In a recent paper on the status of morphology in a generative theory of grammar, Zwicky (1982a) has argued “that processes of cliticization and readjustment together constitute a component of grammatical description in any language, a component related to others by strict principles of precedence…that syntactic rules, as a set, can feed or bleed rules of cliticization/readjustment (but not vice versa)” (Zwicky 1982b:51). Here, I shall consider the question of whether such an assumption of strict directionality can be maintained in a theory of language processing, in which generative rules of syntax and cliticization are replaced with rules of parsing and word recognition. My hypothesis is that given certain natural assumptions about language processing, rules for analyzing complex phrases and rules for analyzing host-clitic groups must be able to feed information to each other, making directionality an untenable assumption. To develop this hypothesis in concrete terms, I shall consider certain problems which contracted auxiliaries pose for a theory of language processing: specifically, I shall show that these problems can be elegantly handled if parsing rules and word recognition rules are permitted to interact in a nondirectional manner.
Stump, Gregory, "Directionality and the Processing of Contracted Auxiliaries" (1983). Linguistics Faculty Publications. 36.