A dynamic stimulus-driven model of signal detection.

Abstract

Signal detection theory forms the core of many current models of cognition, including memory, choice, and categorization. However, the classic signal detection model presumes the a priori existence of fixed stimulus representations-usually Gaussian distributions-even when the observer has no experience with the task. Furthermore, the classic signal detection model requires the observer to place a response criterion along the axis of stimulus strength, and without theoretical elaboration, this criterion is fixed and independent of the observer's experience. We present a dynamic, adaptive model that addresses these 2 long-standing issues. Our model describes how the stimulus representation can develop from a rough subjective prior and thereby explains changes in signal detection performance over time. The model structure also provides a basis for the signal detection decision that does not require the placement of a criterion along the axis of stimulus strength. We present simulations of the model to examine its behavior and several experiments that provide data to test the model. We also fit the model to recognition memory data and discuss the role that feedback plays in establishing stimulus representations.Keywords: signal detection theory, recognition memory, cognitive modeling, dynamic models of information processing Signal detection theory (SDT) is crucial to many important theories in cognitive psychology, especially those theories that deal with performance in two-choice tasks. In such tasks, an observer is presented with a series of trials in which he or she must respond to a stimulus. The stimulus is one of two types, either "noise" (requiring a response of "no") or "signal" (requiring a response of "yes"). What constitutes noise or signal can be very flexible.The SDT framework assumes that the presentation of a stimulus gives rise to a perception of some sensory effect in the cognitive apparatus of the observer. The magnitude of the effect, conceived on some relevant experiential scale such as "loudness" or "familiarity," is used as the basis of the "yes" or "no" decision. Random noise in the observer's perceptual system (or in the stimulus itself) results in varying magnitudes of effects over different stimulus presentations but, on average, signals result in larger effects than noise. Variability in sensory effects is represented by two random variables that often are assumed to follow equal-variance Gaussian distributions, though this assumption is not strictly necessary. To make a decision, traditional SDT assumes that observers place a criterion along the axis of sensory effect. The "yes" or "no" response is determined by whether the perceived effect is above or below this criterion (see Macmillan & Creelman, 2005, for a review).SDT is not confined to the relatively simple problem of detecting the presence of signals. Any two-choice task that can be recast as a magnitude judgment can be shoehorned into the SDT framework. Consider, for example, a lab technician whose job is to examine Pap smears an...

Modulation of additive and interactive effects in lexical decision by trial history.

“…Such a complex constellation of results required a follow-up to assess its reliability and generalizability. Therefore, in Experiment 2, we investigated the question of whether this modulation was the result of an ineluctable characteristic of the word-processing architecture or instead was the product of dynamic adaptation to varying trial difficulty, in line with models such as the ASE (Kinoshita et al, 2008(Kinoshita et al, , 2011 or the Turner et al (2011) account of learning stimulus representations. Rather than randomly mixing clear and degraded targets, subjects in Experiment 2 experienced these two types of display in separate blocks of trials.…”

“…A related approach to accounting for patterns of additivity and interaction emphasizes the nature of stimulus representations that enter into decision processes. As we have discussed, Turner et al (2011) proposed that changes in task performance across trials, including the influence of recent trial history, are determined in part by the development of stimulus representations. We suggest that differential learning about stimuli, potentially as a function of any of the three major factors considered here, can contribute to the nature of the relationship observed between those factors without appeal to any modulation of processing architecture.…”

“…Under these conditions, rapidly discriminating words from nonwords in a lexical decision task, for example, may depend on stimulus learning that accrues over the course of a substantial number of trials. Turner, Van Zandt, and Brown (2011) proposed a formal model of stimulus discrimination (signal vs. noise) in which subjects develop signal-to-noise likelihood ratios for various points along a stimulus strength axis. These ratios are modified by recent trial events in the following way.…”

“…The results described in the previous section clearly document that characteristics of the previous trial (i.e., the lexical status and the stimulus quality of the last target) exert strong effects on response speed in the current trial. According to Turner et al (2011), subjects may develop stimulus representations over time, and this learning increases the ability to discriminate between stimulus classeswords and nonwords. We investigated the possibility that these effects would change across trials in the experiment.…”

Human memory research: Current hypotheses and new perspectives

“…Since its initial proposal as a framework to account for recognition memory, SDT was adopted by several memory theorists as the framework of choice for characterizing this phenomenon (e.g., Morrell, Gaitan, & Wixted, 2002;Turner, Van Zandt, & Brown, 2011). The main reason for this is perhaps the elegance and parsimony of the model, its consistency with a number of theories of memory (e.g., Global Matching Models; Clark, & Gronlund, 1996), and its outstanding capacity of generating useful indexes to measure memory performance (e.g., d' to measure accuracy and C to response bias, Snodgrass & Corwin, 1988).…”