Martin (1992: 576-7):
Basically Bernstein's suggestion […] is social class positions subjects to make meaning in distinctive ways depending on context. Taking up Hallidays' (Thibault 1987: 620) terms quoted above, code "bifurcates" register, with the result that speakers from different classes (or generations, ethnicities and genders) construe context in different ways. In Bernstein's own terms:
… I shall take the view that the code which the linguist invents to explain the formal properties of grammar is capable of generating any number of speech codes, and there is no reason for believing that any one language code is better than another in this respect. On this argument, language is a set of rules to which all speech codes must comply, but which speech codes are realised is a function of the culture acting through social relationships in specific contexts. (1971/1974: 197)
Without an interpretation of these divergent speech codes, or better, fashions of meaning, contextual theory does indeed run the danger of over-determining, homogenising and thereby reifying semiotic communities. The notion of 'fashions of meaning' which has been used to relativise context here is based on work by Whorf who differentiated languages and cultures on the basis of different fashions of speaking …
 To be clear, in terms of SFL theory, Bernstein's codes regulate the selection of meanings in the registers that realise situation types. Halliday (1978: 67, 68):
In terms of our general picture, the codes act as determinants of register, operating on the selection of meanings within situation types: when the systemics of language — the ordered sets of options that constitute the linguistic system — are activated by the situational determinants of text (the field, tenor and mode […]), this process is regulated by the codes. …It is important to avoid reifying the codes, which are not varieties of language in the sense that registers and social dialects are varieties of language. […] The code is actualised in language through register, the clustering of semantic features according to situation type. (Bernstein in fact uses the term ‘variant’, i.e. ‘elaborated variant’, to refer to those characteristics of a register that derive from the choice of code.) But the codes themselves are types of social semiotic, symbolic orders of meaning generated by the social system. Hence they transmit, or control the transmission of, the underlying patterns of a culture and subculture, acting through the primary socialising agencies of family, peer group and school.
 To be clear, in terms of SFL theory, Bernstein's 'fashions of speaking' are registerial varieties, at the level of semantics, as regulated by Bernstein's codes. Halliday (1978: 25):
The ‘fashions of speaking’ are sociosemantic in nature; they are patterns of meaning that emerge more or less strongly, in particular contexts, especially those relating to the socialisation of the child in the family.
 This misconstrues under-specifying as over-determining.
 This misconstrues under-specifying as homogenising.
 This misunderstands the meaning of 'reify'. The meaning of reify is to convert into, or regard as, a concrete thing; that is, to metaphorically construe a phenomenon that is not a thing as a thing. To claim that a community is not congruently a thing, is to claim that is either a quality, a process or a circumstance. Of course, in using the word 'communities' Martin has construed the phenomenon as a thing himself. On the basis of grammatical reactances, Halliday & Matthiessen (1999: 193) classify human collectives as things that are intermediate between conscious things and non-conscious semiotic things (institutions):
Human collectives: intermediate between conscious beings and institutions. These can function as Senser in figures of sensing of all kinds, including those embodying desideration; but they accept either singular or plural pronouns, and if singular pronominalise with it (e.g. the family says it is united/ the family say they are united).