Content Differences 1 Content Differences for Abstract and Concrete Concepts

The content of 36 abstract and concrete noun concepts was explored and compared in a feature generation task with 31 participants. Abstract concepts had significantly fewer entity properties, more properties expressing subjective experiences, and overall less specific features. Situation properties generated for abstract and concrete concepts differed in kind, but not in number. Abstract concepts were predominantly related to social aspects of situations, including agents. Abstractness emerges as a function of types and specificity of conceptual components. Systematic relations of conceptual content were revealed with context availability and imageability. The implications of the findings are discussed with respect to cognitive processes involving abstract concepts. Content Differences for Abstract and Concrete Concepts Abstract words, like indecision, difference, or consideration, are abundant in daily conversation. We use such words to describe, interpret, and explain events or circumstances in our social and physical environment. Abstract concepts contain a variety of categories each of which has sparked its own domain of research, including personality traits, behaviors and social cognition, emotions, cognitive processes, and events. As such, there is considerable interest in abstract concepts across disciplines within and beyond psychology. In spite of this, very little is known about their representation. Abstract entities are not experienced directly: they are not spatially constrained entities that we can see, or touch, or interact with. Instead, abstract concepts represent complex entities such as behaviors, scenes, orwords, like indecision, difference, or consideration, are abundant in daily conversation. We use such words to describe, interpret, and explain events or circumstances in our social and physical environment. Abstract concepts contain a variety of categories each of which has sparked its own domain of research, including personality traits, behaviors and social cognition, emotions, cognitive processes, and events. As such, there is considerable interest in abstract concepts across disciplines within and beyond psychology. In spite of this, very little is known about their representation. Abstract entities are not experienced directly: they are not spatially constrained entities that we can see, or touch, or interact with. Instead, abstract concepts represent complex entities such as behaviors, scenes, or subjective experiences. Thus, they are more aptly described as constructs of the mind which represent and structure experiences. The instances represented by an abstract concept vary considerably. This is reflected in ratings of contextual variety which tend to be higher for abstract concepts (Galbraith & Underwood, 1973). For example, difference can refer to sensory units, such as the result of adding a bit of salt to a soup or a height difference, or to mental units such as two opinions. At first glance, abstract (e.g., difference) and concrete entities (e.g., bucket) are easily distinguished. However, researchers have struggled to outline exactly in what respect they differ. The obvious difference is that only the concrete things are physical entities with defined by spatial boundaries. This ontological difference has important implications. We experience abstract and concrete things as qualitatively different – they seem to belong to different realms. As salient as this distinction may be, however, it mostly reveals what abstract concepts lack (i.e., physical substance, spatial boundaries). This leaves open the interesting question what abstract concepts do represent. The distinction of physical and nonphysical concepts is unsatisfying in a second regard: there are graded differences in concreteness. For example, people perceive scientist to be more abstract than milkbottle, and notion as more abstract than climate. Physicality, being a dichotomous variable, cannot easily explain the more subtle differences. What makes concrete concepts vary in concreteness? These questions motivate the research presented here. We explored the content of abstract versus concrete concepts by systematically analyzing participantgenerated features, to identify differences at a componential level. The analysis focused on the kinds of features that characterize people’s knowledge of concrete versus abstract concepts, the quantity of these features, and their relations to perceived concreteness and associated variables, such as imagery. Since a majority of concept processing models are based on the notion of conceptual components, this analysis is an important step in advancing our understanding of abstract concepts. Variables Associated with Concreteness Differences between abstract and concrete concepts have been studied in detail in form of concreteness effects on different kinds of processes. Generally, abstract concept processing is more challenging than concrete concept processing. Concreteness effects have been reported in a variety of tasks, including studies of learning, memory retrieval, comprehension, lexical decision, translation and semantic deficits. Imageability and context availability. The most influential theories that have been put forward to explain concreteness effects are the dual-coding theory and context availability theory. Both make distinct assumptions about storage of and access to abstract and concrete concepts in memory. Put in very simple terms, the dual-code theory assumes that concreteness effects are due to abstract concepts lacking a perceptual representation (Paivio, 1971; 1986), whereas the context-availability theory attributes concreteness effects to more relevant information stored in memory for concrete concepts, which facilitates their processing (Bransford & Johnson, 1972; Kieras, 1978). Ratings of Content Differences 3 imageability and context availability tend to be highly correlated with concreteness (Paivio, 1986; Rubin, 1980; Schwanenflugel, Harnishfeger, & Stowe, 1988). However, the foundations for imagery and the nature of context effects in memory are not well understood. The basis for both is likely linked to conceptual content. The present analysis may thus shed light on why abstract concepts evoke less or no imagery, and why less information may be stored in memory for abstract concepts. A conceptual content analysis may also be able to resolve the interesting recent finding that concreteness and imageability, and concreteness and context availability are not consistently correlated for the entire range of concreteness (Altarriba, Bauer, & Benvenuto, 1999; Wiemer-Hastings, Krug, & Xu, 2001). While high correlations were found for word samples that span the entire concreteness range, the correlation patterns differed when calculated separately for concrete versus abstract words. In particular, imageability ratings were correlated with concreteness ratings only for abstract words (N=155; r=0.57, p<0.05), but not for concrete words (N=100; r=0.02) (Altarriba, Bauer, & Benvenuto, 1999). In contrast, the association between rated context availability and concreteness was much stronger for concrete words (r=0.68, p<0.05) than for abstract ones (r=0.25, p<0.05). Thus, context availability increases strongly with concreteness of physical items, but only little with concreteness of nonmaterial items. In a similar study with 18 abstract and 18 concrete items, we replicated this pattern for context availability (Wiemer-Hastings, Krug, & Xu, 2001). However, in this sample, neither ratings for imageability (r=0.33) nor ratings for context availability (r=0.17) were significantly associated with concreteness of abstract concepts. Both were significantly correlated with concreteness of physical items (imageability: r=0.48, p<0.05; context availability: r=0.58, p<0.05). The findings suggest a qualitative difference between abstract and concrete concepts, which coincides with the dichotomous physicality dimension. This difference may be the main contributor to the strong correlations that are observed for samples spanning both concrete and abstract items. Further, it seems that the main factor underlying ratings of imageability and context availability is predominately varying among concrete concepts, and that imageability ratings may further be influenced by a separate factor that predominately varies for abstract concepts. These hypotheses can be tested through correlations between these measures and measures based on the features generated for abstract and concrete concepts. Ease of predication. Jones (1985) suggested in the context of dyslexia that imageability effects may be mediated by a semantic factor which he calls ease of predication, which is the ease with which individuals can access predicates of a concept. He found that rated imageability and ease of predication were highly correlated (r=0.88). However, this account is also unsatisfactory as an explanation for concreteness effects, because it is unclear how people rate ease of predication. It is possible that people base their ratings of ease of predication on the ease with which they can access an image (a related argument is made by de Mornay Davies & Funnell, 2000). Jones instructed his participants to ignore other factors than ease of predication in their ratings, but imagery may be an automatic process that mediates the participant ratings without their knowledge. Thus, ease of predication measures, too, require explanation at a fundamental conceptual level. Overview Knowledge of differences in the conceptual content due to varying concreteness could better explain concreteness effects, as well as provide a better understanding of imageability, context availability, ease of predication, and other variables related to concreteness. Further, a componential analysis of abstract concepts may allow for more sophisticated predictions in research involving such concepts, including encoding and recall, semantic relations, semantic access in priming studies, and categorical organization. What types of information may be involved in the representation of abstract entities like a thought or a goal? While our research is largely exploratory, we are building on a few earlier studies from which we have derived a few hypotheses regarding the content of abstract concepts. Psychological situations and action schemata. A large majority of abstract concepts are agent-centered; that is, they are strongly related to a person or group. This is true of large categories of abstract concepts, including emotions, cognition, actions and interaction, communication, character traits and attitudes, and others. Such abstract concepts would likely be associated with aspects of the internal and external context of the agent, such as behaviors, mental processes, another person, and states of affairs. The extent to which agents and agents’ contextual properties are part of the content of abstract concepts has not yet been subject of Content Differences 4 systematic investigation. However, related analyses suggest that they likely are. For example, Hampton (1981) investigated the prototypicality structure of abstract concepts using a property generation task. Many of the generated properties describe a social situation involving an agent and agent-related aspects of the situation. Hampton suggests that abstract concepts would commonly involve behaviors, agent characteristics such as goals, and other aspects of a situation, consistent with our reasoning. Similarly, researchers on personality traits have proposed agents, agent characteristics, and agent experiences as abstract concept components (Chaplin, John, & Goldberg, 1988). Research on psychological situations has revealed quite similar components as the properties of situation prototypes, such as agents, their behaviors, and dispositions, as well as physical attributes such as states, and objects (Cantor, Mischel, & Schwartz, 1982). On the basis of these studies, we predicted that descriptions of abstract concepts would show a strong focus on an agent and agent-related properties, including cognitive and emotional states and behaviors. Subjective experiences. There is evidence that abstract concepts are linked closely to subjective experiences that are only accessible through introspection. Introspective features have recently been proposed as a necessary component of abstract item representation (Barsalou, 1999). For example, a concept like decision cannot be understood without having experienced the mental process of weighing two or more choices against each other in terms of their good and bad sides. A large number of abstract concepts involve such elements of subjective experience, which may be mental, emotional, or even physiological in nature. Recent data also show that such components directly affect the perceived concreteness of an item. The larger the proportion of an item’s introspective elements is (i.e., goals, interpretations, evaluations, or emotions), the higher people judge its concreteness (Wiemer-Hastings, Krug, & Xu, 2001). Based on this finding, we predict that people will list significantly more properties for abstract concepts that express subjective experiences of an event or a situation. Such a finding would further support the finding that concreteness is in part a function of introspective components, and it would indeed be stronger because the measure for the proportion of introspective content would be based on properties generated by the participants, whereas in the above study, it was based on the experimenters’ coding schema for abstract concepts. Context variability. Abstract words occur in a larger variety of contexts, just as abstract entities can occur in a large variety of situations (Galbraith & Underwood, 1973). Contextual variability may be a factor underlying context availability differences for abstract and concrete concepts. Context may be less accessible in memory because the large number contexts stored with an abstract concept are competing for activation, or because contexts need to be actively generated during concept processing. Since abstract concepts represent instances from a larger variety of contexts, their features may place low constraint on context generation. A likely consequence of contextual variability is that abstract concepts involve components that are relatively unspecified, i.e., features that act like slots in a script or schema (Minsky, 1975; Schank & Abelson, 1977) that allow for a variety of specific attributes. For example, the restaurant script contains a slot for an agent bringing the menu to the table, for the type of food served in the restaurant, etc. Abstract concepts resemble scripts in that they refer to actions, goals, and agents at a general level. Thus, we expected that abstract concepts would be characterized by unspecific features. For example, an intuitive analysis suggests that difference occurs in contexts that contain any two or more items that can be compared on some dimension. At the extreme, perhaps some abstract concepts can be described as content-free schemata (Fiske & Taylor, 1991) in that they only specify abstract components and relations between them. Distinct versus graded differences. Abstract versus concrete items are typically defined dichotomously, as physical and non-physical items. Concreteness ratings are consistent with this view only to some extent. The consistent finding is that concreteness ratings tend to form a clearly bimodal distribution (Nelson & Schreiber, 1992; WiemerHastings, Krug, & Xu, 2001) with each mode centered in one of the two halves of the concreteness scale. This suggests a qualitative difference in representation which may be associated with distinct content elements that form part exclusively of either concrete or abstract concepts. This is in line with the finding that imageability and context availability correlate to different extents with concreteness ratings for abstract versus concrete concepts (Altarriba, Bauer, & Benvenuto, 1999). On the other hand, there are many degrees of concreteness, which suggests that there are additional concept characteristics that vary gradually across sections of the concreteness range. Thus, we expected to find semantic components which were exclusively named Content Differences 5 for abstract or concrete items, as well as components which varied gradually with concreteness. Participants in our experiments received a property generation task for 36 nouns spanning the entire scale of concreteness. Property generation tasks have proved useful in concept research. Perhaps the best known example of this is the prototype research by Rosch and colleagues (e.g., Rosch & Mervis, 1975). Our focus of interest was on proportions of different knowledge domains in the content of abstract and concrete concepts, and whether these varied in a graded fashion, or whether there were realms of experience exclusively applicable to just either concrete or abstract concepts. While participant-generated features unlikely represent exact conceptual content, they should accurately reflect these types of knowledge and systematic differences in knowledge domains across the concreteness dimension. Features provide insight into only one aspect of representation. We acknowledge that by focusing on features, we are ignoring more complex aspects such as relations among features, or organizations of features according to individuals’ assumptions about entities, which play an important role for concept processing. However, our hope is that providing this groundwork will facilitate more complex research of this kind.