A long-standing assumption in linguistic analysis is that different languages and constructions can be described in terms of the same grammatical categories and relations. Individual grammatical categories and relations are in fact often assumed to be universal. Grammatical categories and relations display however different properties across different languages and constructions, which challenges the idea that the same categories and relations should actually be posited in each case. These facts have been dealt with in two major ways in the functional-typological literature. In a widespread approach, the same categories and relations are posited for different languages and constructions provided that these all have categories and relations that display some selected properties. In a more recent approach, this idea is abandoned, and grammatical categories and relations are argued to be language-specific and construction-specific. This article provides a critical review of these approaches, and a comparison is made with some generatively oriented approaches. In particular, it is argued that a distinction should be made between two views of grammatical categories and relations. In one view, grammatical categories and relations are classificatory labels indicating that a variety of linguistic elements display some selected property. In another view, grammatical categories and relations are proper components of a speaker's mental grammar. While cross-linguistically valid (or possibly universal) and cross-constructionally valid categories and relations can be posited when classifying linguistic elements based on observed grammatical patterns, there is no obvious evidence that such categories and relations exist at the level of mental representation. This is, however, because of the absence of conclusive evidence about the organization of a speaker's mental grammar, rather than because of the linguistic evidence as such.
See FULL TEXT