Edward J. Wisniewski

When concepts combine

I present a computational level account of how people combine concepts, and I use this account to evaluate current models of conceptual combination. Constrained by this account, I then provide an algorithmic level description of how people combine concepts. The algorithmic level account highlights the importance of two additional processes (comparison and construction) in explaining how some concepts combine and change. I then show that the interpretation of nominal metaphors involves these processes as well. Current approaches to metaphor understanding emphasize the importance of one or the other of these processes, but not both.

On the interaction of theory and data in concept learning

Standard models of concept learning generally focus on deriving statistical properties of a category based on data (i.e., category members and the features that describe them) but fail to give appropriate weight to the contact between peoples intuitive theories and these data. Two experiments explored the role of peoples prior knowledge or intuitive theories on category learning by manipulating the labels associated with the category. Learning differed dramatically when categories of childrens drawings were meaningfully labeled (e.g., “done by creative children”) compared to when they were labeled in a neutral manner. When categories are meaningfully labeled, people bring intuitive theories to the learning context. Learning then involves a process in which people search for evidence in the data that supports abstract features or hypotheses that have been activated by the intuitive theories. In contrast, when categories are labeled in a neutral manner, people search for simple features that distinguish one category from another. Importantly, the final study suggests that learning involves an interaction of peoples intuitive theories with data, in which theories and data mutually influence each other. The results strongly suggest that straight-forward, relatively modular ways of incorporating prior knowledge into models of category learning are inadequate. More telling, the results suggest that standard models may have fundamental limitations. We outline a speculative model of learning in which the interaction of theory and data is tightly coupled. The article concludes by comparing the results to recent artificial intelligence systems that use prior knowledge during learning.

Similar and Different: The Differentiation of Basic-Level Categories

Categories in the middle level of a taxonomic hierarchy tend to be highly differentiated in that they have both high levels of within-category similarity and low levels of between-category similarity. Research on similarity reveals a distinction between pairs of categories that are seen as dissimilar because they have few commonalities and pairs that are seen as dissimilar because they have many psychologically relevant alignable differences. The authors suggest that the low between-category similarity proposed for neighboring basic-level categories is actually a matter of having many psychologically relevant differences. In contrast, the low between-category similarity of superordinates is a result of their having few commonalities. The authors evaluate this claim in 4 experiments using a variety of natural stimuli and converging measures. The data support the importance of aliguable differences for distinguishing between pairs of basic-level categories.

Prior knowledge and functionally relevant features in concept learning.

Empirical learning models have typically focused on statistical aspects of features (e.g., cue and category validity). In general, these models do not address the contact between peoples prior knowledge that lies outside the category and their experiences of the category. A variety of extensions to these models are examined, which combine prior knowledge with empirical learning. Predictions of these models were compared in 4 experiments. These studies contrasted the cue and category validity of features with peoples prior knowledge about the relevance of features to the functions of novel artifacts. The findings suggest that the influences of knowledge and experience are more tightly integrated than some models would predict. Furthermore, relatively straightforward ways of incorporating knowledge into an empirical learning algorithm appear insufficient (e.g., use of knowledge to weight features by general relevance or to individually weight features). Other extensions to these models are suggested that focus on the importance of intermediary features, coherence, and conceptual roles.

What makes a man similar to a tie? Stimulus compatibility with comparison and integration

We argue and show that different properties of stimuli are compatible with different types of processing. Specifically, object pairs from the same taxonomic category (e.g., chair-bed) tend to be alignable and thus compatible with comparison, whereas object pairs that play different roles in thematic relations (e.g., chair-carpenter) tend to be nonalignable and compatible with integration. Using object pairs that varied orthogonally in alignability and thematic relatedness, we demonstrated that stimulus compatibility modulates processing and affects the outcomes of tasks that are currently believed to involve only comparison (similarity ratings, Experiment 1; listing commonalities and differences, Experiment 2) or only integration (thematic relatedness ratings, Experiment 3). Our findings and others that we have reviewed suggest that: (1) many cognitive tasks involve both comparison and integration, and (2) the relative influence of each process is modulated by an interplay between the task-appropriate and the stimulus-compatible process. We believe that single-process models should be extended to take this interplay into account.

A structural account of global and local processing.

The order of processing, whether global forms are processed prior to local forms or vice versa, has been of considerable interest. Many current theories hold that the more perceptually conspicuous form is identified first. An alternative view is presented here in which the stuctural relations among elements are an important factor in explaining the relative speeds of global and local processing. We equated the conspicuity of the global and local forms in three experiments and still found advantages in the processing of global forms. Subjects were able to process the relations among the elements quickly, even before the elements themselves were identified. According to our alternative view, subjects created equivalence classes of similar and proximate local elements before identifying the constituent elements. The experiments required subjects to decide whether two displays were the same or different, and consequently, the results are relevant to work in higher-level cognition that stresses the importance of comparison processes (e.g., analogy and conceptual combination). We conclude by evaluating related work in higher-level cognition in light of our findings.

Typical Versus Atypical Unpacking and Superadditive Probability Judgment

Probability judgments for packed descriptions of events (e.g., the probability that a businessman does business with a European country) are compared with judgments for unpacked descriptions of the same events (e.g., the probability that a businessman does business with England, France, or some other European country). The prediction that unpacking can decrease probability judgments, derived from the hypothesis that category descriptions are interpreted narrowly in terms of typical instances, is contrasted to the prediction of support theory that unpacking will generally increase judged probabilities (A. Tversky & D. J. Koehler, 1994). The authors varied the typicality of unpacked instances and found no effect of unpacking with typical instances (additivity) and a negative effect with atypical instances (superadditivity). Support theory cannot account for these findings in its current formulation.

Harpoons and long sticks: the interaction of theory and similarity in rule induction

Publisher Summary This chapter presents two studies that examined the roles of theoretical expectations and empirical evidence in rule induction. These studies produced several major findings. First, theoretical expectations strongly affect the kinds of rules that people construct for a given category. These rules are qualitatively different from those constructed by people without such expectations. The theoretical expectations activate hypotheses about abstract features that could be true for a category. Second, theoretical knowledge closely interacts with empirical evidence, and these types of knowledge mutually influence each other. In particular, the two sources of knowledge interact to determine hypothetical features upon which induction operates. They also mutually influence the types of explanations that people construct for a category. The chapter describes the way in which a learning system closely integrates these sources of knowledge. The view that peoples theoretical expectations closely interact with experience generally has not been emphasized in machine learning models that integrate these two sources of knowledge.

Frequency of relation type as a determinant of conceptual combination: a reanalysis.

C. L. Gagne and E. J. Shoben (1997) proposed that concepts are combined via external relations and that lexical entries include information about which relations are frequent for every modifying noun. As evidence for this view, they showed that relations associated with the modifier affected the interpretation of combinations in several studies in which subjects had to decide whether the combinations were sensible. The authors evaluated the methods and stimuli used in Gagne and Shobens experiments and present findings suggesting that the effect of relation frequency is likely due to differences between the familiarity and plausibility of different combinations. Although relation frequency could be involved in conceptual combination, the authors concluded that better evidence is needed for this variable, controlling for other more general differences between the combinations.

Property instantiation in conceptual combination

In four experiments, I examined how a property in one concept is transferred to a second concept during conceptual combination. The results suggest that people instantiate properties: that is, they use a specific representation of a property in the modifier concept to construct a new version of that property that is specific to the combination. If people are instantiating properties, then the modifier property should match its counterpart in the combination to the extent that the modifier and head noun are similar. This observation leads to a variety of predictions (supported by the experiments) about interpretations of similar and dissimilar combinations and about plausibility, preference, and similarity judgments associated with such interpretations. The results argue against an alternative view of transfer that posits that, in general, abstract representations of properties are copied from one concept to another. In this paper, I describe various processing accounts of instantiation and discuss the implications of the instantiation view for theories of metaphor, conceptual combination, and induction.