Tests of serial scanning in item recognition.

Abstract

We report two experiments designed to bear critically on Sternberg's (1966) theory that human subjects scan a memorized set of items serially and exhaustively in order to determine whether or not it contains a given probe item. In each experiment, subjects memorized sets of 2,4,6,8, 12, and 16 items, and after memorizing each set were presented with a series of 96 probes for that set. Although mean reaction times increased with set size in a manner largely consistent with serial scanning, analysis of reaction-time distributions showed that the effect of increasing set size was mainly to increase the positive skew of the distribution rather than to shift the distribution as a whole. The slope of the function relating mean reaction time to set size was influenced by repetition of negative probes in the series of probes (Experiment I), and was higher when a probe was a two-syllable non-noun that when it was a one-syllable noun (Experiment II). The latter effect was independent of whether the sets consisted wholly of nouns or of non-nouns, or whether they were mixtures of the two. The only serial-scanning model that can be reconciled with these findings is one in which subjects scan on only a proportion of trials. The data are also generally consistent with Ratcliff's (1978) theory, which assumes parallel search through items in memory. RESUME Compte rendu de deux experiences mettant en cause la theorie de Sternberg (1966) selon laquelle les sujets humains precedent par balayage seriel et exhaustif d'un ensemble memorise d'items pour determiner s'il contient ou non un item-sonde donne. Dans les deux experiences, les sujets doivent memoriser des ensembles 2, 4, 6, 8, 12, et 16 items et, apres avoir memorise chacun, doivent reagir a une serie de 96 items-sonde pour chaque ensemble. Meme si le temps de reaction moyen augmente avec la dimension des ensembles d'une maniere consistante avec I'hypothese d'un balayage seriel, I'analyse des distributions des temps de reaction montre que I'effet du a I'augmentation de la dimension des ensembles est principalement d'accrottre I'asymetrie positive de la distribution plutot que de modifier la distribution dans sa totalite. La pente de la fonction reliant le temps de reaction moyen a la dimension des ensembles est influencee par la repetition d'items-sonde negatifs (experience I) et est plus accusee quand I'item-sonde est un non-nom de deux syllabes que lorsqu'il s'agit d'un nom d'une syllabe (experience II); ceci, que les ensembles soient composes de noms, de non-noms ou d'un melange des deux. Le seul modele de balayage seriel qui puisse etre concilie avec ces donnees est celui dans lequel les sujets effectueraient un balayage sur une partie des essais seulement. Dans I'ensemble, les donnees sont egalement compatibles avec la theorie de Ratcliff (1978) selon laquelle il existe une recherche parallele des items en memoire.

Scanning two memorized lists

“…We also included a control condition, in which the words in the two sets were incorporated into a single list, and the response choice was reduced to a dichotomous one. The conditions were similar to those of previous experiments (e.g., Corballis, Katz, & Schwartz, 1980;Corballis & Miller, 1973;Hockley & Corballis, 1982) that have yielded linear set-size functions for single memorized lists, at least over the range of total memory loads (two to eight words) used in the present study.…”

“…Although evidence from experiments on the Sternberg paradigm, and variants of it, continue to confirm the linear relation between RT and list length, at least over limited ranges of list lengths, some authors have argued strongly against interpretation in terms of scanning, or have at least shown that there are severe restrictions on the kinds of scanning process that can be plausibly entertained (e.g., Hockley & Corballis, 1982). For example, the shape of the RT distributions makes it unlikely that scanning is more than an intermittent strategy.…”

Choice and response time processes in the identification and categorization of unidimensional stimuli

“…6) proposed, the exponential process may be conceptualized as the decision component (i.e., the time required to decide which response to make) whereas the Gaussian component may be conceptualized as the transduction component (i.e., the sum of the time required by the sensory process and the time required to physically make the response). Although the theoretical proposition that cognitive process RTs are the sum of two additive processes is difficult to test, several researchers have demonstrated that the ex-Gaussian function provides a very good fit to several empirical RT distributions (Hockley, 1984;Hockley & Corballis, 1982;Luce, 1986;Ratcliff & Murdock, 1976). Other authors have suggested that, in the absence of any theoretical assumptions, the exGaussian function can be used effectively to characterize an arbitrary RT distribution (see Heathcote, Popiel, & Mewhort, 1991).…”

Retrieval of item frequency information in a continuous memory task

“…The additional variable included in Experiment 2 was a noun/nonnoun stimulus manipulation. Hockley (1982) and Hockley and Corballis (1982) found that nonnouns were responded to more slowly than were nouns in item recognition. It is reasonable to assume that this manipulation affected the availability of item information, because it does not seem likely that response information would differ for nouns and nonnouns.…”