Abstract
The canonical approach is designed to avoid two dangers in typology: the premature use of statistics and the failure to compare like with like. It involves taking definitions to their logical end point, in order to build theoretical spaces of possibilities. The canonical instances are not necessarily the most frequent, and may indeed be rare, but they fix a point from which occurring phenomena can be calibrated. This approach is demonstrated first in syntax, by looking at agreement. It is then applied to morphology, starting from the notion of a canonical paradigm, and showing how four major morphological phenomena can be seen as mismatches between the canonical expectation and the variety of real inflectional systems. Syncretism and suppletion are investigated as examples. The latter demonstrates how typology can be applied to an extreme case (lexical items whose constituent parts have no shared phonology), which can be extended by interactions with other morphological phenomena. The relevance of this approach is shown in its application to the construction of typological databases.