Categories of environmental scenes

COGNITIVE

PSYCHOLOGY

15, 121- 149 (1983)

Categories BARBARATVERSKY

of Environmental AND KATHLEEN Stanford

Scenes HEMENWAY

University

Environmental scenes are the settings in which human action occurs; since they constrain behavior, they are of interest to social, personality, and environmental psychologists. Scenes can also be viewed as a spatial generalization of objects, as well as the spatial contexts in which objects appear. As such, they are studied in perception and memory. Previous approaches to characterizing environments have relied on scaling techniques to yield a manageable number of dimensions or attributes by which environments can be compared. In contrast, the present research demonstrates development of a taxonomy of kinds of environmental scenes, where perceived attributes are obtained as a byproduct. A basic or preferred level of categorization in the taxonomy is also identified, based on measures of cognition, behavior, and communication. The basic level, for example, school, home, beach, mountains, corresponds to the level commonly used in the study of scene schemas in perception, memory, and environmental psychology, as well as to the level apparently most useful in other domains of knowledge concerned with environments, for example, architecture and geography.

INTRODUCTION Basic Level Categorization In the fifties, there was a popular satire of noncommunicative teenagers called “ ‘Where did you go?’ ‘Out.’ ‘What did you do?’ ‘Nothing”’ (Smith, 1958). Their parents, of course, expected more informative answers, such as, “Well, first I went to school, then to the beach for a run, then to a restaurant for a bite to eat, and then I came home.” But their parents did not expect answers more informative than that, either (unless they were suspicious), so they probably were not interested in which beach or what kind of restaurant. Why school, beach, restaurant, home? Why not just “Out,” and why not answers that are more specific? A similar question was posed by Roger Brown (1958), in his classic paper “How shall a thing be called?” Why is it that we refer to the thing we sit on as a chair, and not as desk chair or a piece offurniture; to our This research was supported by NSF Grant BSN 8002012 and by NIMH Grant MH-34248 to Stanford University. Kathleen Hemenway was supported by an NSF Fellowship. She is now at Bell Laboratories, Piscataway, NJ. The authors are grateful to Roslyn Banish, Edward Baer, and Richard Gans for generously providing photographs, and especially to Homer James, for turning his weekend outings into photography expeditions. We also appreciate helpful comments from Roger Brown, Herbert H. Clark, and Gregory Murphy. 121 OOlO-0285/83/010121-29$07.50/O Copyright 0 1983 by Academic Press, Inc. All rights of reproduction in any form reserved.

122

TVERSKY

AND

HEMENWAY

midafternoon snack as an apple, and not apippin apple or a piece offruir? Brown argued persuasively against some of the more obvious explanations for this phenomenon. True, we generally prefer short labels to long ones, but length in turn is dependent on frequency, as reflected in Zipf s law. However, the preferred level of naming systematically violates frequency; for example, pineapple instead offruit, hammer instead of tool. The preferred level of naming corresponds to the level that parents choose in describing the world to children. Since children are notoriously concrete, perhaps a concrete level is selected for their benefit. But no, our labels are not as concrete as they could be, they seem to be at an intermediate level. We normally refer to green seedless grapes as simply grapes, to a ballpeen hammer as just a hammer. After rejecting these explanations, Brown argued that the level selected seems to be the level that is most “useful” in most contexts, in picking out the referent from the scene. Naturally, what is useful may change, depending on the context and on mutual knowledge or common ground (Clark & Carlson, 1981). Both psychologists and anthropologists have sought answers to Brown’s question (Berlin, Breedlove, & Raven, 1973; Berlin, 1978; Brown, 1977, 1979; Rosch, Mervis, Gray, Johnson, & Boyes-Braem, 1976; Rosch, 1978). The work of Rosch and her collaborators has been especially influential in providing insight into what makes one level of description more “useful” than other levels. Objects may be grouped into categories varying in level of abstraction or degree of inclusiveness, for example, pippin apple, or an apple, or fruit. In general, people refer to objects at what Rosch et al. termed the basic level, for instance, chair or apple. This basic level of categorization is a compromise between two opposing goals of categorization (Rosch, 1978). On the one hand, we want categories to be informative, so that knowing what category a thing belongs to will reveal many of its attributes. Knowing that your car is a 1975 blue Datsun 210 station wagon is more informative than knowing that it is a car. The more specific the category, the more informative. On the other hand, we want to minimize the number of different categories we have to deal with, and, particularly, to minimize categories based on irrelevant distinctions. The more general the categories, the fewer their number, and the fewer distinctions needed to categorize. If you offer to give me a lift, all I need to know is that you have a car. Although the informativeness of a category increases with its specificity in a taxonomy, it has been shown in a variety of object taxonomies that some shifts in taxonomic level yield relatively larger gains in informativeness than others. In particular, going from the superordinate level, for example, furniture, fruit, to the basic level, for example, chair, apple, of categorization yields a relatively large increase in informativeness, whereas going from the basic level to the subordinate level, for example, desk chair, pippin apple, yields only a

CATEGORIES

OF SCENES

123

small addition to informativeness (Rosch et al., 1976). Informativeness is reflected in people’s perception or representations of the objects as well as in people’s behavior toward the objects. That is, people perceive that members of superordinate categories have very few common attributes and very few common human behaviors, but that members of basic level categories have many attributes and behaviors in common. Members of more specific, subordinate categories share a few more attributes and behaviors than basic level categories, but the increase is relatively slight. Many other experimental operations converge on the basic level. It is the highest level of abstraction where people can identify an averaged shape, the level yielding fastest naming by adults, and the level first named by children (Rosch et al., 1976). Objects at the basic level are typically named by simple, short labels, termed primary lexemes (Berlin et al., 1973). In the development of a taxonomy of a knowledge domain, basic level terms are generally differentiated first, and only later, more specific and more general terms enter the lexicon (Berlin et al., 1973). Thus, measures of our perception of stimuli, our responses to those stimuli, and our communication about them all converge on the same basic level. Characterization of Environments Let us now return to our opening example. Why are the parents of teenagers interested in where they were anyway? Presumably because knowing where they were would reveal what they had been doing. At the library, studying. At a restaurant, eating. This intuition, that the environment accounts for a considerable portion of our behavior has been substantiated by personality, social, organizational, and environmental psychologists alike (e.g., Barker, 1968; Bandura, 1977; Mischel, 1968, 1973; Moos, 1973). Evidence that behavior settings play a substantial role in determining behavior has led to numerous attempts to characterize environments (see reviews by Craik, 1973; Fredricksen, 1972; Moos, 1973; Russell & Ward, 1982; Stokols, 1978; Ward & Russell, 1981). Many of these attempts to characterize environments have applied techniques like multidimensional scaling and factor analysis to uncover a manageable set of dimensions or factors by which to compare environments. In other words, researchers have sought to uncover the “personalities” of environments by applying the techniques of individual difference research. Many of the reviewers of these attempts have lamented the lack of consensus across investigations as to the important or basic dimensions of environments. Of course, different investigations have explored different environments, at different levels of analysis. Some studies have compared highly diverse sets of environments, including situations for work, for recreation, for living, and so on, whereas other studies have compared highly similar and specialized environments, such as groups of prisons or

124

TVERSKY

AND

HEMENWAY

of industrial settings. It is not surprising to find that the dimensions that emerge from studying prisons differ from those that emerge from studying industry, which, in turn, differ from those that emerge from studying a broader set of situations. The recent research on categorization, discussed in the first section, suggests a different, but complementary, approach to the problem of characterizing environments. Instead of looking for a set of underlying dimensions that describe a set of stimuli, the research on natural categories attempts to construct a taxonomy of kinds of environments. The construction is based on the assumption that the distinctions that are important for human perception, behavior, and communication will be reflected in language (Brown, Note 1). Moreover, by Zipf s law, the more frequent, presumably more useful categories, will have short labels. Secondly, because categories are related by inclusiveness in a taxonomy of different levels of abstraction, there is no reason to believe that the same dimensions or attributes appear at each taxonomic level. For instance, for object taxonomies, the attributes subjects list for basic level and subordinate categories tend to be perceptual attributes, particularly parts, whereas the attributes listed for superordinate categories tend to be abstract and/or functional (Rosch et al., 1976; Hemenway, 1981; Hemenway & Tversky, Note 2; Murphy & Smith, 1982). Furthermore, even at a given taxonomic level, the attributes informants list differ from category to category, and sets of attributes tend to co-occur. In the world, things that have beaks tend to have feathers rather than fur (Brunei-, Goodnow, & Austin, 1956; Rosch, 1978), and perception of attributes tends to follow this fact about the world (Malt & Smith, Note 3). Finally, a taxonomy of categories related by inclusiveness may have a preferred level of reference, where most of the early differentiations of interest are made. As a taxonomy is developed, more distinctions are made both above and below the basic level (Berlin et al., 1973). In short, many previous approaches to characterizing environments have attempted to uncover a set of dimensions according to which environments could be cross-classified. In contrast, we attempt to uncover a taxonomy of environmental categories from perception of attributes and activities of behavior settings, and from communication about them. One byproduct of this analysis is a set of attributes that informants agree describe various environments. The taxonomy developed here will be a skeletal, demonstrative taxonomy, rather than a comprehensive one that includes all the environments potentially important to environmental psychology. However, it should be readily apparent how to extend the methods to other behavior settings of interest.

CATEGORIES

OF SCENES

125

Constructing a Taxonomy of Environments Can the notion of basic level be extended to other, more abstract, categories, such as environments or scenes? An object is a concrete, typically moveable entity with an identifiable shape. Scenes are a perceptual, spatial generalization of objects; they are the setting or context for objects, the background where objects are figural. Scenes are typically composed of objects, but in particular combinations and configurations. A school contains desks and chairs, baseballs and basketballs, pianos and trumpets, as well as other things, like walls and windows. A restaurant has tables and chairs, tomatoes and lettuce, and cooks, customers, and cash registers. So it seems possible that a taxonomy of scenes might have a preferred level of usefulness, a basic level, based on informativeness in cognition, behavior, and communication. Not all of the operations converging on a basic level for object concepts should be expected to apply to scenes. Scenes, like objects, have primarily perceptual referents, though scenes have fewer constraints on the arrangements of components than objects. In a school scene, the desks, chairs, books, and blackboards can be arranged and rearranged and still be a school, but many rearrangements of the legs of a table will yield something that is no longer a table. Scenes, like objects, are three-dimensional, though typically, we, as human observers and actors, are immersed in scenes, are potentially parts of scenes, whereas we are typically neither immersed in nor potentially parts of objects. Thus we would not expect scenes to have recognizable averaged shapes. One of the reservations expressed about the notion of a basic level of categorization has been that the level of categorization or description selected depends on the context of description. So, for example, in a store specializing in office furniture, even the term, desk chair, would be too general, whereas in a catalog of export items, a term likefurniture might conceivably be too specific, and a term like manufactured item might be preferred. That the level of description depends on context is an undeniable fact, but not one that obviates the establishment of a preferred level in a “neutral” (Cruse, 1977) or typical or average context. To strengthen the evidence for a basic level, Rosch (1978) embarked on a study of the temporal contexts of objects, events. Another approach to the study of contexts of objects is spatial, rather than temporal,

and the spatial con-

texts of objects are one focus of this project on environmental scenes. The present research is an attempt to build a taxonomy for scene categories, and to provide evidence for a basic or preferred level of categorization for scenes. The first set of experiments develops a skeletal taxonomy of scenes and reports evidence for a preferred or basic level of

126

TVERSKY

AND

HEMENWAY

scene category based on perceived attributes, activities, and parts. These measures reflect the perceptual and behavioral aspects of categorization. Our investigations attempt to characterize knowledge about classes of scenes, not memory for any particular scene or group of scenes. The second set of studies reports evidence for basic level scene categories derived from naming, in both a pictorial and a verbal context, reflecting the communicative aspects of categories. CATEGORIES,

ATTRIBUTES, A. Selection

ACTIVITIES,

AND PARTS

of Categories

The first step was to generate a list of scene categories. Indoors and outdoors were selected as superordinates because they come close to being mutually exclusive and jointly exhaustive of scene categories. Then, a group of 80 Stanford students in a psychology of language class were asked to generate categories and subcategories either of indoor or of outdoor scenes. The presumed basic level categories selected were the four most frequently mentioned indoor scenes and the four most frequently mentioned outdoor scenes. This procedure yielded category labels with high frequency in the language (33 or more in the KuceraFrancis (1967) corpus). For subordinate categories, we could not rely entirely on our informants since in some cases there was little consensus and idiosyncratic responding (e.g., naming places near their residences). Based on a pilot study, informal discussion with other informants, and our own intuitions, the subordinates displayed in Fig. 1 were selected. We attempted to find familiar, distinctive subordinates and to avoid proper names wherever possible (e.g., Chicago, Lake Michigan, Macy’s), but were unable to find familiar names of kinds of mountains. A potential problem in the taxonomy produced by our subjects is that the indoor categories can also be conceived of as instances of the category building as well as of indoor scenes. We collected another set of attribute, part, and activity norms using building as a superordinate rather than indoor scene. The quantitative aspects of the data were essentially identical to those to be reported using indoor as the superordinate. B. Attribute

and Activity

Norms

The purpose of collecting these norms was two-fold: first, in order to identify a basic or preferred level of categorization, and second, in order to provide a representative list of attributes and activities that subjects agree apply to scene categories at various levels of abstraction. The attributes subjects list are presumed to derive from their internal representations of the scenes; the attributes refer primarily to visual features of the scenes. Thus, we interpret the attributes listed as reflecting the appearance or perception or cognition of the scenes, and the activities listed

127

CATEGORIES OF SCENES

single-family home

apartment

/ elementary school

\

high sch001

gmcery/ store

\department store

/ fast food reSta”rant

\

d! mO”“tal”S

’ Rocky mOUntalnS

fancy restaurant

Outdoors

,

Park \ A

c,ty park

neighborhood park

~\“t.l”s /eac\

//pi\ midwestern coy

mdustrial city

lake beach

ocean beach

FIG. 1. Hierarchy of scenes.

as reflecting responses or behavior in the scenes. Rosch and her collaborators (1976) found that subjects listed very few attributes or activities (motor programs) common to members of superordinate categories but listed a relatively large number of attributes or activities common to members of basic level categories. Increasing the specificity of the categories by descending the taxonomy to the subordinate level did not appreciably increase the numbers of attributes and activities listed. Going from the superordinate level to the basic level, then, yielded a large increase in the informativeness of the categories, but going from the basic to the subordinate level yielded only a minor increase in informativeness at the cost of a large increase in number of categories. In compiling these norms, we attempted to replicate the procedures of Rosch et al. (1976). Methods Compiling

Norms

A total of 210 subjects, Stanford students completing a course requirement, participated in the study. Subjects were run in groups on this and other unrelated experiments. Subjects produced either attributes or activities for categories at a single level of abstraction. Two groups of 15 subjects listed attributes or activities at the superordinate level, four groups of 15 subjects each listed attributes or activities for either the indoor or the outdoor basic level categories, and eight groups of 15 subjects each listed attributes or activities for either the indoor or the outdoor categories. Subjects given subordinate categories listed attributes or activities for one category from each basic level category. Subjects completed booklets with instructions on the cover and a separate page for each category’s listing, collated in random order for each subject. The cover instructions explained the task. In addition, subjects listing arrributes were instructed “On each of the following pages is the name of a setting or a kind of scene. You will have two minutes to write down the names of as many ATTRIBUTES or

TVERSKY AND HEMENWAY

128

FEATURES of that setting you can think of. For instance, if you had the setting you might write down, ‘is dark, has trees, has animals,’ and so on.” The instructions to subjects listing activities requested “the names of as many ACTIVITIES that are appropriate to that setting as you can think of. For instance, if you had the settingforest, you might write down hunting, hiking, climbing trees, and so on.”

forest,

The instructions and examples varied to tit the level of categorization and the superordinate of the requested listing. Subjects were given 2 min to list attributes or activities for each category. In compiling the norms, attributes or activities listed by five or more subjects were included in the final raw data tally. In addition, attributes listed for a higher level category were included in the listings of its subordinate categories.

Judgment of Attributes Two new groups of subjects were asked to judge the “truth” of the attributes or the “appropriateness” of the activities listed in the raw data tallies. This procedure is typically adopted to ensure that attributes true of two settings but mentioned for only one will be included in the list of attributes of both settings. So, for example, bedroom was listed only for single-family home, but was judged true of both single-family home and apartment, so it is not included in the raw data tallies for apartment, but it is included in the judge-amended norms. The subjects were Stanford students fulfilling a course requirement, and they were run in two groups. Nine subjects judged attribute truth and 10 subjects judged activity appropriateness. They were given pages with categories listed across the top of the page; each page contained one basic level category, its superordinate, and its subordinates. The attributes or activities from the raw data tallies were listed in random order down the page. Subjects were asked to judge the degree of truth of the attributes or the degree of appropriateness of the activities for each scene listed at the top on a 7-point scale, where 7 was very true or very appropriate to the setting, and 1 was not true or inappropriate. They were given detailed oral and written instructions, explaining the origin of the attributes or activities and explaining how to make the ratings. Subjects entered a number in every cell corresponding to their judgment of that attribute or activity for that category. Attributes or activities with a median rating of 6 or more were included in the final judge-amended tally and attributes or activities judged true or appropriate to a higher level category, but not judged true or appropriate to its subordinate categories were eliminated from the higher level categories.

Results The raw data and judge-amended tallies of scene attributes are displayed in Table 1, and the raw data and judge-amended tallies of scene activities are displayed in Table 2. In ail but one case, the increases in number of attributes or activities between the superordinate and basic levels were significantly greater than the increases in number of attributes or activities between the basic and subordinate levels (raw attributes t(7) = 2.71,~ < .05; judge-amended attributes t(7) = 1.74, n.s.; raw activities t(7) = 3.19, p < .05; judge-amended activities t(7) = 4.99, p < .05). Examples of attributes and activities from the judge-amended tallies are listed in the Appendix.

CATEGORIES

129

OF SCENES

TABLE 1 Scene Attributes Subordinate Mountain Park Beach City Home Restaurant Store School r(7) = 2.71,~

Basic

Raw data tally 10 13 11 11 12.5 16 11.5 13

7 10 9 9 11 13 8 10

Superordinate

Outdoors 5

Indoors 5

< .05

Mountain Park Beach City Home Restaurant Store School f(7) = 1.74, n.s.

Judge-amended 7 13.5 8.5 8.5 10.5 12 10.5 14.5

tally 5 10 5 6 8 5 5 13

Outdoors 1

Indoors 1

C. Part Norms

Inspection of the attribute norms revealed that 95% of the attribute types (and even more of the tokens) listed (in the judge-amended list) were parts of scenes. The only nonpart attributes listed were high for muuntains and its subordinates, and cold, tall, small, and warm listed for various subordinates. For object categories, research has demonstrated that an increase in shared parts is particularly diagnostic of basic levelness and that other attributes, mainly descriptors, increase at the subordinate level (Hemenway, 1981; Hemenway & Tversky, Note 1). When the attributes listed by subjects in the Rosch et al. (1976) study are divided into parts (e.g., wheel, stem, handle) and nonparts (e.g., red, large), it is apparent that most of the dramatic increase in attributes listed from the superordinate to the basic level is due to parts, rather than other attributes. On a theoretical level, parts seem to underlie the convergence of perceptual and behavioral measures onto the same basic level. Presumably, part configuration determines object shape, and thereby, the perceptual aspect of a preferred level. When we interact with objects, we typically interact with their parts; we sit on the seat of a couch, lean on the back, remove

130

TVERSKY AND HEMENWAY TABLE 2 Scene Activities Activities Scene

Mountain Park Beach City Home Restaurant Store School f(7) = 3.19, p < .05 Mountain Park Beach City Home Restaurant Store School r(7) = 4.99,p < .05

Subordinate Raw data tally 6 6.5 6.5 6 9 7 8.5 9 Judge-amended tally 7 6 10 3.5 9 4 4.5 6

Basic

Superordinate

Outdoors 2

Indoors 5

Outdoors 0 Indoors 1

the peel of a banana, and eat the pulp. Thus, parts also determine our responses toward objects to a large extent, and thereby underlie the behavioral aspect of a preferred level. Because of the special role of parts in determining the basic level, we decided to compile norms for parts of scenes directly. Method Our methods for collecting raw and judge-amended tallies for parts were nearly identical to our methods of collecting attribute and activity norms. One hundred and five Stanford students completing a course requirement participated in the experiment. Any one student produced parts at a single level of abstraction, and, except for the superordinate group, for a single superordinate. The instructions, varying slightly depending on the particular category, went as follows: “On each of the following pages is the name of a setting, like a forest. You will have two minutes to write down the names of as many PARTS of the setting as you can think of. For instance, you might write down such things as trees, shrubs, grass, wildlife, etc., for parts of a forest.” As before, a scene name was listed at the top of each page, and pages were randomly ordered for each subject. Eleven judges evaluated the truth of the parts obtained in the raw data tallies.

131

CATEGORIES OF SCENES TABLE 3 Scene Parts Parts Scene Mountain Park Beach City Home Restaurant Store School t(7) = 3.21,~ < .05 Mountain Park Beach City Home Restaurant Store School t(7) = 2.41,~ c .05

Basic

Subordinate Raw data tally 16.5 19 22 29 33 30.5 23.5 29.5

16 16 21 27 28 27 19 26

Judge-amended tally 14.5 15.5 11 22.5 32.5 26 21 29

13 15 6 17 30 17 11 26

Superordinate

Outdoors 14 Indoors 15

Outdoors 3

Indoors 6

Results

The raw data and judge-amended parts tallies are displayed in Table 3. As before, the increase in parts from the superordinate to the basic level was significantly greater than the increase from the basic to the subordinate level (raw data tally t(7) = 3.21,~ < .05; judge-amended r(7) = 2.41, p < .OS.>.Examples of parts from the judge-amended tallies are listed in the Appendix. Both the attribute and the part data were analyzed for the level of TABLE 4 Level of Categorization of Objects Listed in Judge-Amended Tallies Tally Level

Attributes

Parts

Superordinate Basic Subordinate

2 21 1

8 43 1

Note. Entries are frequencies.

132

TVERSKY

AND

HEMENWAY

categorization, superordinate, basic, or subordinate, of the objects listed in the judge-amended tallies. Only those elements whose level of categorization had been ascertained by Rosch et al. (1976) or by Hemenway (1981) were included in the count, displayed in Table 4. The vast majority of objects listed in the scene norms were listed at the basic level. Moreover, the preferred level of description of the objects did not depend on the taxonomic level of the scene. Discussion A taxonomy of kinds of environmental scenes was developed, within which a basic or preferred level has been established. The basic level was determined using the major operation defining basic level, a substantial gain in informativeness at that level accompanied by only a minor gain in informativeness at a more specific level of categorization. The result was obtained for norms whose content was derived from the representation of the appearance of the scenes, that is, attributes and parts, as well as for norms based on the activities or behavior appropriate to the scenes. Scenes are not only categories in and of themselves, they are also the spatial contexts in which basic level objects appear. As such, they allow us to ascertain whether objects are referred to at the same level of description across a set of contexts. We found that the preferred level of description for objects embedded in spatial contexts, varying in both content and level of abstraction, is the level previously determined to be basic. Currently, there is interest (Rosch, 1978; Rifkin, Note 4) in determining a basic level in taxonomies of activities or events. The activities listed by our subjects appear to be at a single level of analysis; it would be interesting to determine if that level corresponds to a basic level in an action taxonomy. In spite of the fact that classes of scenes were referred to by name only, most of the attributes, parts, and activities produced were, in fact, observable, perceptible properties and activities of scenes, rather than more abstract features. Ninety-five percent of the attribute types listed were in fact parts of scenes, indicating that subjects viewed parts as a major subset of attributes. However, asking subjects to list parts as opposed to attributes yielded considerably longer lists, despite the fact that, potentially, there are more attributes than parts. Presumably, asking for parts broduced a concrete retrieval plan, namely, search the scene for its parts, but a request for attributes was apparently too abstract to produce a thorough search strategy. Like labels for objects and scenes, retrieval cues may be most effective at an intermediate level of specificity. A basic level for scenes has been demonstrated, based on perceived appearance of the scenes and perceived behavior in them. The basic level is preferred in categorization because for a relatively small increase in the

CATEGORIES

OF SCENES

133

number of categories, or, for a relatively small increase in number of discriminations, a large increase in informativeness is obtained. Generic outdoor and indoor scenes share relatively few attributes and activities; more specific categories, such as beach or school, have many common attributes and activities, but still more specific ones do not share appreciably more qualities. To a certain extent, nature gives us these categories, whether through natural correlations and discontinuities of features in the real world or through our perceptual apparatus, or both (see Rosch, 1978; Mervis & Rosch, 1981). Just as things that have wings also have feathers and fly, scenes that have swings tend to have slides, grass, and children. It is significant that the basic level defined in terms of appearance coincides with the basic level defined in terms of behavior. Elsewhere (Hemenway, 1981; Hemenway & Tversky, Note l), we have presented evidence that shared parts is particularly diagnostic of the preferred level in object categories. Part configuration forms a natural bridge between the perceptual aspect of categories and the behavioral aspect of categories of objects. On the perceptual side, part configuration strongly determines shape; on the behavioral side, part configuration strongly determines human responses. Thus, in the case of objects, natural and manufactured, part configuration seems to underlie the coincidence of the basic level determined perceptually and the basic level determined behaviorally. A similar argument can be developed for scene categories. The desks and chairs in a school, the tables, chairs, food, and cash register in a restaurant, the sand and water at a beach, the streets and buildings in a city, seem to play a major role in both the appearance of a school, restaurant, beach, or city scene and in the activities appropriate to them. The basic level for scenes, then, like the basic level for objects seems to reflect perceived correlations and discontinuities of naturally occurring attributes and activities. Moreover, for scenes as well as objects, part configurations seems to underlie the convergence of the basic level determined by appearance and the basic level determined by behavior. NAMING

SCENES

Is the level determined to be basic in terms of appearance and behavior also the level most useful in communication? Basic level categories, whether objects or scenes, seem to correspond to naturally occurring correlations and discontinuities of features, at least as perceived by our perceptual apparatus (Rosch, 1978; Mervis & Rosch, 1981). Communication about objects or scenes undoubtedly utilizes terms that capture these naturally occurring correlations and discontinuities. However, communication is also context bound in a way that mental representations of categories and taxonomies of categories need not be; in communicating

134

TVERSKY

AND

HEMENWAY

about something, we use a term of reference that selects it from other possible things, from a contrast set (Olson, 1970; Krauss & Glucksberg, 1977; Clark, Schreuder, & Buttrick, Note 5.). The other possibilities may not be a constant set, so the term of reference may vary with the contrast categories. To return to chairs, when purchasing a chair from a furniture store, we may have to refer to it as “the orange tweed rocking chair with a teak base” whereas when it arrives first to an otherwise empty house, we may ask visitors how they like the “furniture.” Several experiments have found that subjects prefer to label objects at the basic level even when they are embedded in a series including other objects at the same level (Rosch et al., 1976; Murphy & Smith, 1982.) In the latter studies, however, subjects were asked to label the objects; they were not asked to provide labels that would distinguish each object from all of the others. Carroll (1980) has shown that subjects describe things differently depending on whether they are instructed simply to label the objects, or instructed to label the objects in order to discriminate one from another, where the objects in a set are fairly similar to one another. The term of reference, then, depends not just on the context or contrast set, but also on the habitual term as well. The habitual term would be that term that discriminates or distinguishes the target object in its typical or neutral or average context (Cruse, 1977). The typical or neutral or average contrast set for chairs would be other furniture or other elements of a home or school scene, presumably, rather than other chairs. The preferred level of communication, then, coincides with the basic level determined by perceived appearance and behavior because the distinctions made at the basic level correspond to the distinctions needed in the typical or average or neutral context. For objects, the preferred level of labeling coincides with the basic level determined by other operations. In the next two experiments, we attempt to demonstrate that this also occurs for scenes. Communication about scenes was explored in two different situations: preferred level of description for photographs of scenes, and preferred level of description in sentences concerning activities occurring in scenes. The photographs were selected so that they could be described at any of the three levels of specificity, and the sentences were designed so that they could be completed at the basic or subordinate levels of specificity, and, in many cases, at the superordinate level as well. The photographs were selected to reflect the general appearance of each of the scene categories, and the sentences were designed to reflect the general behavioral aspects of scenes, by describing activities appropriate to particular settings. In both cases, subjects were free to provide appropriate labels.

CATEGORIES

OF SCENES

135

Methods

Photographs Stimuli. One photograph for each subordinate level scene was selected from a large pool of photographs. Some of the photographs were obtained from the collections of friends, and the remainder were obtained from a professional photographer who volunteered. He was instructed to take general, representative pictures of each of the scenes. Photographs were selected to be clear and typical, and to emphasize a sweeping view of the entire scene. Photographs where the emphasis was on people or on a particular object were excluded. For the indoor scenes, it was decided to use indoor pictures, although it seems compelling a priori that subjects would recognize, and label as expected, typical outdoor photographs of school, restaurant, home, and store buildings. The decision to use indoor shots was not problematic for the store or restaurant pictures, since each of these can be one large room, more or less captured in a single slide. However, it does not seem possible to capture an entire, typical house or school in single photographs. Instead, only what was considered to be the most typical, salient part of each was photographed, a living room for single-famiiy house and for apartment, and a classroom for high school and elementary school. The decision to use indoor shots of indoor scenes, however, led us to difficulties in interpreting the naming data for home and school pictures. The 16 photographs were divided randomly into two groups, where each group included one exemplar from each basic level category. Two different divisions were used, for approximately half the subjects each. Subjects. Sixty-six Stanford students tilling a course requirement served as subjects. They were run in small groups on this and other unrelated experiments. Procedure. Subjects were told they would see slides of common and familiar scenes. They were asked to provide a “very simple common name or label for each of the slides,” and were told, “We do not want elaborate or creative answers. Instead, we want the most simple, obvious, direct sort of name that ordinary people would give for each scene.” The instructions were designed to discourage idiosyncratic, artistic titles. Subjects were shown the entire set of stimuli at a rate of 2 set per slide, and then shown each slide for approximately 10 set each, sufficient time for written responding.

Sentences Stimuli. For each subordinate level scene, a sentence was composed describing an activity occurring in a setting; the description of the setting was to be tilled in by the subjects. For example, for apartment, the sentence was: The Kingstons furnished their ~ with furniture they built themselves. At issue was which level of specificity would be selected. Either a specific description, apartment, or a basic level description, home, could be entered in the blank (or any other response deemed appropriate by the subject, for that matter). Some more examples of sentences follow. ForRocky Mountains, the sentence was: Skiers and rock climbers seek work in Denver to take advantage of the nearby __. For industrial city, the sentence was: Because he feared pollution, filth, crime, and crowding, Eric refused to consider a job in ~. For elementary school, the sentence was: Since the Hudsons’ children were aged 6 and 8, they wanted their new home to be near ~. While an attempt was made to construct sentences where labels at all three levels of generality would be appropriate, this was not always possible. For all the sentences, however, labels at both the basic and subordinate levels were appropriate.

136

TVERSKY

AND

HEMENWAY

subjects. Forty-seven Stanford students participated in this and other unrelated experiments either to fulfill a course requirement or for pay. Procedure. Each subject responded to one sentence from each basic level category. Sentences were typed on separate pages and randomly ordered in a booklet. The first page of the booklets contained the followng instructions: This is a simple experiment investigating the terms people naturally use in common, everyday language. Please fill in the blanks in the following sentences in the way you think most ordinary people would complete them. Sometimes, you will need more than one word to complete the sentence. We do NOT want elaborate or creative answers, just the most direct, natural and obvious answers you can think of.

Results Agreement Subjects’ responses were tallied and scored for content, as basic level (target term) or some other response. The two most frequent responses and the percentage of responses falling into those and other frequency categories are displayed in Table 5 for pictures and in Table 6 for sentences. There was a high degree of consensus in the responses. For 13 out of 16 pictures and for 13 out of 16 sentences, a majority of subjects agreed on the same response. In general, when the majority response was not given, responding tended to be idiosyncratic. While the greatest proportion of responses for most of the pictures and sentences were the majority responses, the second greatest proportion were idiosyncratic responses, responses given by only a single subject, for 13 out of 16 of the pictures and for 13 out of 16 of the sentences. When subjects failed to give the target response, they tended to give idiosyncratic responses at a lower level of specificity. For example, the sentence targeting midwestern city evoked “Des Moines,” “St. Louis,” and “Cinncinnati” in three different respondents. One respondent called the photograph of lake beach “summer evening,” another called it “lake front,” and a third described it as “coast.” Pictures Since the photographs for school and home categories could only exhibit parts of each, and not the whole, it is not unreasonable that subjects labeled the part, rather than the whole. It is analogous to showing a picture of only the wing of an airplane or only the pocket of a shirt; these pictures would probably elicit “wing” and “pocket” as labels rather than “airplane” and “shirt.” At this point, we were left with two alternatives. We could eliminate these stimuli from our analysis, or we could redefine the basic level term of reference for these two stimuli. Thus, the target term for home would be “livingroom” and the target term for school

137

CATEGORIES OF SCENES TABLE 5 Picture Naming Percentage of Responses to each Response Category

Category Grocery STORE Department STORE Elementary SCHOOL High SCHOOL Fast Food RESTAURANT Fancy RESTAURANT Apartment HOME Single-family HOME Ocean BEACH Lake BEACH Rocky MOUNTAINS Sierra MOUNTAINS Midwestern CITY Industrial CITY City PARK Neighborhood PARK

Supermarket

52

Department store 67 Classroom

55

Classroom

64

Resta”rant

Two subjects agree

Next most frequent response

Most frequent response

46

Grocery store

24

Store

12

Elementary school classroom 15

Cafeteria

20

18

Idiosyncratic 6 21

18

12

6

30

6

29

13

29

Restaurant

58

Living room

60

House- home

11

6

23

Living room

48

House-home

13

13

26

Beach

69

Beach

58

Mountain(s)

64

Mountain(s)

61

City

52

Inner city

City

52

Park Park

31 6

35

6

21

18

21

12

6

30

Downtown

9

6

33

36

Playground

36

27

70

Field

9

21

Mountain range 9

Note. Rows sum to 100 (except for rounding). Average n = 33.

would be “classroom.” More general terms, never obtained, would have been terms like “indoors” or “building,” and more specific labels, obtained, though infrequently, were, for example, “elementary school classroom” and “middle class livingroom.” For the city pictures, “cityscape” and “city scene” were counted as city responses, but “city skyline” and “city buildings” were not. This was because the former terms refer to the whole abstract scene, and the “scape” and “scene”

TVERSKY AND HEMENWAY

138

TABLE 6 Sentence Completion Percentage of Responses to each Response Category

Category Grocery STORE Department STORE Elementary SCHOOL High SCHOOL Fast Food RESTAURANT Fancy RESTAURANT Apartment HOME Single-family HOME Ocean BEACH Lake BEACH Rocky MOUNTAINS Sierra MOUNTAINS Midwestern CITY Industrial CITY City PARK Neighborhood PARK

Two subjects agree

Next most frequent response

Most frequent response

13

8

Idiosyncratic

Store

71 Grocery store

Store

74

School

54 Elementary school 13

School

70 High school

Restaurant

58

Restaurant

78

13

House- home

79

13

House- home

70

22

Beach

71

29

Beach

19 Park

Mountain(s)

63

Mountain(s)

61

City

33 Chicago

City

48

Park

88

13

Park

87

13

8 26

8

22

9 25

19

25

17

10

52

17

21 39

29

8

29

17

35

Note. Rows sum to 100 (except for rounding). Average n = 23.4.

seemed to be added to make the picture title more literary, while the latter terms refer to only part of the abstract scene. Omitting these cases would have meant that the target term, city, would have been the modal but not the majority response for these two pictures. For labeling photographs, the preferred level was the basic level. The target response was the modal response for 14 out of 16 cases (10 out of 12 excluding home and school) and the majority response for 11 out of 16

CATEGORIES

OF SCENES

139

cases (10 out of 12 excluding home and school). The two most frequent labels for each photograph and the percentage of subjects responding with them are displayed in Table 5. The notable exceptions in labeling photographs were grocery store, which was called “supermarket,” and department store, which was thus labeled. The photograph of thefastfood restaurant was called “cafeteria” by seven subjects, so, although “restaurant” was the modal response, it was not quite the majority response. The photograph of the single-family living room included a hallway and parts of the dining room and kitchen, so that while the modal response was “livingroom;” this was not the majority response because of the prevalence of “house” or “home” responses. Finally, the city park photograph showed only a part of the park, the playground, and it was thus labeled by as many subjects as labeled it “park.” Sentences. In sentence completion, the modal response was the target basic level in 16 out of 16 cases, and the majority response was basic level in 13 out of 16 cases (significant by a sign test, p < .05). The two most frequent responses to each sentence and the percentages of subjects reporting them are displayed in Table 6. The sentence exceptions were not revealing: both city sentences, and the lake beach sentence were exceptions to the majority criterion, but not to the modal criterion. Through an oversight, the lake beach sentence mentioned sunning but failed to mention an activity dependent on water, such as swimming. As a consequence, this sentence elicited “park” as frequently as “beach” and also elicited other idiosyncratic places to sun, such as “field” and “tanning spot.” Discussion

In both labeling photographs of scenes and in completing sentences describing activities performed in scenes, subjects prefer the basic level terms even though more specific or more general terms would have been appropriate. Thus, the level preferred in communication corresponds to the level determined to be basic by virtue of perceived appearance and perceived behaviors. The labels given to pictures exhibited greater variance than those used to complete sentences. This is probably due in part to the difficulty of capturing an entire scene in a single photograph, a problem that does not exist for objects. This was particularly apparent for the home and school scenes, and to a lesser degree, for the park scenes. Another factor contributing to the variability of responses to pictures was the tendency of subjects to compose literary titles for the pictures, despite instructions to be mundane and ordinary. In a pilot study, simply requesting titles for the pictures, we found an even higher proportion of idiosyncratic, literary responses. We succeeded in reducing the creativity of our respondents by instructing them to answer in the ordinary way that other people might

140

TVERSKY

AND

HEMENWAY

answer, but we did not eliminate it entirely. By being ordinary and mundane themselves, our sentences apparently succeeded in diminishing the creative tendencies of our subjects, so that responding was more stereotyped. Of all the exceptions, only store seems to approach a genuine violation. Our two kinds of store, grocery store and department store, differ considerably on measures related to represented appearance, (raw) parts and photograph labeling, though not, significantly, on the measures related to behavior, activities, and sentence completion. In fact, grocery store is referred to by a single word term, supermarket, another indication that it may be in between basic and subordinate status. GENERAL

DISCUSSION

This investigation has elicited and explored a taxonomy for environmental scenes, and has presented evidence for a preferred or basic level in scene categories. The evidence has come from judgments of the attributes of scenes, judgments of activities characteristic of the scenes, and labels used for scenes captured in photographs and referred to in sentences. Subjects perceive and judge that superordinate categories, indoor and outdoor, have very few attributes, parts, or activities in common, whereas basic level categories, beach, mountains, home, school, share many attributes, parts, and activities. Still more specific categories, lake beach, elementary school, do not share appreciably more of these features. So, for scenes as for objects (Rosch et al., 1976), categories at the basic level are highly informative, but not at the expense of a multitude of categories and classifications. Moreover, when subjects are asked to label photographs or to describe the setting of some activity, they prefer basic level terms. These measures of human cognition, human behavior, and human communication all converge on the same level. More general categories, such as indoors and outdoors are apparently too impoverished to be of general use, and more specific categories, such as lake beach and elementary school are apparently too discriminitive to be widely applied. Naturally, both more general and more specific categories are useful in certain contexts; we are speaking here of a preference for a level of reference in a typical or neutral context (Cruse, 1977). Another indicator that a set of terms in a taxonomy is basic is the nature of the terminology itself (Berlin, 1978); basic level terms tend to be primary lexemes, short, one-word terms, whereas subordinate terms tend to be secondary lexemes that are formed from the basic level term and a modifier as well. This indicator also holds for our scene categories: the basic level terms are primary lexemes, while the subordinate terms are generally compound terms, like “lake beach” and “department store.” Like objects, scenes are concrete stimuli composed of parts. Other research (Hemenway,

CATEGORIES

141

OF SCENES

1981; Hemenway & Tversky, Note 1.) has indicated that an increase in part attributes is particularly diagnostic of the basic level for objects; again, this holds for scenes. Thus, this project has extended the evidence for a preferred level of categorization and reference from objects, to the spatial contexts of objects, scenes. Other researchers have attempted to ascertain a basic level for events (Rosch, 1978; Rifkin, Note 4), for personality (Cantor & Mischel, 1979), and for psychiatric diagnostic categories (Cantor, Smith, French, & Mezzich, 1980). Spatial

and Temporal

Contexts

While scenes form the spatial context for objects, event structures form the temporal context in which objects are props (Rosch, 1978). Scenes, then, are analogous to scripts (Shank & Abelson, 1977; Bower, Black, & Turner, 1979); both are information-rich bundles of information, both are contexts for objects. In scripts, the information is primarily about action sequences, in scenes, about component structures, but also about activities; in scripts, the context is time, in scenes, space. Despite shifts in context, whether in degree of abstraction or in specific content, objects appear to be referred to at the same, basic level. This had been observed for temporal contexts (Rosch, 1978) and was demonstrated here for spatial contexts as well. Although it is easy to conceive of contexts in which objects are referred to by either a more general or a more specific term of reference, we have shown that when context is systematically varied over a broad range of abstraction and content, the level of reference for objects nevertheless remains the same. Parts and the Convergence of Measures and Communication

of Cognition,

Behavior,

Why is it that these three different kinds of measures, one of our cognition of scenes (or objects), one of our behavior in them, and the third, of our communication about them, converge on the same level? We would like to offer some speculations, first about why measures of representation and behavior converge, and then about why measures of communication converge to the others. Parts underlie both the appearance of scenes (and objects) as well as the activities appropriate to them. So, parks have playgrounds and grass; we play in the playgrounds and picnic on the grass. Schools have teachers, desks, and books; we learn from the teachers, write at our desks, and read books. Objects and scenes at the superordinate level are perceived to have few parts in common, while objects and scenes at the basic level are perceived to share many parts. The increase in perceived parts at the subordinate level is relatively inconsequential, for both objects and scenes. The same pattern holds for perceived attributes and activities, and parts compose the majority of

142

TVERSKY

AND

HEMENWAY

attributes listed for objects, and the overwhelming majority of attributes listed for scenes. Parts act as a natural bridge between cognition and behavior. On the one hand, parts determine the appearance of objects and scenes to a large degree, and thereby determine our representations of them. On the other hand, parts determine our behavior to a significant extent, since most of our behavior is directed toward components of an object or scene (See also, Hemenway, 1981; Hemenway & Tversky, Note 2). Cognition and behavior, then, may converge because part contiguration is the basis for both of them. The issue of communication, and why measures of communication also converge to the same, preferred level of abstraction, brings us full circle to our starting point, “How shall a thing be called?” (Brown, 1958). In communication, a speaker selects a term of reference that will identify an object (or scene) and will also distinguish it from other possible objects or scenes. But first and foremost, the speaker wants to communicate, that is, to select a term that will be interpretable to the addressee. If, for a community of speakers, certain categories are distinctive by virtue of both appearance and behavior, then communication would be facilitated if each of these distinctive categories had a distinctive label. In other words, classifications made in language should reflect classifications made in perception and behavior. Thus, basic level terms should be preferred a priori. Naturally, context also affects labeling. In a chair store, “I like the chair” probably will not be specific enough to communicate, but in a small office, “Have a chair” probably will do. But a chair is a kind of furniture or a part of a home scene or a restaurant scene. The more general categories, of which chair is a part or a kind, seem to form the most natural or typical or neutral context for chair. The other possibilities or natural contrast set for chair, then, would be other items of furniture or other parts of a home or school or restaurant scene. Similarly, a school is a kind of indoor scene or a part of a city, so, again, these more inclusive categories would form a natural, neutral, typical context for school. In this and other research varying context (Rosch et al., 1976; Murphy & Smith, 1982), the basic level terms of reference were preferred for scenes as well as objects. All other things being equal, then, basic level terms of reference will be preferred because they capture the categories of appearance and of behavior. Context can alter selection of a term of reference, but it appears that the default or neutral (Cruse, 1977) or typical contexts of an object or scene demand discrimination, and thus labels, at the basic level of abstraction. Taxonomy

of Environmental

Scenes

Classification and the construction of taxonomies are ubiquitous in academic disciplines, if not in all cognitive activity. First, categories si-

CATEGORIES

OF SCENES

143

multaneously group entities that are similar and separate entities that are different. Then, categories are nested into more inclusive and abstract groupings. Together, these organize and simplify knowledge, and form a foundation on which theories can be built. The search for a way to characterize environments is not new (e.g., Craik, 1973; Fredricksen, 1972; Moos, 1973; Russell & Ward, 1982; Stokols, 1978). Most previous approaches have used factor analysis or multidimensional scaling techniques to derive a small set of factors or dimensions capable of describing a large set of environments. There have been many different studies employing these techniques, and many different sets of factors or dimensions obtained. The situation improves when the same stimuli are used for both factor analysis and multidimensional scaling (Price & Blashfield, 1975; Ward & Russell, 1981). Then, there appears to be a high degree of overlap between the factors and dimensions obtained by the different techniques. Both sets of researchers commented on the correspondence of factors and dimensions obtained by different techniques for the same stimuli: in fact, there is a fair degree of overlap of the factors and dimensions obtained across those two projects with those obtained in the present study. The clustering analyses tended to produce clusters approximately equivalent to our subordinate categories (e.g., elementary and high school settings), whereas the dimensional approach tended to produce dimensions corresponding to our most global categories (e.g., outdoors-indoors; natural-man-made). However, the interpretation of the factors or dimensions is not determined by the scaling methods; for instance, Ward and Russell (1981) have argued that the factors they obtained could be given a compelling affective interpretation or an equally compelling, but qualitatively different, cognitive interpretation. Our approach has been to construct a taxonomy of environmental scenes, in a manner analogous to the construction of taxonomies of natural kindsplants and animals-and to more recent taxonomies of objects. We have begun with comprehensive superordinate categories, indoors and outdoors, but other global categories could serve as a starting point. We have let subjects produce the stimuli and the features (dimensions) as well. The method used here allows features to be correlated, acknowledges that scenes with cars are likely to be bustling, but scenes with water and sand are likely to be calm. It also allows for different features, even different kinds of features, to appear at different levels of abstraction. It provides a way of directly determining, rather than inferring, the features in the minds of observers. It takes into account measures of appearance as well as measures of behavior (activities). Certain natural extensions are suggested by the categorization approach. One, to a taxonomy of action and to the relations of such a taxonomy to activities appropriate to scenes, was raised earlier. Another possibility is to examine the internal structure

144

TVERSKY

AND

HEMENWAY

of scene categories. Are elementary schools perceived to be more typical than beauty schools? Do more typical exemplars also have more of the shared features than less typical? Information about how typical a particular setting is for its class of settings may also be useful to environmental psychologists. We have by no means accomplished the Linnaean task of classifying all environments, but we have suggested a way to go about it that should complement and supplement previous approaches. A Broader Look Our research indicates that knowledge about scenes, like knowledge about objects, has a hierarchical organization, in which an intermediate level of abstraction is basic, or preferred, in measures of cognition, behavior, and communication. Basic level scenes, like basic level objects, are information-rich bundles of features, primarily parts. The taxonomy obtained provides a way to organize knowledge about the multitude of behavior settings which environmental, social, and personality psychologists maintain are strong determinants of behavior. It is not surprising that the taxonomy obtained and the level found to be basic articulate well with settings and scenes used by scholars in other fields, each for different reasons. The kinds of scenes selected by researchers on scene schemas in perception and memory (Biederman, 1981; Friedman, 1979; Loftus & Kallman, 1979; Mandler & Johnson, 1976) correspond well to our basic level scenes. These researchers have shown, among other phenomena, that when the configuration or organization of parts of a scene is disrupted, recognition and memory are impaired. Another relevant finding is that when parts or elements from one scene are introduced into pictures of other scenes, the imported parts are more likely to be noticed and recognized than when they appear in their natural contexts. This kind of manipulation has typically extracted a part from one basic level scene, and added it to a scene from another basic level category. Presumably, moving a chair from one home scene to another home scene would not be as noticeable as moving a fire hydrant to a home or as moving an octopus to a farmyard. The list of indoor scenes we obtained looks like the table of contents of a textbook on architecture (see, for example, de Chiara & Callender, 1973): home, school, restaurant, and so on. The list of outdoor scenes looks like the table of contents of a geography text or the key to symbols on a map. It also resembles the first question a travel agent would ask a prospective client: “Would you like to go to the mountains, or the seashore, or perhaps you’d prefer a city setting for your vacation?” Thus, the taxonomy of scenes, as well as the preferred level of reference to scenes generated by our subjects seem to correspond to the taxonomy and preferred level useful in the comprehension of environmental settings as well as implicit in the work of psychologists, architects, geographers, and other experts who work with scenes.

CATEGORIES

APPENDIX:

145

OF SCENES

JUDGE-AMENDED LISTS OF ATTRIBUTES, ACTIVITIES, AND PARTS

Items listed at a high-level

category

are also included in the category’s

subsets.

A. ATTRIBUTES

MOUNTAIN sun rock boulder high ROCKIES cold snow SIERRAS lake tree

RESTAURANT table menu cash register food FANCY office good food silverware waitress lamp tablecloth waiter maitre’d chair music FAST-FOOD french fries concrete grease hamburgers

OUTDOOR skv -BEACH PARK sun water fountain bird sun people towel water bench playground LAKE flower tree tree OCEAN seagull grass bird sand CITY sunbathers slides seaweed wave children swings shell NEIGHBORHOOD picnic slides children swings

STORE cash register money people aisle DEPARTMENT office concrete elevator clothes escalator table GROCERY office concrete food shopping carts meat

INDOOR window HOME table lamp chair sofa furniture bedroom kitchen SINGLE-FAMILY family living room warm APARTMENT tall small

CITY people restaurant car building tall building INDUSTRIAL factory smokestack smog bus MIDWESTERN tree

SCHOOL office concrete chair teacher desk friend library cafeteria playground building book table HIGH SCHOOL football ELEMENTARY children recess

146

TVERSKY AND HEMENWAY B. ACTIVITIES

MOUNTAIN picnics backpacking fishing skiing playing hiking camping ROCKIES (none) SIERRAS (none)

RESTAURANT buying eating FANCY drinking entertaining FAST-FOOD (none)

OUTDOOR (none) PARK BEACH picnics picnics running running playing frisbee playing frisbee playing volleyball playing volleyball playing playing swimming playing football CITY playing football camping (none) building sandNEIGHBORHOOD (none) castles LAKE fishing OCEAN surfing INDOOR talking HOME STORE relaxing buying sleeping shopping watching TV walking DEPARTMENT learning trying on clothes cooking cleaning GROCERY eating (none) SINGLE-FAMILY playing APARTMENT studying

CITY going to movies going shopping working INDUSTRIAL going to plays MIDWESTERN (none)

SCHOOL learning reading writing walking HIGH SCHOOL studying ELEMENTARY playing

C. PARTS

MOUNTAIN bird tree rock grass stream dirt animal trail snow valley ROCKIES flower water

OUTDOOR sky sun insect BEACH PARK bird bird water tree sand rock LAKE bench tree grass lake people grass dirt people cloud dirt swing set OCEAN picnic table cloud flower driftwood water lifeguard CITY

CITY people dirt cloud water sidewalk car police station fire station store building restaurant street bus

CATEGORIES

SIERRAS lake

RESTAURANT water kitchen customers table bathroom cash register chair food napkins cooks room FANCY menu plant picture carpet silverware tablecloth glasses maitre’d rug furniture waitress good food waiter paintings candles FAST-FOOD

counters french fries hamburgers

fountain NEIGHBORHOOD (none)

147

OF SCENES

seaweed frisbee

INDOOR ceiling roof wall floor door window STORE HOME plant customers picture cash register aisles kitchen carpet sales people check-out hallway couch counters DEPARTMENT table picture TV bathroom table TV rug furniture rug stairs chair furniture room chair stereo room lamp clothing livingroom stereo curtains elevator bed lamp sink escalator book refrigerator jewelry shoes bedroom GROCERY stove meat closet vegetables SINGLE-FAMILY fruit desk shopping carts garage room dining room family backyard APARTMENT

(none)

school INDUSTRIAL smokestack pollution smog factory MIDWESTERN bird tree rock park house grass

SCHOOL picture hallway table bathroom furniture chair office desk room book chalk cafeteria blackboard athletic field teacher playground library student classroom papers HIGH SCHOOL auditorium gym locker rug stairs plant ELEMENTARY

(none)

148

TVERSKY AND HEMENWAY

REFERENCES Barker, R. G. Ecological psychology. Stanford, CA: Stanford Univ. Press, 1968. Bandura, A. Social learning theory. New York: Prentice-Hall, 1977. Berlin, B. Ethnobiological classification. In E. Rosch & B. B. Lloyd (Eds.), Cognition and categorizafion. Hillsdale, NJ: Erlbaum, 1978. Berlin, B., Breedlove, D. E., & Raven, P. H. General principles of classification and 1973, 75, 214-242. nomenclature in folk biology. American Anthropologist, Biederman, I. On the semantics of a glance at a scene. In M. Kubovy & J. R. Pomerantz (Eds.), Perceptual organization. Hillsdale, NJ: Erlbaum, 1981. Bower, G. H., Black, J. B., & Turner, T. J. Scripts in memory for text. Cognitive Psychology, 1979, 11, 177-220. Brown, C. Folk botanical life-forms: Their universality and growth. American Anthropologist, 1977, 79, 317-342. Brown, C. Folk zoological life-forms: Their universality and growth. American Anthropologist, 1979, 81, 791-817. Brown, R. How shall a thing be called? Psychological Review, 1958, 65, 14-21. Bruner, J. S., Goodnow, J. J., & Austin, G. A. A Study ofthinking. New York: Wiley, 1956. Cantor, N., & Mischel, W. Prototypes in person perception. In L. Berkowitz (Ed.), Advances in experimental social psychology, Vol. 12. New York: Academic Press, 1979. Cantor, N., Smith, E. E., French, R., & Mezzich, J. Psychiatric diagnosis as prototype categorization. Journal of Abnormal Psychology, 1980, 89, 181- 193. Carroll, J. M. “Purpose” in a cognitive theory of reference. Bulletin of the Psychonomic Society, 1980, 16, 37-40. Clark, H. H., & Carlson, T. B. Context for comprehension. In J. Long & A. Baddeley (Eds.), Attention and performance IX. Hillsdale, NJ: Erlbaum, 1981. Craik, K. H. Environmental psychology. Annual Review ofPsychology, 1973,24,403-421. Cruse, D. A. The pragmatics of lexical specificity. Journal of Linguistics, 1977, 13, 153- 164. de Chiara, J., & Callender, J. H. Time-saver standards for building types. New York: McGraw-Hill, 1973. Fredricksen, N. Toward a taxonomy of situations. American Psychologist, 1972, 27, 114- 127. Friedman, A. Framing pictures: The role of default knowledge in automatized encoding and memory for gist. Journal of Experimental Psychology: General, 1979, 108, 316-355. Hemenway, K. The role of perceived parts in categorization. Unpublished doctoral dissertation, Stanford University, 1981. Krauss, R., & Glucksberg, S. Social and nonsocial speech. Scientific American, 1977, 236, loo- 105. Kucera, H., & Francis, W. N. Computational analysis of present-day American English. Providence, RI: Brown Univ. Press, 1967. Loftus, G. R., & Kallman, H. J. Encoding and use of detail information in picture recognition. Journal of Experimental Psychology: Human Learning and Memory. 1979, 5, 197-211. Mandler, J., & Johnson, N. S. Some of the thousand words a picture is worth. Journal of Experimenral

Psychology:

Human Learning

and Memory,

1976, 2, 529-540.

Mervis, C. B., & Rosch, E. Categorization of natural objects. Annual Review of Psychology, 1981, 32, 89- 115. Mischel, W. Persona&y and assessment. New York: Wiley, 1968. Mischel, W. Toward a cognitive social learning reconceptualization of personality. Psychological RevieMa, 1973, 80, 252-283.

CATEGORIES

OF SCENES

149

Moos, R. Conceptualizations of human environments. American Psychologist, 1973, 28, 652-663. Murphy, G. L., & Smith, E. E. Basic level superiority in picture categorization. Journal of Verbal Learning and Verbal Behavior, 1982, 21, l-20. Olson, D. R. Language and thought: Aspects of a cognitive theory of semantics. Psychological Review, 1970, 77, 2577273. Price, R. H., & Blashfield, R. K. Explorations in the taxonomy of behavior settings: Analysis of dimensions and classifications of settings. American Journal of Community 1975, 3, 335-351. Psychology, Rosch, E. Principles of categorization. In E. Rosch & B. B. Lloyd (Eds.), Cognition und categorization. Hillsdale, NJ: Erlbaum, 1978. Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., & Boyes-Braem, P. Basic objects in natural categories. Cognifive Psychology, 1976, 8, 382-439. Russell, J. A., & Ward, L. M. Environmental psychology. Annual Review, of Psychology, 1982, 33, 651-688. Shank, R., & Abelson, R. Scripts, plans, goals, and understanding. Hillsdale, NJ: Erlbaum, 1977. Smith, E. E., & Medin, D. L. Categories and concepts. Cambridge, MA: Harvard Univ. Press, 1981. Smith, R. P. “Where did you go?” “Out.” “What did you do?” “Nothing.” New York: Norton, 1958. Stokols, D. Environmental psychology. Annual Review of Psychology, 1978,29, 253-295. Ward, L. W., & Russell, J. A. The psychological representation of molar physical environments. Journal of Experimental Psychology: General, 1981, 110, 121- 152.

REFERENCE

NOTES

1. Brown, R. Nutural cafegories and basic objects in the domain of persons. Katz-Newcomb Lecture, University of Michigan, 1980. 2. Hemenway, K., & Tversky, B. What’s basic about the basic level? Paper presented at the Psychonomic Science Society Meetings, Philadelphia, 1981. 3. Malt, B., & Smith, E. E. Correlated properties in natural categories. Manuscript, Stanford University, 1982. 4. Rifkin, A. J. Event categories, event taxonomies, and basic level events: An initial investigntion. Manuscript, City University of New York Graduate Center, 1981. 5. Clark, H. H., Schreuder, R., & Buttrick, S. Common ground and the anderstanding of demonstratives. Manuscript, Stanford University, 1981.