Fontaine (2017: 8):
One interpretation of lexis as most delicate grammar is that they are somehow the end point of a bundle of system choices, which leads to a single lexical option. Hunston and Francis (2000:28) argue against this view:
If words have their typical phraseologies, such that words are not selected in isolation but in variable phrases, then it is unsatisfactory to propose that each lexical item is the end-point of an individual bundle of systemic choices. The best compromise that could be reached would be to propose that each bundle of system choices should end, not in a lexical item per se but in a ‘unit of meaning’. As units of meaning are, by their nature, indeterminate in extent, however, such an interpretation would involve system-choices leading to fuzzy-edged and overlapping units, at best.
Blogger Comments:
[1] This is a serious misunderstanding of SFL Theory. To be clear, the specification of lexical items by bundles of the most delicate lexicogrammatical features does not entail that words are selected in isolation. Grammatically, words serve as functions in group and phrase structures, and so they are selected with other words in those structures. Lexically, words have co-occurrence tendencies, and these collocations are modelled in SFL Theory as serving a cohesive function.
[2] Again, this is a serious misunderstanding of SFL Theory. To be clear, in SFL Theory, lexicogrammar (wording) is modelled in terms of the meaning it realises. That is, the lexicogrammatical features that specify lexical items are modelled in terms of meaning.
[3] To be clear, from the perspective of SFL Theory, units of meaning are not "by their very nature, indeterminate in extent". For example, the "units" of ideational meaning are the three types of phenomenon: sequence, figure, and element, which are congruently realised in wording as clause complex, clause and group/phrase, respectively (Halliday & Matthiessen 1999: 49). Their realisation in wording delimits, and so determines, the extent of each unit of meaning.