In a study using eye-voice span (EVS) measures to determine the relationship between the types and numbers of transformations in sentences and the predictability of the sentences for readers, sentences with varying types and numbers of transformations were embedded in paragraphs projected onto a screen and read aloud by the subjects, 20 graduate students. When subjects reached predetermined critical points in the target sentences, the paragraphs were immediately removed from the screen; the number of correct consecutive words that subjects remembered seeing beyond the last word read orally was taken to be the EVS measurement for that paragraph. Analysis of the results indicated that a complex interrelationship exists among the three independent variables--type of sentence transformations, basic sentences versus transformations, and critical point--with regard to the EVS measurement. The results of this study do not coincide with expected results based on the work of previous researchers; rather, they indicate that such previously proposed language processing models as the deep structure model, the chunking model, and the model suggested by Levin and Kaplan offer overly simplistic explanations of language processing. (GW)