Introductory remarks

Network models constitute a family of formats for representing lexical polysemy. Thus, they provide a particular type of answer to the question concerning the appropriate level of granularity of semantic definitions, i.e. “the level at which the relatedness of senses can be best observed and captured”. As pointed out by Lewandowska-Tomaszczyk, the monosemy position (espoused by neostructuralists) strives for the most parsimonious or schematic representation (a single sense), while the homonymy position (associated with generativists) is primarily concerned with deriving (any number of) senses from sets of matrices of linguistic properties (2007: 152). Both approaches are obviously motivated by different theoretical assumptions. I have little to say about the generativist approach (see J. Taylor 2007; and more generally Seuren 1998), which in CL plays mostly the part of the polemical “bogeyman” (as will be exemplified in the next paragraph). That approach seems to be motivated by the idea that linguistics should be chiefly concerned with “the underlying reality of language” (cf. Katz 1971), thus (in semantics) it neglects many of its surface manifestations. The level of granularity considered appropriate in particular cases is achieved by combinations of atomic properties.

The motivation for the monosemy position is much more complex and important in our context in view of the claims that CL is solidly based on Saussurian foundations.