Deriving Inflectional Irregularity

Conventional wisdom holds that productive morphology is regular morphology. Drawing evidence from French, we argue that the description of many lexeme formation processes is simplified if we hold that a productive rule may give rise to inflectionally irregular lexemes. We argue that the notion of a stem space allows for a straightforward description of this phenomenon: each lexeme comes equipped with a vector of possibly distinct stems, which serve as bases for inflectional form construction. The stem space is structured by default relations which encode the regular pattern of inflection; (partial) irregularities occur when a lexeme specifies a stem space violating the default relations. Derived irregularity is then the effect of a productive lexeme formation rule which specifies an irregular stem space for its output. 1 Productive irregular inflection 1.1 Background A central issue in the modeling of inflection is how the notion of (ir)regular inflection is taken into account. A distinct possibility is to give the notion no theoretical status (see e.g. Stump, 2001). For morphologists that seek to preserve the intuition that irregular inflection necessitate specific modeling, there are two options. Either we take regularity to be a design property of morphological systems, and thus try to model every possible process as regular, limiting irregularity to the description of blatant suppletion phenomena; this position is the default for most morphophonological work in the tradition of Chomsky and Halle (1968), and is strongly defended by proponents of distributed morphology (Halle and Marantz, 1993). Or we take regularity to be an empirical property, that is manifest in performance: speakers are able to inflect an unknown lexeme according to a regular pattern, but will not be able to inflect a lexeme according to an irregular pattern. This position is assumed in much of the psycholinguistic litterature on inflection, and defended forcefully, if somewhat partially, by Pinker (1999) and work cited therein. In this paper we assume the second position. Note that we do not commit ourselves to any particular view of the processing of inflection, but simply assume that (ir)regularity is a real grammatical phenomenon, that is manifest not only in psycholinguistic behavior but also in language change and in synchronic grammar. Our main empirical argument concerns the status of lexeme formation rules: we show that despite conventional wisdom, the output of lexeme formation rules should not always be considered inflectionally regular.