Categories and Concepts
Category label and feature combination
How do category and feature information combine when making an inference? Imagine, for example, trying to determine whether it was safe to approach a dog (the inference). To make this determination, you can take into account whether the dog is snarling or wagging its tail (feature information) and whether the dog is a pit bull or golden retriever (category information). A number of previous studies have suggested that category information is privileged, that is, takes precedence over feature information. Few (if any) studies, however, have looked at the form of interaction between these two sources of information. Borrowing from the information integration literature, we compared two ways that feature and category information could combine: optimally, i.e., a Bayesian combination, and linearly, i.e., a weighted linear combination. We find, contrary to past research, that participants seems to treat feature and category information equally and combine them in an optimal fashion.
Decision-making in categorization
The generalized context model (GCM) is arguably the most successful model of categorization. The model implicitly breaks categorization into two parts: a comparison phase, in which a to-be-categorized item is compared to all stored exemplars from the competing categories, and a decision phase, in which a category decision is made based on the outcome of the comparison phase. The decision phase is based on the Luce choice rule. This decision rule has been shown to be inadequate in a number of judgment and decision-making (JDM) experiments. For example, suppose there are two fairly dissimilar widgets, A and B, on the market. A company is considering producing a new kind of widget, C, that is similar to A and dissimilar to B. A common finding is that, under certain constraints, the introduction of C will hurt the market share of A more than B. In psychological terms, people are less likely to choose A than B after the introduction of C. This change in preferential choice is called the similarity effect. Different similarity relations between A, B, and C will produce different shifts in preferential choice. In this line of research, we replicated the problematic JDM phenomena in a categorization setting, suggesting ways in which the decision component of the GCM needs to be updated.
Context and meaning in noun-noun combinations
What is the meaning of the noun-noun combination "dictionary treatment"? Coming up with a meaning is probably difficult. What if I were to put this combination in context? "When I came home late last night, my father yelled at me with words I had to look up. He gave me the dictionary treatment last night." Putting the combination in context probably made the task easier. The goal of this project is to explore how the effect of context in such meaning creation over time. That is, does context ameliorate the task immediately or does it take some time to incorporate context? An eye-tracking experiment examined processing of novel noun-noun combinations (e.g., dictionary treatment, olive signals). Readers’ eye movements were monitored as they read sentences containing such combinations, or control sentences containing easy-to-interpret adjective-noun combinations. A preceding context sentence either did or did not support a specific interpretation of the critical noun-noun combination. The earliest eye-tracking measures revealed processing difficulty for the critical noun-noun combinations, which was not modulated by context. On later measures, however, the potentially helpful context did alleviate difficulty. The results suggest a two-stage model in which a noun-noun combination is initially interpreted locally, i.e., without taking into account the extra-sentential context, but that this context may later play a critical role.
Conceptual reinterpretation
Perception of the visual world depends on neural activity across multiple brain regions (Desimone & Duncan, 1995). The visual system needs to coordinate the activity across these regions in order to reach a stable interpretation of the visual world (Reynolds & Desimone, 1999). Recent research suggests that switching attention from one aspect of a scene to another, for example, from color to motion, requires dissolving the current stable state and the creation of a new stable state. Serences & Yantis (2006) suggested that such switches are initiated by a “transient control signal that nudges the visual system from one coherent state to another” (p. 38) and have localized this signal to the medial superior parietal cortex. Importantly, these researchers claim that the switching signal is independent of the type of switch required. If that is the case, the switching signal should also play an important role for higher-level cognitive switches that require moving from one coherent state to another. The goal of this project is to determine whether the same transient signal that is used to switch the perceptual system from one coherent state to another also coordinates switching at a conceptual level. To test this idea, participants were be asked to switch conceptual interpretations of simple visual line drawings while in an fMRI scanner. A pattern of brain activity that replicates the previous perceptual work would suggest that the transient signal is also involved in higher-level switches. A different pattern would indicate that switches at the cognitive and perceptual levels do not involve the same switching mechanism.
Decision-tree models of categorization
The goal of this project is to develop decision-tree models of categorization. In a decision-tree model, it is assumed that people make a classification by going through a sequential series of rule-based decisions. In particular, the goal was to use decision-tree models to provide a quantitative account of categorization response times, choice proportions, and typicality judgments at the individual-participant level. The decision trees explained a very high proportion of variance in the data and compare favorably with two leading exemplar models. A process tracing method called the “four-questions game” (Sayeki, 1969) was used in a post-test phase to identify a decision tree for each participant. Cross-validation tests showed that the decision-tree models had a good predictive accuracy—an important capability that other models of categorization have yet to demonstrate.
Emotion and categorization
How does emotion effect categorization? Emotion can influence many of the factors that determine classification behavior. People in a positive mood, for example, are both better able to differentiate between and integrate unusual and diverse information (Murray, et al 1990; Vosburg, 1998). We are taking a modeling approach to determine the specific classification processes influenced by emotion. In particular, we are applying the generalized context model (GCM) of categorization to classification data collected under various moods. Analysis of the GCM will allow us to determine, for example, whether emotion modulates the likelihood that a person will attend to multiple stimulus features, the overall similarity between stimuli, or the degree to which people respond deterministically.
Model Selection
Grouped or individual data
A model of a psychological phenomenon is a mathematical formalization of a theory which can produce precise experimental predictions. The main goal of this research is to determine whether averaged group or individual subject model analyses better uncover the “correct” model for an experiment. Because it is well known that averaging can distort the form of data, there has been a trend away from modeling averaged data and towards modeling individual data. However, there are conditions under which fitting averaged data may recover the correct model more often than fitting individual data. For example, if there is little data per subject, the noise associated with fitting individual subjects may be worse than the distortion created by averaging across subjects. Using a simulation technique in which the correct model is known, we are exploring how these analyses interact with various experimental factors and modeling techniques such as number of subjects, number of trials/condition, model selection criterion, parametric variation across subjects, and the choice of competing models.
Models of remember-know
In a remember-know memory task, participants are asked to determine which of a set of stimuli they previously experienced and whether they remember, i.e., give specific details about, or know, i.e., have only a general sense that they remember the stimulus, the ones they recognize. Remember-know judgments provide additional information in recognition memory tests, but the nature of this information and the attendant decision process is in dispute. Competing models propose that remember judgments reflect a sum of familiarity and recollective information (the one-dimensional model), are based on a difference between these strengths (STREAK), or are purely recollective (the dual-process model). A choice among these accounts is sometimes made by comparing the precision of their fits to data, but this strategy may be muddied by differences in model complexity: some models that appear to provide good fits may be better able to mimic data produced by other models. To evaluate this possibility, we simulated data with each of the models in each of three popular remember-know paradigms, then fit those data with each of the models. We found that the one-dimensional model is generally less complex than the others; despite this handicap, it dominates the other models as the best-fitting model. For both reasons, the one-dimensional model is preferred. In addition, we found that some empirical paradigms are ill-suited for distinguishing among models. For example, data collected by soliciting remember/know/new judgments, i.e., the trinary task, provide particularly weak grounds for distinguishing the models.
Perception
Feature induction
This project utilizes a number of new experimental and mathematical techniques to tackle one of the most fundamental questions of psychology: what are the basic building blocks, the features, of perception? That is, what are the parts of an object that are treated as unitary wholes when recognizing or discriminating objects? For example, consider a task classifying a visual target presented in pixel noise as a ‘P’ or a ‘Q’. The features may correspond to particular shapes of the target letters. Two such features for ‘P’, for example, might be a vertical line and upper right-facing curve. The decision may be encoded in terms of particular values of such features, and an appropriate combination of these values may determine how the expression is perceived. The empirical work is based on the response-classification technique. On each of numerous trials, an observer's task is to identify a stimulus presented in a noise field. On many trials, the noise will cause observers to make judgment errors. To determine which pixels cause the observer to make a particular response, the noise fields (without the actual stimulus) from trials on which the observer gives a particular response are summed and subtracted from the result of a similar sum for the other choices. The resulting map is called a classification image and shows which pixel locations the observer used to make a classification judgment. The classification image, however, is a map of all of the pixels used to make a response and does not discriminate between the subsets of pixels that define a feature. The next step is to define a generative model of object recognition, e.g., a mathematical formalization of how pixels are grouped into features and then how the features are used to make a response. The problem now becomes one of recovering the generative model, including the features, from the classification image data. On the surface, this induction step is an enormously computationally extensive problem. Recent advances in techniques of data manipulation, however, have made this computation feasible. We have used two main modeling approaches: Gaussian mixture models and Bayesian networks.
Relative mass
How do people "see" mass? Unlike velocities and angles, mass cannot be observed visually. The central research issue here is how people use visually presented information to determine mass. In particular, it has been found that observers have some ability to make relative mass judgments after viewing two-ball collisions. On a typical trial of a colliding balls experiment, two (computer-simulated) balls roll across a flat surface, collide, and then roll away from each other. A typical observer’s task is to determine which of the two balls is heavier. We are exploring the potential contributions of invariants, heuristics, and exemplars in the perception of dynamic properties as realized in the colliding balls paradigm. The invariant approach assumes that people can learn to detect complex visual patterns that reliably specify which ball is heavier. The heuristic approach assumes that observers only have access to simple motion cues. The exemplar-based approach assumes that people store particular exemplars of collisions in memory, which are later retrieved to perform the task. We have compared formal mathematical models of these theories. One line of research suggests that people rely on exemplar processing when the task involves relatively few, similar collisions. Observers switch to invariant processing when there are large numbers of dissimilar collisions. When confidence in the invariant is low, however, observers revert to the errorful strategies they used before training. Another line of research, using a newly developed technique of psychological Markov chain Monte Carlo sampling (Sanborn & Griffiths, 2008), explores participants’ perceptions of different collision mass ratios. The results reveal inter-participant differences and a qualitative distinction between the perception of 1:1 and 1:2 ratios. The results strongly suggest that participants’ perceptions of 1:1 collisions are described by simple heuristics. The evidence for 1:2 collisions favors heuristic perception models that are sensitive to the sign but not the magnitude of perceived mass differences.
Information integration
The goal of this research is to uncover the fundamental mechanisms by which a judgment is fashioned from multiple sources of information. For example, how does a physician formulate a diagnosis from numerous test results? Of particular theoretical and historical importance is whether information is combined in a way that maximizes some performance measure, such as the probability of a correct diagnosis. Such questions of optimality permeate the judgment and decision making, categorization, memory, and perception literatures, among others. Typically, optimal and suboptimal outcomes occur in quite different tasks. For example, poor combination of information is common in verbal probability problems and good combination is often found in category judgments of colors varying in hue and saturation. This research explores how the nature of the system performing the combination affects the manner in which the information is combined. The working hypothesis is that people are good (closer to optimal) at combining information when it can be done automatically, but that conscious, cognitive combination results in integration that is farther from optimal. We explored this hypothesis in series of information experiment using perceptual stimuli. In one set of conditions, the stimuli were designed to promote automatic processing. In another set of conditions, this automatic processing was disrupted. Our results suggest that (1) people can indeed use different modes of processing, e.g., automatic and intentional, to integrate information, (2) that automatic processing results in integration that is closer to optimal, but cannot always be used, and (3) that different stimuli (quantitative, perceptual, etc.) can facilitate different processing modes, leading to the conflicting findings.
Emotion and perception
How does emotion affect perception? People have an uncanny ability to detect patterns where they do not exist. The goal of this project is to determine whether this ability is modulated by emotion. The basic idea is that people in a threatened state may be more likely to "impose" patterns upon the world. We are employing a signal detection analysis to determine whether emotion changes perception or introduces a decision bias.
|
