How conceptual are semantic maps?

The question addressed in this paper is whether (and to what extent) a semantic map aimed at representing the multifunctionality of a given construction (or set of constructions) in discourse can be thought of as endowed with conceptual reality. To be considered as a mental representation that is essentially similar in all human brains, such a map should meet two requirements: (i) its nodes should be bundles of semantic and pragmatic properties that form conceptual archetypes, that is, ways of conceptualizing and categorizing dynamic or static configurations that are fundamental to human experience; (ii) there should be a high degree of regularity in the data material, i.e. each construction should be associated with a node or a contiguous set of nodes in a regular way. However, observing the use of grammatical constructions in discourse provides us with compelling evidence that discourse contexts are complex entities involving many different variables, and that “a perfect fit is not the usual state of affairs for models of complex human behavior (including language)” (Croft and Poole 2008:6). Based on a previous analysis of various passive and impersonal constructions in a parallel corpus of five European languages, I will argue that a first-generation semantic map representing the distribution of these constructions in discourse and comprising a few conceptual archetypes may be only an idealized abstraction over the conflicting evidence of the association between discourse contexts and construction types. As an idealization, such a map is not particularly informative as to language-specific tendencies and idiosyncrasies and does not allow us to analyze all the datasets that we might be interested in analyzing. On the other hand, a second-generation semantic map proves to be a more reliable tool for representing variation in discourse and does not force the analyst to posit (and multiply) conceptual structures where there may be none.