The unitary base hypothesis

In discussions of word formation it is widely assumed that certain affixes only attach to bases of a certain syntactic category. For example, -ness is said to attach only to adjectives to form nouns as in empty-ness, -able attaches only to verbs to form adjectives as in breakable, -al is suffixed only to nouns to form adjectives as in constitutional. In the generative literature, such facts have led to the formulation of the so-called unitary base hypothesis (UBH), which claims that "The syntacticosemantic specification of the base ... is always unique. A WFR [Word Formation Rule, I. P.] will never operate on this or that" (Aronoff 1976:48, see also Booij 1977:140 141). The UBH is a strong hypothesis that can be refuted by showing that a certain word formation process operates on two distinct classes of bases. According to Aronoff, however, a rule that operates, for example, on nouns and adjectives does not necessarily speak against the UBH because nouns and adjectives form a natural class sharing the feature [+ N] to which the rule could refer unitarily. An example of this kind is the adjective-forming suffix -ly which attaches to nouns (as in manly, weekly), as well as to adjectives (goodly, northerly, see e.g. Marchand 1969:329-331).

In those cases where lexical categories do not form a natural class, an entirely different homophonous process must be assumed. Aronoff (1976:48) illustrates the latter point with the English suffix -able, which combines with verbal stems, as in breakable, perishable, as well as with nouns, as in serviceable, fashionable. For Aronoff, there are two -able rules, one deverbal with the meaning 'can be VERBed', the other denominal with the meaning 'characterized by NOUN'.¹

The UBH faces two main problems. The first is that Aronoffs escape hatch, namely the formation of natural classes, makes the UBH practically vacuous. Depending on the system of features and categories selected, even seemingly disjunct classes can be made into natural ones. Thus, in standard generative grammar natural classes can be formed on the basis of the categories [± N] and [± V], which leads to the conclusion that nouns and verbs can never form a natural class. In Jackendoff (1977), however, nouns and verbs form a natural class on the basis of the feature [+ Subj]. In essence, by choosing the appropriate feature system the UBH can be immunized against refutation.

The second problem is of a more empirical nature and is known as affix-generalization. Plank (1981:43-65) discusses a number of affixes (from a number of different languages) that can be found on the basis of more than one category and argues that the meaning of the derivatives is constant across the different categories of the base words. According to Plank, the preponderance of bases of a certain category is therefore best viewed as the consequence of the meaning of the process and not the result of a stipulated general condition on possible bases like the UBH.

¹ See Akmajian et al. (1979), Anderson (1992) for similar approaches to -able. The reader may have observed that Aronoffs deverbal rule runs into problems with intransitive verbs.