A recurrent theme in studying the interaction between human and formal
ism is the understanding of how people interact with representations i
n reasoning and communication. In contrast to the best known theories,
which approach the question of the impact of representation upon reas
oning through explanations in terms of human computational architectur
e, we present here a more fundamental approach. This approach separate
s the problem into two parts-issues about computational complexity ari
sing from the nature of the semantic interpretation (issues which are
abstract with regard to architecture); and issues about how human comp
utational architecture in particular can be brought to bear on differe
nt representations. On this view, for example, diagrams are often logi
cally inexpressive and this is why they lead to efficient inference. T
his paper presents experiences in applying this semantic approach to t
he empirical study of modality assignment in three disparate domains:
logic teaching, safety critical software engineering and the teaching
of formality. We show how, in each of these cases, an account of the s
emantics of representations in simple formal terms permits the analysi
s and modelling of what would otherwise be incomprehensibly complicate
d behavioural phenomena. The results of these apparently diverse studi
es indicate that individual differences in what might be termed cognit
ive styles have a significant effect upon a humans use and understandi
ng of various formalisms. This, we argue, is evidence that HCI researc
hers require a more analytical means to relate the cognitive and socia
l sides of HCI than has previously been available. Furthermore, we als
o take the studies presented here as evidence that our approach is a s
ubstantial step towards providing such a means of analysis. (C) 1997 E
lsevier Science B.V.