Martin (1992: 273):
The folk-conception of a dictionary entry as a "form" and the definition as its "meaning" is of course a misleading one. In fact the relationship between words and definitions is not stratal (i.e. words are not being associated with the semantic primes they realise). All that is going on is that one wording is being reformulated as another.
Halliday & Matthiessen (2004: 237) see this rather differently, because they apply SFL theory to definitions:
Naming and defining are linguistic exercises, in which the word is Token and its meaning is its Value. …
a ‘gazebo’ is a pavilion on an eminence Token Value
(= ‘the word gazebo means [names, is defined as] a pavilion on an eminence')