The unitary output hypothesis

The unitary output hypothesis (UOH) captures the idea that the derivatives formed on the basis of a certain word formation process can be characterized uniquely in terms of their phonological, semantic, and syntactic properties (cf. Aronoff 1976:22, Scalise 1984:137, 1988:232, Szymanek 1985:95). Two main objections can be raised against the UOH, one phonological and one semantic.

Given the wide-spreadedness of stem and affix allomorphy it can be seriously doubted that affixation indeed produces a phonologically unique output. It has been claimed, however, that processes which involve allomorphy are typically less productive or completely unproductive. For example, Cutler (1981) has argued that phonologically less transparent processes are also less productive. This position is corroborated by the findings in Kettemann (1988), who shows that many of the readjustment rules proposed in SPE do not extend to nonce words.

We will see, however, that even productive rules may tolerate a high degree of allomorphy. In particular, many of the derivatives featuring the verbalizing suffix -ize display a wide range of different stem alternations (e.g. truncation and stress reduction), which make the class of derivatives look phonologically disparate. It will be demonstrated, however, that these stem alternations can be accounted for in a straightforward unitary manner, thereby lending further support to the UOH.

Modifying his earlier position, Scalise (1988:244f, note 3) claims that the UOH is valid only "for the 'formal' part of a WFR [Word Formation Rule, I.P.], not for the 'semantic' part". To illustrate this point he cites the Italian verb forming suffix -are which is phonologically unitary, but, according to the source he cites (Lepschy 1981), semantically diverse.

Scalise's point is unconvincing in several respects. First of all it is uncontroversial that truly homophonous processes should be kept apart. No one would, for example, argue that English deverbal nominalizing -al (as in arrival) is in any sense semantically related to adjectival -al (as in conventional). The affixes differ in their meaning but each type of derivative is semantically uniform. Secondly, it is far from clear whether the examples of Italian -are derivatives Scalise cites cannot be accounted for in a unitary fashion by a proper semantic analysis (perhaps with certain extension rules that explain the polysemy involved). Interestingly, a similar problem arises for English -ize derivatives, whose possible meanings look considerably diverse, but which, at closer inspection, turn out to be a case of polysemy. In any case, it seems that the output of a rule is not only 'formally' but also semantically uniform.