Phantom Recollection of Bridging and Elaborative Inferences

Abstract

The phantom recollection model is a multiprocess analysis according to which memory judgments are collaboratively supported by one's recollection of an item in its context, a vaguer sense of stimulus familiarity, and the phantom recollection of the substance and even perceptual details of unstudied but related lures. Phantom recollection has previously been documented to support readers’ judgments of certain logical implications of sentences and discourse bridging inferences. This study evaluated the representational status of elaborative inferences in a comparable way. Subjects read blocks of brief passages and then judged explicit, bridging inference, elaborative inference, and control test items. Different subject groups were instructed to make their judgments with (1) a verbatim criterion, or (2) a gist criterion or (3) to accept only items implied but not stated in their passages. Multinomial tree processing analysis was applied to the data. Consistent with numerous other behavioral measures, these procedures provided evidence of phantom recollection support for bridging but not elaborative inferences. This outcome bears on the representational basis of elaborative inferences. The merits but also some hazards of the present methodology are discussed.

Acknowledgments

This research was supported by Discovery Grant OGP/9800 from the Natural Sciences and Engineering Research Council of Canada and internal funding from the University of Manitoba. These data were presented at the meeting of the Society for Text and Discourse, Chicago, August 2014.

Notes

¹ The appearance of recollection in the top branches of Figure 1a and 1b, implies that given sufficient processing time, recollection overrules the other processes. It does not bear on the issue of whether recollection is the faster process (Brainerd et al., 2001, footnote 3).

² We thank Mark McDaniel for providing the corpus of passages.

³PD^λ is a family of statistics that asymptotically has the X² distribution. The goodness-of-fit statistic, G², is equivalent to PD with λ = 0 (Moshagen, 2010).