The notion of lexis as most delicate grammar is made difficult to understand by the fact that we don't have the grammar elaborated sufficiently delicately to the features that specify individual lexical items and the lexical sets that they form through shared features. But the principle can be understood by looking at articulatory phonology, where systems are sufficiently delicate.
In lexicogrammar, 'word' conflates two abstractions: word as grammatical rank and word as lexical item. The same conflation can be applied to the phoneme in articulatory phonology: a phoneme can be understood as both a phonological rank and an articulatory item.
As a phonological rank unit, the phoneme is a constituent of the higher rank unit, the syllable, and classes of phoneme, consonants and vowels, realise elements of syllable structure, Onset, Nucleus and Coda — just as a grammatical rank unit, the word is a constituent of the higher rank unit, the group, and classes of word, nominal, verbal etc., realise elements of group structure, Thing, Event etc.
As an articulatory item, the phoneme is the synthetic realisation of the most delicate articulatory features. For example, the phoneme /b/ is the synthetic realisation of the features [voiced, bilabial, stop] and phonemes can be grouped into articulatory sets on the basis of shared features, such as [voiced] and/or [bilabial] and/or [stop]. For example, the [voiced] set includes {b d g v z m n a e i o u w y}, the [bilabial] set includes {p b m} and the [stop] set includes {p b m t d n k g}.
It is in this sense that the lexical item is the synthetic realisation of the most delicate grammatical features, and that lexical items form lexical sets on the basis of shared features.
