There are at least three distinct interpretations of the statement in (1):

    (1)  "identify is the base of identifiable. "

On a first interpretation, identify is the morphosyntactic base of the derived word identifiable: a syntactic property of identify - the fact that it is a transitive verb - is a prerequisite for the affixation of -able. On a second interpretation, identify is the semantic base of identifiable: the semantics of the -able form are a function of those of the verb contained within it. A third interpretation of (1) has to do with phonology : the phonological shape of the -able word is a function of the shape of its inner constituent.

One goal of the talk is to show that the usual conflation of the concepts of phonological and morphosyntactic or semantic base cannot be maintained. This can be shown by examining a cases in which complex expressions are formed by reference to several distinct reference terms: this is the split-base effect. For instance, in forming a novel adjective such as remediable (recorded with this stress in the OED but treated as a nonce form by most speakers), the speaker must consult not only the verb remedy, which serves as its morphosyntactic base, but also the adjective remedial, which serves its phonological base, in lending its stress to the -able form. I argue that split-base effects arise when the morphosyntactic or semantic base of affixation lacks a phonological property that is desirable in the derivative: in such cases, the derivative may adopt the phonology of a distinct listed allomorph, here the stress of remedial. The stress pattern adopted in the novel form must however have a lexical precedent, a precedent in a listed allomorph: it is this requirement that blocks (for many speakers) forms like remediable. Split-base effects will be documented for a large class of English suffixes - previously classified as belonging to Level 1 (e.g. -atory, -ify), Level 2 (e.g. -ize, -ism) and variably Level 1 or 2 (e.g. -able). In all these cases, novel affixed forms display the systematic option of metrical improvement, but only in lexical paradigms that include more than one stress pattern. It is shown that the notion of lexical conservatism (avoidance of phonologically innovative forms) provides a better classification of English affixes than the Level 1/Level 2 distinction or the recent OT successors of these notions (output-output correspondence to a unique base form).

The talk's second point is to outline an analytical framework that uses the notion of lexical conservatism to characterize in unified fashion both split-base and single-base derivatives.