Martin (1992: 505-6):
(i) Establishing genre as a level of semiosis which is not itself metafunctionally organised means that texts can be classified in ways which cut across metafunctional components in language. This strengthens Halliday's suggestion that field is strongly predictive of experiential choices, tenor of interpersonal choices and mode of textual ones without sacrificing the classification of texts into generic types such as narrative, exposition, procedure, report etc. … Generic labels such as narrative or exposition are impossible to tie satisfactorally [sic] to any one type of meaning; their realisation cuts across metafunctions. For this reason, it is useful not to associate genre too closely with any one register variable (e.g. mode in Halliday's work or field in Hasan's).
 Modelling 'genre as a level of semiosis' is inconsistent with the architecture of SFL theory and the meanings of the theoretical terms. Strata represent levels of symbolic abstraction, but a type of text (genre) is not more abstract than the semantics and lexicogrammar of the text type.
 This is inconsistent with one of the most fundamental postulates of SFL theory. Halliday and Matthiessen (2004: 60):
In fact, the threefold pattern of meaning is not simply characteristic of the clause; these three kinds of meaning run throughout the whole of language, and in a fundamental respect they determine the way language has evolved. They are referred to in systemic accounts of grammar as metafunctions, and the concept of 'metafunction' is one of the basic concepts around which the theory is constructed.
 In SFL theory, texts can be identified according to features of the three metafunctional dimensions — field, tenor and mode — of the context of situation that they realise. To classify texts in ways which "cut across metafunctional components" is to classify texts in a less principled way, and according to fewer and less elaborated features. Theoretically, it is a step backwards.
 This is manifestly untrue. Classifying texts without regard to metafunction has no bearing on metafunctional predictions across strata.
 No such sacrifice is required. To the extent that such classifications refer to the rôle played by a text type in a situation type, they are either features of the system of mode, or else the synthetic outputs of mode systems.
 This confuses context with language. Generic labels are features of context, whereas their realisation is in language. Moreover, "generic labels" can be modelled in terms of the textual metafunction (see ), and their realisation in language involves all three metafunctions (see ).
 Given the falsity of the reason, the conclusion does not follow.