Wi 
a further loosening of feature theories in that there is no 
claim to exhaustive understanding of the features of a 
concept. Fodor et al. (1980) have recently reviewed 
definitional semantic theories, in both their strong and 
weak forms, and have faulted them for both theoretical and 
empirical reasons. 
Nevertheless, Armstrong, Gleitman & Gleitman (1983) 
have recently proposed a dualistic semantic theory. Their 
theoretical and empirical work suggests that we use 
rough-and-ready identification heuristics as well as formal 
and exact descriptions. They found, for example, that 
subjects can formally and infallibly classify numbers as odd 
or even, but that they can also identify numbers that are 
are better exemplars of "oddness" than others. McNamara & 
Sternberg (1983) have presented evidence for a two stage. 
weighted model of semantic representation. By their model, 
informal charactericstic criteria are processed first, and 
then, if necessary, the formal necessary ‘and sufficient 
attributes are processed in a weighted manner. This is an 
empirical restatement of the 'Ugly Duckling Theorem!’ 
(Watanabe, 1969), that if all features are equally 
important, then a duck shares as many features with a swan 
as a swan does with a swan.