ACT-R models of a delayed match-to sample task - eScholarship

ACT-R models of a delayed match-to sample task Sarah Cebulski (sarahcebulski@cmail.carleton.ca) Institute of Cognitive Science, Carleton University, 1125 Colonel By Drive Ottawa, On., Canada Sterling Somers (sterling@sterlingsomers.com) Institute of Cognitive Science, Carleton University, 1125 Colonel By Drive Ottawa, On., Canada to examine, as a primary focus, the rehearsal mechanism involved in actively maintaining complex visual stimuli in memory for a brief period of time. Specifically, we are interested in determining whether an ACT-R model implementing a serial rehearsal strategy can account for human performance differences observed across two versions of a delayed match-to sample task. Versions of the delayed match-to sample task exist throughout the literature (Della Sala, Gray, Baddeley, Allamano, & Wilson, 1999; Warrington & James, 1967). In its most basic form, the task requires participants to encode a matrix grid pattern, rehearse it across a delay period, and compare it to a test grid. This task was selected for a number of reasons. First, its simplicity reduces many of the major confounds introduced by individual differences in strategy use, such as the tendency to recode presented visual information verbally. This notion is supported by the finding that articulatory suppression does not impair performance on similar tasks (Salway & Logie, 1995; Vandierendonck, Kemps, Fastame, & Szmalec, 2004). Second, the randomized nature of the grid pattern ensures that the structure does not become more familiar with time, so there is no expectation that implicit learning occurs resulting in faster and more efficient linking of environmental features to object-locations (Winkelholz & Schlick, 2006). Third, the instituted delay period between encoding and retrieval is longer than the time visual information is purported to survive in sensory memory (Phillips, 1974). This necessitates some form of active maintenance or rehearsal strategy. Finally, it is possible to create different versions of the selected task that vary only in complexity, such that a high-workload version contains more visual data to be encoded and rehearsed than a low- workload version. The present paper describes two ACT-R models of visual rehearsal. As a starting point, both models assume similar low-level processes, with absolute screen position used to encode visual stimuli in a serial fashion (i.e. objects are encoded as single chunks, without any Gestalt-type grouping). If model performance employing this serial encoding and rehearsal strategy does not fit the experimental data, it would suggest differences in encoding strategies (i.e., perceptual grouping of visual information) should be investigated in future work. The two models diverge in their implementation insofar as whether they represent each trial as an episode. While one model allows Abstract The current paper presents two ACT-R models of a delayed match-to sample task, and performs equivalence testing against human performance data to evaluate them. Success of an episodic model which avoids interference from previously encountered visual stimuli, and implements a serial search and rehearsal strategy lends insight into how individuals may encode, maintain and retrieve visual information. Keywords: ACT-R, visual memory, rehearsal Introduction ACT-R (Anderson & Lebiere, 1998) is a cognitive architecture that includes a theory of how higher-level processes interact with a visual system. ACT-R’s visual module identifies objects in the visual environment and through the use of buffers passes this information to the declarative memory module in the form of chunks. A chunk is a vector representation of individual properties, and in the case of visual information, is often represented with vector locations of the presented stimuli. Once visual information is represented in declarative memory, it can be retrieved according to task demands. In the past there has been little in the way of research which connects low-level visual processes with high-level cognition. Fortunately, this trend has been reversing over the last several decades and a wealth of research in the ACT-R community examines exactly how low-level processing constrains and influences visual encoding. These constraints include, among others: the time required for visual attentional shifts, the noise accompanying conjunction searches and the feature scale directing object recognition (Anderson, Matessa, & Lebiere, 1997). Despite strides towards understanding encoding constraints, most computational models of high-level visual processing continue to take visual representations for granted. Many of these models assume representations are deposited into declarative memory once they have been successfully encoded without accounting for intermediate processes between encoding and chunk formation. Often, for example, models do not account for rehearsal strategies that actively maintain complex visual stimuli in memory in order to prevent their decay. Extant models that do include visual rehearsal processes (e.g., Winkelholz & Schlick, 2006) do not do so as a primary research focus, and it is thus difficult to disentangle observed effects owing to rehearsal from those owing to other lines of inquiry. It is thus our aim

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.