This paper, written in September 1998, has been published in Hanne Gram Simonsen and Rolf Theil Endresen (eds) 2001 "A Cognitive Approach to theVerb: Morphological and Constructional Perspectives" (Mouton de Gruyter), pp. 49-72. This collection of papers arose out of a workshop on `The verb in cognitive grammar' in June 1998. My talk at that workshop, called `A cognitive network for verbs', is the basis for the present paper, but length constraints have required me to omit a certain amount of material, and while working it up I found myself developing the ideas on networks further than in the talk; so I decided it needed a different title.
The paper develops the leading idea of Word Grammar and other "cognitive" theories of language, which is that language is a network. It reviews some of the consequences of this view: spreading activation, effects of conceptual distance, default inheritance, the unity of grammar and lexicon and, more generally, non-modularity; the unity of permanent and temporary representations, degrees of accessibility and binary relations. It then shows briefly how these ideas apply to two specific areas of language analysis: the contrast between polysemy and homonymy, and the treatment of regular and irregular morphology. The last section discusses Pinker's contrast between "mentalese" and "connectoplasm", and argues that the networks defined in this paper have all the symbolic qualities of mentalese, so maybe the mind uses "networks all the way down".