The effects of complexity on confidence ratings in linguistic integration

Abstract

Bransford and Franks (1971) introduced a paradigm for the study of linguistic integration. Their primary measure was subjects' rated confidence in their responses. The present paper considers the appropriateness of this dependent variable, and concludes that the relation between sentence complexity and rated confidence is too small and unstable to justify the emphasis which confidence ratings have received. Further, certain differences between concrete and abstract sentences were obscured by a confidence analysis. The proportion of old responses is championed as a more appropriate dependent variable. Finally, it is shown that increasing the amount of specific memory necessarily decreases the size of the complexity effect.There is growing literature concerned with the issue of linguistic integration, primarily stimulated by the work of Franks (1971, 1973; Franks & Bransford, 1972 In acquisition, sentences derived from each of several fours are presented, usually under a procedure not encouraging memorization of form. During recognition, both old and new sentences are tested. (New sentences are permissable combinations of ideas derivable from the four which were not presented during acquisition.) Subjects are required to label each recognition sentence as "old" or "new" and to assign a confidence rating to their categorizations. Bransford and Franks (1971) reported two facts of primary interest: Confidence ratings (CRs) increased sharply with complexity of the recognition sentence, and the functions for olds and news were quite similar; ratings for new ones were lower than for old ones, but otherwise it appeared that subjects could not distinguish olds from news. Bransford and Franks concluded that the complex idea was abstracted from the exemplars, no traces of which were retained. Hence, olds and news were indistinguishable, and the closer

Surface information loss in comprehension

“…Neither were the results due to subjects employing response biases or adopting guessing strategies. Although these criticisms have been aimed toward some of the language-based studies (Clifton, Kurcz, & Jenkins, 1965;Hayes-Roth & Hayes-Roth, 1977;James, Hillinger, & Murphy, 1977;Soli & Balch, 1976), none of them are warranted here.…”

Is memory schematic?

“…In fact, the Bransford and Franks (1971) findings are difficult to replicate unless their methods are followed closely. Changes in presentation modality (Flagg & Reynolds, 1977; see also Katz & Gruenewald, 1974), in presentation of materials (Flagg, 1976;Flagg & Reynolds, 1977;Katz, Atkeson, & Lee, 1974), in instructions (James, Hillinger, & Murphy, 1977), and in testing procedures (Anderson & Bower, 1973;Griggs & Keen, 1977; greatly reduce or eliminate the integration effects found with the original Bransford-Franks paradigm. Finally, "integration" will occur for nonsemantic or arbitrary materials (e.g., letter-digits, nonsense syllables) if the original procedures are used (Flagg, 1976;Katz & Gruenewald, 1974;Reitman & Bower, 1973;Small, 1975).…”

Integration versus decomposition in the retention of complex ideas

“…For instance, if each acquisition sentence contained only one or two elements of a complex idea, test sentences containing three or four elements might be rejected as being inconsistent with the general style of the acquisition material. This was subsequently confirmed in several different investigations Flagg, 1976;Pezdek, 1978).Nevertheless, in the absence of any information about the specific acquisition sentences, and in the absence of any concomitant information about the general style of the acquisition material, James, Hillinger, and Murphy (1977) showed that the overall relationship between the subjects' responses and the number of elements in a test sentence was purely the result of variations in response bias. Such biases can apparently be overridden by appropriate experimental instructions; thus, the confidencerating function is essentially horizontal when subjects are required to judge whether test sentences have the same meaning as acquisition sentences (Katz, 1973;Moeser, 1975).…”

“…Nevertheless, in the absence of any information about the specific acquisition sentences, and in the absence of any concomitant information about the general style of the acquisition material, James, Hillinger, and Murphy (1977) showed that the overall relationship between the subjects' responses and the number of elements in a test sentence was purely the result of variations in response bias. Such biases can apparently be overridden by appropriate experimental instructions; thus, the confidencerating function is essentially horizontal when subjects are required to judge whether test sentences have the same meaning as acquisition sentences (Katz, 1973;Moeser, 1975).…”