Reading time for the second clause of a conjoined sentence was found to be faster when the clause was structurally similar to the first clause than when the clausal structures differed. This "parallel structure" effect was found for each of several types of structures, including active versus passive constructions, direct object versus sentential complement (minimal vs. nonminimal attachment), nonshifted versus shifted heavy noun phrase, agent versus theme, and animate versus inanimate noun phrase. The pervasiveness ofthe effect ruled out some hypotheses about its basis, including the hypothesis that it would occur only when a subject's just having processed a structture would affect how temporary ambiguities are resolved. Detailed analysis of the data suggested the existence of several distinct sources of the effect and provided indirect evidence that people typically compute both a surface structure and an S-structure representation of a sentence. (la) Readers and listeners strongly prefer coordinated elements of sentences to beparallel in structure. The interpretations assigned to sentences such as (1) and (2) indicate that this preference does not amount simply to an aesthetic judgment or stylistic convention. The ambiguous sentence(1) Joshua hit the girl with a book and the boy with a bat is easily parsed as a pair of conjoined sentences, in which the prepositional phrases modify the verb phrases, as indicated in (1a), or as a clause including an object consisting of conjoined noun phrases with prepositional phrase modifiers, as in (lb). However, as we trust will be clear to the reader, there seems to belittle temptation to interpret this sentence as in (1c), in which the internal structures of the conjoined categories differ.

Proceedings of the Eleventh Annual Meeting of the Berkeley Linguistics Society (1985), pp. 519-528