Revised 9 January 2009
This paper is included in a collection of papers on grammatical constructions edited by Jan-Ola Oestman and Mirjam Fried: Construction Grammars. Cognitive grounding and theoretical extensions (Benjamins, 2005, pp 243-72). The first draft was written in November 1999, but in the light of comments (from Joe Hilferty and two readers) we have completely rewritten it. This draft dates from October 2001.
In this paper we consider the relationship between Word Grammar (WG) and Construction Grammar (CG). We first of all argue that the two frameworks share all of their fundamental assumptions, naming specifically the following: a requirement to deal with 'noncore' as well as central patterns, rejection of the traditional division between rules and separate lexical items, declarative monostratal grammatical structure (no transformations), full integration of syntactic and semantic structures, decomposition of linguistic structures into simple constructions which (by an inheritance mechanism) determine the structures and properties of utterances, and a requirement to include all relevant information (including illocutionary force, presupposition etc) among the properties of a linguistic item. Next we consider two differences between the two frameworks: WG uses dependency structure where CG uses phrase structure, and WG uses default inheritance where CG (in some versions) uses strict unification-based inheritance. The first of these differences is held to be a material, but not a fundamental, difference between the frameworks. We show that a dependency-based network approach with 'normal' mode inheritance (Flickinger, Pollard, and Wasow 1985) allows for a more flexible characterisation of linguistic structures and that the box-based approach usually applied in CG analyses both excludes some patterns found in existing linguistic structures and requires arbitrary stipulations to rule out patterns that are not found.
In order to demonstrate the similarities and differences between the two frameworks we present a WG analysis of two structures which have both been covered in the CG literature. Using Kay and Fillmore 1999 as a source of data, we give a WG analysis of the What's X doing Y? construction as a combination of (specialisations of) five smaller constructions which all have uses in other structures. We go on to compare a WG analysis of the double object construction with that given in Goldberg 1995. The double object construction in WG is a network of morphosyntactic and semantic properties structured around the indirect object relationship and we consider some core uses of this construction, as well as some motivated extensions which exploit particular semantic relationships associated with the indirect object (namely beneficiary and owner). We conclude that the descriptive mechanisms of WG provide for a full and flexible characterisation of the kind of explicit non-derivational grammar envisaged by Kay and Fillmore 1999.