Civil Engineering Reference
In-Depth Information
TABLE 5.1 Evaluation of Cognitive Models (Cells of Matrix Refer to Requirement Fulfillment Level, the Goodness-
of-Fit is the Weighted Sum of Fulfillment Levels)
Model
Requirement
COSIMO
GOMS
SRK
UCT
ACT-R
Priority
/
Complex tasks
2
2
1
2
3
1
Resource
3
2
3
1
1
2
Different time bases
1
2
1
1
3
1
Multiple modalities
3
1
3
2
3
1
Learning
2
2
1
1
1
2
Human error
2
3
1
3
3
1
Time consumption
1
1
2
1
1
1
Stress
/
Strain
3
3
1
3
1
3
Semiotics
3
3
1
3
3
1
Software support
2
1
3
2
2
3
Spreading
2
1
2
2
3
2
Goodness-of-fit
0.53
0.61
0.73
0.62
0.30
second step, the corresponding weights of the requirements were defined with the help of an exponential
priority scale (first priority has twice the weight of second priority, second priority has twice the weight of
third priority). The goodness-of-fit for each model was calculated as the sum of the weighted fulfillment
levels (Table 5.1).
With regard to human information processing in complex task situations, the SRK approach rep-
resents the construct of mental model in terms of a detailed means-ends abstraction hierarchy (Rasmussen,
1985) and, therefore, fully fulfills the first requirement. However, COSIMO, UCT, and GOMS define
different levels of cognitive abstraction and, therefore, partially fulfill this requirement. ACT-R uses the
action-rules monolithically and, therefore, does not fulfill this requirement. Concerning the allocation of
mental resources, the UCT, as well as ACT-R provide features to activate and schedule cognitive rules in
parallel and, therefore, cope well with limited cognitive resources. Also GOMS-extensions, for example,
Gray et al. (1993), can model parallel execution threads, but only with a restricted horizon of resource
allocation. The other models offer limited features only and, therefore, do not fulfill this requirement.
Multiple time bases concerning task coupling are represented well in SRK, COSIMO, and UCT, which
rely on a multi-layered architecture of cognition. GOMS has restricted abilities when considering differ-
ent partial models (like unit task and keystroke level). Multiple sensory modalities can be modeled with
GOMS appropriately. UCT differentiates between sensory input and output channels, but is lacking
specific operators. The other models do not fulfill this requirement. Learning processes are represented
well with SRK, ACT-R, and UCT. The SRK-framework interprets human learning as a shift towards lower
levels of cognitive control while ACT-R and UCT include chunking mechanisms. The other approaches
only partially fulfill this requirement.
Aspects of human error and reliability are represented well with extensions of SRK such as Reason
(1987) and Hannaman et al. (1985). COSIMO also copes with human reliability, but is restricted to
error detection and recovery. The other models are lacking sufficient mechanisms for human error mod-
eling. COSIMO, UCT, GOMS, and ACT-R provide strict formalisms for modeling of mental time con-
sumption and, therefore, fulfill this requirement completely. SRK offers limited functions in terms of
rules of thumb. Aspects of mental stress and strain are a fundamental part of ACT-R, which controls
rational behavior. Therefore, this model fulfills this requirement completely. Moreover, the work of
Moray et al. (1988a, b) must be taken into account, which enriches SRK with workload modeling. The
other models are lacking such evaluation mechanisms and do not fulfill this requirement. Semiotic
aspects of human-machine communication are represented well in SRK, which differentiates among
signals, signs, and symbols for information exchange. The other models do not integrate semiotic aspects
and, therefore, do not fulfill this requirement.
Concerning the first utility-driven requirement, the GOMS approach offers a variety of software tools
for analysis, modeling and evaluation and, therefore, fulfills this requirement completely. However,
Search WWH ::
Custom Search