Kevin Dunbar

On the control of automatic processes: A parallel distributed processing account of the Stroop effect.

Traditional views of automaticity are in need of revision. For example, automaticity often has been treated as an all-or-none phenomenon, and traditional theories have held that automatic processes are independent of attention. Yet recent empirical data suggest that automatic processes are continuous, and furthermore are subject to attentional control. A model of attention is presented to address these issues. Within a parallel distributed processing framework, it is proposed that the attributes of automaticity depend on the strength of a processing pathway and that strength increases with training. With the Stroop effect as an example, automatic processes are shown to be continuous and to emerge gradually with practice. Specifically, a computational model of the Stroop task simulates the time course of processing as well as the effects of learning. This was accomplished by combining the cascade mechanism described by McClelland (1979) with the backpropagation learning algorithm (Rumelhart, Hinton, & Williams, 1986). The model can simulate performance in the standard Stroop task, as well as aspects of performance in variants of this task that manipulate stimulus-onset asynchrony, response set, and degree of practice. The model presented is contrasted against other models, and its relation to many of the central issues in the literature on attention, automaticity, and interference is discussed.

Dual Space Search During Scientific Reasoning

The purpose of the two studies reported here wos to develop an integrated model of the scientific reasoning process. Subjects were placed in a simulated scientific discovery context by first teaching them how to use an electronic device and then asking them to discover how a hitherto unencountered function worked. To do this task. subjects had to formulate hypotheses bas’ed on their prior knowledge, conduct experiments, and evaluate the results of their experiments. In t,he first study, using 20 adult subjects, we identified two main strategies that subjects used to generate new hypotheses. One strategy was to scorch memory and the other was to generalize from the results of previous experiments. We described the former group as searching an hypothesis space, and the latter as searching on experiment space. In a second study, with 10 adults, we investigated how subjects search the hypothesis space by instructing them to state all the hypotheses that they could think of prior to conducting any experiments. Following this phase, subjects were then allowed to conduct experiments. Subjects who could not think of the correct rule in the hypothesis generation phase discovered the correct rule only by generalizing from the results of experiments in the experimentol phase. Both studies provide support for the view that scientific reasoning can be characterized as search in two problem spaces. By extending Simon and Lea’s (1974) Generalized Rule Inducer, we present a general model of Scientific Discovery as Dual Search (SDDS) that shows how search in two problem spaces (an hypothesis space and an experiment space) shapes hypothesis generation, experimental design, and the evaluation of hypotheses. The model also shows how these processes interact with each other. Finally, we interpret earlier findings about the psychology of scientific reasoning In terms of the SDDS model.

Training and Stroop-Like Interference: Evidence for a Continuum of Automaticity

Three experiments varied the extent of practice in an analog of the Stroop color-work task. Each experiment involved four phases: (a) baseline naming of four familiar colors, (b) training in consistently naming four novel shapes by using the names of the same four colors, (c) naming the colors when they appeared in the form of the shapes, and (d) naming the shapes when they appeared in color. In Experiment 1, with up to 2 hr of training in shape naming, colors were named much faster than shapes. Interference was observed only in Phase 4. In Experiment 2, with 5 hr of training, shape naming sped up, but was still slower than color naming. Nevertheless, there was symmetrical interference in Phases 3 and 4, and this persisted 3 months later without further training. Experiment 3 replicated this pattern and then extended practice to 20 hr, by which time shape and color naming were equally rapid. After 20 hr, interference appeared only in Phase 3, reversing the original asymmetry. The overall pattern is inconsistent with a simple speed of processing account of interference. The alternative idea of a continuum of automaticity--a direct consequence of training--remains plausible, and the implications of this perspective are considered.

Heuristics for Scientific Experimentation: A Developmental Study.

Scientific discovery involves search in a space of hypotheses and a space of experiments. We describe an investigation of developmental differences in the search constraint heuristics used in scientific reasoning. Sixty-four subjects (technically trained college students, community college students with little technical training, 6th graders, and 3rd graders) were taught how to use a programmable robot. Then they were presented with a new operation, provided with a hypothesis about how it might work, and asked to conduct experiments to discover how the new operation really did work. The suggested hypothesis was always incorrect, as subjects could discover if they wrote informative experiments, and it was either plausible or implausible. The rule for how the unknown operation actually worked was either very similar or very dissimilar to the given hypothesis. Children focused primarily on plausible hypotheses, conducted a limited set of experiments, designed experiments that were difficult to interpret, and were unable to induce implausible (but correct) hypotheses from data. Adults were much better than children in discovering implausible rules. The performance deficits we found were not simply the result of childrens inadequate encoding or mnemonic skills. Instead, the adults appear to use domain-general skills that go beyond the logic of confirmation and disconfirmation and deal with the coordination of search in two spaces: a space of hypotheses and a space of experiments.

A Horse Race of a Different Color: Stroop Interference Patterns With Transformed Words

Four experiments investigated Stroop interference using geometrically transformed words. Over experiments, reading was made increasingly difficult by manipulating orientation uncertainty and the number of noncolor words. As a consequence, time to read color words aloud increased dramatically. Yet, even when reading a color word was considerably slower than naming the color of ink in which the word was printed, Stroop interference persisted virtually unaltered. This result is incompatible with the simple horse race model widely used to explain color-word interference. When reading became extremely slow, a reversed Stroop effect--interference in reading the word due to an incongruent ink color--appeared for one transformation together with the standard Stroop interference. Whether or not the concept of automaticity is invoked, relative speed of processing the word versus the color does not provide an adequate overall explanation of the Stroop phenomenon.

Frontopolar cortex mediates abstract integration in analogy

Integration of abstractly similar relations during analogical reasoning was investigated using functional magnetic resonance imaging. Activation elicited by an analogical reasoning task that required both complex working memory and integration of abstractly similar relations was compared to activation elicited by a non-analogical task that required complex working memory in the absence of abstract relational integration. A left-sided region of the frontal pole of the brain (BA 9/10) was selectively active for the abstract relational integration component of analogical reasoning. Analogical reasoning also engaged a left-sided network of parieto-frontal regions. Activity in this network during analogical reasoning is hypothesized to reflect categorical alignment of individual component terms that make up analogies. This parieto-frontal network was also engaged by the complex control task, which involved explicit categorization, but not by a simpler control task, which did not involve categorization. We hypothesize that frontopolar cortex mediates abstract relational integration in complex reasoning while parieto-frontal regions mediate working memory processes, including manipulation of terms for the purpose of categorical alignment, that facilitate this integration.

How analogies are generated: The roles of structural and superficial similarity

Laboratory studies of analogical reasoning have shown that subjects are mostly influenced by superficial similarity in the retrieval of source analogs. However, real-world investigations have demonstrated that people generate analogies using deep structural features. We conducted three experiments to determine why laboratory and real-world studies have yielded different results. In the first two experiments, we used a “production paradigm” in which subjects were asked to generate sources for a given target. Results show that the majority of the analogies that were generated displayed low levels of superficial similarity with the target problem. Moreover, most of the analogies were based on complex underlying structures. The third experiment used a “reception paradigm” methodology. The subjects had to retrieve predetermined sources instead of generate their own. In this case, retrieval was largely constrained by surface similarity. We conclude that people can use structural relations when given an appropriate task and that this ability has been underestimated in previous research on analogy.

Connecting Long Distance: Semantic Distance in Analogical Reasoning Modulates Frontopolar Cortex Activity

Solving problems often requires seeing new connections between concepts or events that seemed unrelated at first. Innovative solutions of this kind depend on analogical reasoning, a relational reasoning process that involves mapping similarities between concepts. Brain-based evidence has implicated the frontal pole of the brain as important for analogical mapping. Separately, cognitive research has identified semantic distance as a key characteristic of the kind of analogical mapping that can support innovation (i.e., identifying similarities across greater semantic distance reveals connections that support more innovative solutions and models). However, the neural substrates of semantically distant analogical mapping are not well understood. Here, we used functional magnetic resonance imaging (fMRI) to measure brain activity during an analogical reasoning task, in which we parametrically varied the semantic distance between the items in the analogies. Semantic distance was derived quantitatively from latent semantic analysis. Across 23 participants, activity in an a priori region of interest (ROI) in left frontopolar cortex covaried parametrically with increasing semantic distance, even after removing effects of task difficulty. This ROI was centered on a functional peak that we previously associated with analogical mapping. To our knowledge, these data represent a first empirical characterization of how the brain mediates semantically distant analogical mapping.

How Scientists Think in the Real World: Implications for Science Education

Abstract Research on scientific thinking and science education is often based on introspections about what science is, interviews with scientists, prescriptive accounts of science, and historical data. Although each of these approaches is valuable, each lacks some key components of what scientists do, which makes it difficult to determine what scientists are being trained for and what essential thinking and reasoning tools they must have. The research reported herein sought to determine the cognitive processes underlying reasoning in science using two approaches. The first is to bring participants into the laboratory and give them scientific problems to work on. The second is to investigate real scientists as they work at their own problems. Both approaches make it possible to propose several thinking and reasoning strategies that are conducive to making discoveries. Both also make it possible to understand some of the basic cognitive mechanisms underlying scientific thinking.

Priming, analogy, and awareness in complex reasoning

The mechanisms by which a concept used in solving one complex task can influence performance on another complex task were investigated. We tested the hypothesis that even when subjects do not spontaneously make an analogy between two domains, knowledge of one domain can still spontaneously influence reasoning about the other domain via the mechanism of priming. Four groups of subjects (two experimental and two control) were given a simulated biochemistry problem on Day 1 and a simulated molecular genetics problem on Day 2. For the two experimental groups, the solution to the biochemistry problem involved inhibition. For the two control groups, the solution did not involve inhibition. On Day 2, all subjects received the same version of the molecular genetics problem in which the solution involved the concept of inhibition. Subjects in the experimental conditions were more likely to attain the correct answer, to propose inhibition, and to propose inhibition early in the problemsolving session than were subjects in the control conditions. However, subjects in the experimental conditions made no reference to the biochemistry problem either in their verbal protocols or in a posttask questionnaire. The results are interpreted as demonstrating that an implicit process—priming— can make old knowledge available for current problem solving.