第102页 6.1 The homogeneity assumption

There’s no empirical reason why we should expect the ‘best’ feature system to be homogenous, with all features of only one type or another. But there is a kind of a priori reason why linguists often want or expect feature systems to be homogeneous, or insist that they’re defective if they’re not; and this boils down in the end to a matter of aesthetic preference plus philosophical commitment (occasionally masquerading as ‘scientific conscience’). The assumption that systems of primitives ought to be formally homogeneous is the result of what I like to call ‘the atomic fallacy’: the view that – what we get down to a basic enough level – the primitive elements of whatever universe we’re concerned with will turn out to be of one type only.