Halliday (2002 [1984]: 307-8):
To understand [grammatical] categories, it is no use asking what they mean. The question is not ‘what is the meaning of this or that function or feature in the grammar?’; but rather ‘what is encoded in this language, or in this register (functional variety) of the language?’ This reverses the perspective derived from the history of linguistics, in which a language is a system of forms, with meanings attached to make sense of them. Instead, a language is treated as a system of meanings, with forms attached to express them. Not grammatical paradigms with their interpretation, but semantic paradigms with their realisations.
So if we are interested in the grammatical function of Subject, rather than asking ‘what does this category mean?’, we need to ask ‘what are the choices in meaning in whose realisation the Subject plays some part?’ We look for a semantic paradigm which is realised, inter alia, by systematic variation involving the Subject: in this case, that of speech function, in the interpersonal area of meaning. This recalls Firth’s notion of ‘meaning as function in a context’; the relevant context is that of the higher stratum – in other words, the context for understanding the Subject is not the clause, which is its grammatical environment, but the text, which is its semantic environment.
