Martin (1992: 116):
Similarly, as with superset reference, comparison in terms of quality is recursive, so that a relevant participant can be presumed with respect to an indefinite number of qualities. Again, this does not mean that each comparison presumes an additional relevant participant. If a smaller, lighter coloured, cuter frog in [3:39] is qualified, it is qualified with respect to another or another set of participants, not three different participants, one bigger, on darker and a third less cute: a smaller, lighter coloured, cuter frog that the one(s) I saw.
[3:39] It was a smaller lighter coloured, cuter frog.
 To be clear, this is the logical modification of Epithets in nominal groups, which is an aspect of the grammatical construal of experience. Here it is misunderstood as an aspect of textual reference and rebranded as a discourse semantic system. The confusions are in terms of metafunction and stratum.
 This claim is based on the mistaking of nominal groups for reference items. It is each of the comparatives that functions as a reference item, and the question of how each comparison is individually resolved depends entirely on the co-text.
Halliday & Matthiessen (2014: 632-3) explain comparative reference as follows:
In comparative reference, the reference item still signals ‘you know which’; not because the same entity is being referred to over again but rather because there is a frame of reference – something by reference to which what I am now talking about is the same or different, like or unlike, equal or unequal, more or less. Comparative reference items function in nominal and adverbial groups; and the comparison is made with reference either to general features of identity, similarity and difference or to particular features of quality and quantity.
Any expression such as the same, another, similar, different, as big, bigger, less big, and related adverbs such as likewise, differently, equally, presumes some standard of reference in the preceding text.