Martin (1992: 104):
Because adjectives typically have a classifying rather than a descriptive function in generic nominal groups (realising Classifier rather than Epithet in Halliday's 1985: 163-165 terms), relevance phoricity is less commonly realised than in specific groups. However it is possible…
 This is a bear assertion (a guess) unsupported by evidence from corpora. According to Martin (1992 103):
Generic reference is selected when the whole of some experiential class of participants is at stake rather than just a manifestation of that class.
Martin's claim, then, is that when the whole of some experiential class of participants, such as 'people', is "at stake", rather than just a particular person, an adjective that modifies 'people' is more typically to function as a Classifier (e.g. 'boat people') than an Epithet (e.g. 'stupid people').
 To be clear, "relevance phoricity" is Martin's rebranding of his misunderstanding of Halliday's comparative reference and Martin's conclusion is based solely on the fact that 'Classifiers do not accept degrees of comparison' (Halliday & Matthiessen 2014: 377). However, comparative reference items are not limited to adjectives serving as Epithets, as demonstrated by such post-Deictics as same, similar, other, and nor are they even limited to nominal groups, as demonstrated by such adverbial groups as identically, similarly, otherwise; see Halliday & Matthiessen (2014: 626).