Memory & Cognition1990, 18 (4), 428-429Notes and CommentDetectionless processing with semantic activation?A footnote to Greenwald, Klinger, andLiu (1989)J. R. DOYLEUniversity of Bath, Bath, EnglandRecently published research has suggested that, in a patternmasking task, semantic activation caused by the target may con-tinue to exist even though subjects cannot detect the target. Theexperiments are reassessed as an exceptional case of the moregeneral rule that subjects are able to use residual semantic acti-vation to actually detect targets. Furthermore, residual graphicinformation is far less effective at supporting near-chance tar-get detections.In a recent paper that appeared in this journal, Green-wald, Klinger, and Liu (1989) presented findings fromthree experiments. They claimed the following:Coupled with previous research by others using thelexical decision task, these findings converge in es-tablishing the reliability of the empirical phenome-non of semantic processing of words that are renderedundetectable by dichoptic pattern masking. (p. 35)Facilitation oflexical decision by the "detectionless pro-cessing" of a prime is essentially the same claim as thatmade by Marcel (1983a, 1983b) and Fowler, Wolford,Slade, and Tassinary (1981). What makes Greenwaldet al. 's paper important is the considerably greater rigorwith which they establish the nondetectability of stimuli.Notwithstanding these positive findings, others havefound no evidence for detectionless processing. For in-stance, Cheesman and Merikle (1984, 1986) have arguedfor the distinction between a subjective and an objective(detectionless) assessment of whether stimuli are uncon-scious. They found that presentations that had used asubjective assessment showed evidence of high-level acti-vation, whereas the "objective" assessment did not. Sum-marizing this and other evidence, Doyle and Leach (1988)hypothesized thatsemantic activation of words is used to make (some)detections .... A d' of zero would only be possibleif there were no semantic activation, for if there weresemantic activation in the absence of graphic activa-tion, subjects might use the semantic activation tomake detection responses, thus pushing d' abovezero. (p. 295)Yet Greenwald et al. claim to have found semantic acti-vation when d' = O. The purpose of the present note isCorrespondence may be addressed to J. R. Doyle, School of Manage-ment, University of Bath, Claverton Down, Bath BA2 7AY, U.K.to show how these apparently contradictory findings areactually entirely consistent.Greenwald et al. used a two-alternative forced-choiceposition judgment to determine whether or not stimuliwere detected. Words appeared either to the left or to theright of a fixation point. Subjects responded by pressingone of two buttons, the left button if they thought the tar-get appeared to the left of fixation and the right buttonif they thought it appeared to the right. The particularlyinteresting manipulation was that the words LEFT andRIGHT appeared to the left or right of fixation: On halfof the test trials, they were on the "correct" side of fixa-tion (consonant presentation); on the other half, they wereon the "incorrect" side of fixation (nonconsonant presen-tation).Clearly, if no semantic activation occurs at d' = 0, itcould make no difference whether the word LEFT waspresented to the left of fixation or to the right of fixation,and the same holds true for the word RIGHT. However,there was an asymmetry, indicating not only that seman-tic activation was present but also that the semantic acti-vation was able to affect a voluntary response. This isdifferent from the customary method of observing seman-tic activation indirectly (e.g., through its ability to primeanother word).Let us assume a simple additive model, in which s is thestrength of the semantic information and g is the strengthof the graphic information in making detections, measuredrelative to the 50% baseline of chance responding. Con-sonant trials should yield performance of (50 + g + s) %correct; nonconsonant trials should yield (50+g-s)%correct. Overall, an equal number of consonant and non-consonant trials were presented, so that overall detectionsshould be their average [i.e., (50 + g) %]. From the aver-ages given in Greenwald et al. 's (1989) Results subsec-tions "Masked distractor effects on positionjudgments"(pp. 38,40, and 42), we can compute sand g. These cal-culations are performed for their three experiments in Ta-ble 1. In their Experiment 1, "the procedures obliged per-formance to be at least slightly below chance" (p. 38),but was only assessed prior to the presentation of prim-ing trials. The asterisked figures are therefore estimates.In Experiments 2 and 3, checks for adaptation effects weremade in the middle of and after the priming trials, andexact overall performance figures are given.It is interesting to note that s is about three times moreeffective in facilitating detections than is g, a finding thatlends some support to Marcel's original claim (labeledhypothesis HI in Doyle & Leach, 1988) that, as target-mask SOA was reduced, semantic judgments about amasked word could be sustained much better than couldgraphic judgments.Copyright 1990 Psychonomic Society, Inc.428
