Constructions in language description

This paper addresses some rarely-discussed theoretical and methodological issues in the application of a monostratal, construction-based linguistic model to language description. Constructions, in this approach, are defined as complex, schematic signs which are non-compositional, i.e. which have to be learned in a way which is similar to lexical items. In principle, the construction-based model is well suited to language description, since constructions which are specific to a language can be dealt with in the same framework as constructions which are common cross-linguistically. Yet, identifying a construction and distinguishing it from a compositional combination of signs, or from a formally and functionally related construction, is not always a straightforward task in descriptive practice. This is particularly true for spoken language, where the interaction of prosody and grammar has to be taken into account. I will propose here that gestalt-based constructions or G-Constructions, for which coherence is ensured by contiguity and prosodic means, should be distinguished from R-Constructions, where a relational element is responsible for the coherence of the construction, and which may therefore be distributed among two (or more) intonation units. These issues are illustrated by applying a construction-based description to various multi-predicate expressions in the Australian language Jaminjung.