Cloze, Transformational Theory and Redundancy
Frederick Bowers, Phil L. Nacke · Journal of Reading Behavior · 1971
Recent linguistic theory raises some important issues of relevance to reading researchers, particularly to those concerned with measuring reading skill or with calculating the relative readability of texts. In particular, the generative transformational theory of language initiated by Chomsky (1957, 1965) makes one reappraise some of the assumptions which lie behind the construction of experiments using cloze procedures. Of course, since Taylor's original proposals were made in 1953, several researchers, in particular Rankin, Weaver, Coleman and Treisman, have modified the raw cloze procedure to allow for the difference between the structural and referential morphemes of a language, but even they have not met the considerable counters to the theoretic basis of the cloze procedure which generative theory presents. On the other hand, there are some strategies indicated by generative theory which researchers would find useful, but which seem not to have been adopted. In section 2—redundancy and information theory— we present, the reasons for which we think one must doubt the validity of the simple cloze procedure, and in section 3—linguistic redundancy—we present the linguistic motivation for regarding certain items of discourse as redundant on syntactic or semantic grounds. A tentative algorithm for the deletion of redundant words from written English prose is presented finally, based on the linguistic evidence that recent and current linguistic theory has illuminated. The practical reason for our exploring linguistic theory was to find a means of arriving at a straightforward deletion procedure for an experiment to test the hypothesis that fast reading was dependent on the reader's ability to reject words and strings of words of no information (Nacke, 1970). A cloze procedure of the n word deletion kind would clearly have been quite unhelpful in establishing whether or not a selective deletion strategy was being used by a fast reader; what we needed was a deletion process, (1) which reflected linguistic behavior, and (2) whereby successive scans from surface structures to deep structure could be made so that different qualities of restoration might be revealed. The results of the experiment, which will be published separately, encouraged us to offer the linguistic motivation and resulting algorithm for other researchers to test, in the spirit of Bor-