as a conversational technology, serving both the representational requirements of knowledge acquisition, and the wider analytic imperative of arriving at an understanding. Knowledge acquisition therefore is completed only in part from the representational use of the SGN, further information comes from explanations, justifications, excuses and so on. This is the teachback phase.
The basic teachback model has the expert teaching something to the elicitor who, in turn, explains the same thing right back to the expert. The procedure continues until the expert is satisfied with the elicitor's explanation.
The expert can exert some control over their contribution by checking their interpretation of the material they have provided. This can increase the experts' sense of control and ownership of the system they are helping to develop, and maintain their confidence in the analysis.
Teachback is based on Pask's conversation theory, which aims at making ideas and knowledge objective and public.
The basic model of conversation theory
The idea of a competence theory comes from linguistics and is aimed at discovering an underlying abstract structure, grammar or competence from which all actual performances of that langauge, expertise, behaviour or whatever, can be derived.
It assumes expert problem-solving to be model-based, at least with some degree of abstraction, but distinguishes performance (the outward manifestation of the model), from competence (the essence of the model), to arrive at an understanding of knowledge acquisition akin to reverse engineering. Because theunderlying (competence) model must be abstracted from a series of performances, and analysis must proceed with a series of interactions in which the analyst plays a crucial intellective role.
