What's so funny about that?: The domains-interaction approach as a model of incongruity and resolution in humor

Motivation and Emotion, VoL 18, No. 1, 1994

What's So Funny About That?: The Domains-Interact ion Approach as a Model of Incongruity and Resolution in Humor 1

Tim R. Hillson 2 and Rod A. Martin 2~

The domains-interaction approach, originally developed in metaphor research, was used to model humor processing. Using 'yokes" of the structure "A is the B of A's domain" (e.g., "John Candy is the hamburger of actors"), this approach allows for precise operationalizations of incongruity and resolution in terms of between-domain and within-domain semantic distance, respectively. Distances between 26 terms used in 250 of these simple jokes were derived from semantic differential ratings. As predicted, humor ratings of the jokes were positively correlated with incongruity (between-domain distance) but not with resolution (within-domain distance). Also as predicted, a significant interaction revealed that jokes with both incongruity and resolution were rated as most humorous. These findings were replicated in a second study. A third study indicated that humor perception differs from metaphor appreciation, showing main effects for both types of semantic distance, but no interaction, when the sentences were presented as metaphors instead of jokes and rated for aptness.

This paper describes three studies that were designed to model the incongruity and resolution components of humor comprehension and appreciation, using the domains-interaction approach. This approach was borrowed from research on metaphors (Tourangeau & Sternberg, 1981, 1982). We begin by discussing theoretical and empirical work relating to the incongruity-resolution theory of humor. We then describe the domains-interac-

IThe authors wish to thank Richard Harshman and Albert Katz for their helpful suggestions in conducting this research.

2Department of Psychology, University of Western Ontario, London, Ontario, Canada N6A 5C2.

31"o whom correspondence should be addressed.

1

0146-7239/94/0300-0001507.00/0 © 1994 Plenum Publishing Corporation

2 Hillson and Martin

tion approach to metaphor and show how it can be applied to model basic humor processing. While the domains-interaction approach is certainly unable to capture all or even most of the subtle and complex nuances of humor, we propose that the approach may prove useful in modelling the more basic aspects of humor.

THE INCONGRUITY-RESOLUTION THEORY OF HUMOR

Although it is generally recognized that humor involves a variety of emotional, motivational, and esthetic aspects, cognitive theories of humor have emphasized the importance of incongruity as an essential component. For example, James Beattie, the 18th-century Scottish philosopher, suggested that "laughter arises from the view of two or more inconsistent, unsuitable, or incongruous parts or circumstances considered as united in a complex object or as acquiring a sort of mutual relation from the peculiar manner in which the mind takes notice of them" (quoted by Suls, 1983, p. 40).

Forabosco (1992), in a recent discussion of the concept of incongruity in humor, has suggested that "a stimulus is incongruous when it differs from the cognitive model of reference" (p. 46). In other words, incongruity occurs when there is a discrepancy between the cognitive model (set, schema, script, etc.) of an individual and an incoming stimulus against which it is compared.

A number of theorists have argued that incongruity alone is sufficient for producing humor. For example, Nerhardt (1976, 1977) has proposed that humor results from incongruity, defined as a discrepancy between cognitive representations of expectations (e.g., as derived from the body of a joke) and actual stimuli or events (e.g., the punchline of the joke). Accord- ing to this view, the mere perception of incongruity within a safe and nonthreatening environment is necessary and sufficient to produce humor. A similar view has been proposed by Rothbart (1976).

Other theorists have argued that, although incongruity alone is often sufficient for humor, in more effective forms of humor the incongruity must in some sense be resolved (Schultz, 1972; Suls, 1972, 1983). According to this "incongruity-resolution" view, humor results when the discrepancy present in a joke or cartoon is found to "make sense" in some way. As an example of this approach, Suls (1972) proposed a two-stage model of humor comprehension, based on an information-processing type of analysis. The first stage in this model involves the disconfirming of expectations, or perception of incongruity, as in Nerhardt's theory. The second stage involves a problem-solving process to find a cognitive rule that explains or

Domains-Interaction Approach to Humor

"solves" the basis of the incongruity and provides a joke resolution. This cognitive rule might be a logical proposition, a definition, or a fact of ex- perience. In Suls' (1983) model, then, most humor arises from seeing the sense in what at first appeared to be nonsense.

To illustrate this process, Suls (1983) offers the following joke: "Pro- fessor: 'Mr. Twerp, what do you know about French syntax?' Student: 'Gosh, I didn't know they had to pay for their fun.'" In processing this joke, the listener, on hearing the first part of the joke, would first make some prediction of how it might end. For example, one might expect some dull academic jargon concerning the nature of the French language. The punchline, however, is discrepant from this expectation, resulting in incongruity and puzzlement. In the second stage, the discrepancy is resolved by recognizing the alternate interpretation of "syntax" as "sin tax."

Researchers have taken a number of approaches to studying the role of incongruity and resolution in humor. Nerhardt (1970) manipulated subjects' expectations and the discrepancies of stimuli from these expectations in a series of psychophysical experiments. For example, in one study subjects lifted a series of weights in which the final weight was greatly discrepant from the others (either much heavier or much lighter). The results showed that many of the subjects smiled and laughed after lifting this final weight and that the amount of such mirth was a function of the degree of discrepancy from the expected weight range. These findings were taken to support the theory that incongruity alone is sufficient for humor.

While we do not dispute Nerhardt's suggestion that incongruity alone is sufficient for humor, it should be noted that this approach has some limitations. First, it examines only a nonverbal form of humor stimuli as well as nonverbal responses (i.e., smiling and laughter), which do not nee- essarily indicate a perception of humor. (Indeed, they may have been in- dicative of embarrassment, for example.) In addition, although this approach allows for a fairly precise way of measuring incongruity (i.e., the discrepancy between actual weight and expected weight), it does not provide a means of assessing the possible resolution of the incongruity. Thus, it is not surprising that Nerhardt did not find any evidence for the resolution of incongruity in humor, since his methodology did not provide a means of measuring this resolution.

As an alternative approach, Schuitz (1972, 1974; Schultz & Horibe, 1974) conducted a series of studies designed to assess the importance of incongruity and resolution to children's ratings of the funniness of cartoons, riddles and jokes. In these studies, the punch lines of jokes were altered in order to remove either the incongruity or the resolution. For example, one original riddle was, "Question: Why did the cookie cry? Answer: Be- cause its mother was a wafer so long." In the resolution-removed version,


the answer was changed to "Because its mother was a wafer," to produce a riddle presumably matched with the original on degree of incongruity, but without resolution. These studies showed that children over the age of 8 generally found the original jokes to be funnier than the resolution-removed versions and the resolution-removed jokes to be funnier than the incongruity-removed jokes.

Although Schultz's approach attempts to manipulate incongruity and resolution independently, a number of researchers have pointed out prob- lems with the methodology. In particular, Wilson (1979) has argued that it is impossible to separate the influences of incongruity and resolution of authentic jokes. For example, the resolution-removed versions may also be less incongruous than the originals. A s a result, it is impossible to tell whether the lower funniness ratings are due to the removal of resolution or to the reduction of incongruity. Similar criticisms have been made by Godkewitsch (1974), Pien and Rothbart (1977), and Nerhardt (1977).

Another research approach that made use of authentic jokes is found in the work of Wicker, Thorelli, Barron, and Ponder (1981). These inves- tigators had subjects rate a series of jokes on a number of dimensions, including funniness, the degree to which the punch line was surprising, and the degree to which each joke made sense. The results showed that funniness ratings were significantly correlated with both "surprise" and "sense" ratings, which the authors viewed as reflecting incongruity and resolution, respectively. A factor analysis of the ratings showed that surprise, resolution, and funniness all loaded on the same factor, a finding that the authors viewed as support for the incongruity-resolution model.

However, it could also be argued that the results of Wicker et al. may be due to a confounding of incongruity and resolution at a stimulus level. This difficulty seems inherent in the use of subject ratings obtained following the presentation of entire, unaltered jokes. Given the theorized almost- instantaneous perception of incongruity and resolution in humor (Long & Graesser, 1988), it seems unlikely that subjects examining complete jokes would be able to evaluate subsequently the processes of incongruity and resolution independently of the total context of the joke. In addition, this approach does not provide for a precise definition of either incongruity or resolution but, instead, leaves that up to the subjects themselves.

While a naturalistic approach to the study of humor undoubtedly has a great deal to offer, Godkewitsch (1974) has pointed out that humor researchers have generally neglected the use of artificial "jokes," in which a large number of cognitive, affective, and ecological aspects of a joke that contribute to funniness can be separated and quantified. Accordingly, God- kewitsch argued that more artificial stimuli should be used in research on humor, allowing for precise definitions, measurements, and manipulation

Domains-Interaction Approach to l lumor

of the aspects under investigation. While these experimental "jokes" are certainly less humorous than more naturalistic ones, the reduction of ecological validity in such an approach could be considered to be offset by the increased precision and methodological rigor.

In keeping with these suggestions, Godkewitsch (1974) conducted several experiments in which he made use of simple adjective-noun pairs as humor stimuli. He selected the nouns and adjectives from a list of words that had previously been subjected to semantic differential analyses. Factor loadings were available for each word on Osgood's three orthogonal factors of Activity, Evaluation, and Potency (of. Osgood, Suci, & Tannenbaum, 1957). By using these factor loadings, Godkewitsch was able to compute the total semantic distance between the adjectives and nouns. To simplify the analyses, semantic distance was assessed in terms of whether the two words shared the same sign (i.e., positive or negative factor loading) on zero, one, two, or three factors. For example, in the adjective-noun pair "happy child," in which both words shared the same sign on all three factors, the pair had a low semantic distance, whereas "wise egg" had an intermediate level of distance, and "hot poet" had high semantic distance. Thus, the incongruity present in each adjective-noun pair was quantified in terms of the semantic distance between the two terms.

Subjects were asked to rate each adjective-noun pair on a number of scales, including funniness and wittiness. The results revealed that these humor ratings were significantly related to the semantic distance in the word pairs, such that the pairs with greater distance received higher humor ratings. Thus, Godkewitsch was able to demonstrate the importance of incongruity in humor, using artificial stimuli that allowed for precise quantification and manipulation of incongruity. Although the adjective-noun pairs were clearly much less humorous than authentic jokes, the humor that they did evoke was systematically related to structural properties that could be readily measured. However, a major drawback in Godkewitsch's method is that it does not offer a way of quantifying the resolution of incongruity. If overall semantic distance is used to measure incongruity, then it is not readily clear how a resolution component might be added to the design.

To summarize the discussion thus far, a number of research approaches have been taken to investigate the role of incongruity and resolution in humor. Although support for cognitive theories has been found, each approach suffers from certain limitations. Studies that make use of naturalistic jokes (e.g., Schultz, Wicker et al.) are unable to quantify and manipulate incongruity and resolution independently. On the other hand, studies using artificial stimuli (e.g., Nerhardt, Godkewitsch) have rigorously quantified and manipulated incongruity, but have thus far not provided a


means to assess the resolution of incongruity in an equally precise manner. As a result, the latter research approaches have tended to favor the incongruity-only theory of humor.

The approach taken in the present research is very similar to that of Godkewitsch (1974) but has the additional advantage of providing for the assessment of the independent and interactive contributions of both incongruity and resolution. This approach is an adaptation of the domains-interaction model, which was originally developed to account for the cognitive processing of metaphors. We now turn to a discussion of this model.

THE DOMAINS-INTERACTION APPROACH TO METAPHOR

Tourangeau and Sternberg (1981, 1982) have theorized that a metaphor involves a comparison between two concepts (a topic and a vehicle) as well as between the two different higher-order domains or conceptual categories to which the concepts belong. In understanding a metaphor, an individual must compare the similarity of the two higher-order domains as well as the similarity of the relative positions of the two concepts within their respective domains. Therefore, in operational terms, a metaphor is considered particularly apt if the domains are discrepant from one another in semantic space (i.e., there is high between-domain distance) and/or if the concepts occupy similar positions within their respective domains (i.e., low within-domain distance).

We can illustrate this model with the metaphor "Saddam Hussein is the buzzard of world leaders." The two concepts being compared in this metaphor are Saddam Hussein (the topic) and a buzzard (the vehicle), while their domains are world leaders and birds, respectively. In assessing the aptness of this metaphor, one would first note the degree of discrepancy between the two domains on some domain-distinguishing dimension. For example, on the dimension of humanness, world leaders and birds may be viewed as quite discrepant. Subsequently, the domain of the vehicle (birds) is used as a template through which the domain of the topic (world leaders) is viewed. A common within-domain dimension is found that allows a comparison of the relative position of topic and vehicle within their respective domains. In the example, a common within-domain factor might be the like-dislike (evaluative) dimension. Since buzzards tend to be disliked within the domain of birds and Saddam Hussein is similarly disliked within the domain of world leaders, the topic and vehicle tend to be close to one another on this within-domain dimension.

As this example demonstrates, the domains-interaction approach makes some basic assumptions about the way in which concepts and their

Domains-lnteraction Approach to llumor

domains are represented cognitively. First, it is assumed that information is represented in a multidimensional semantic space similar to a Euclidean space. Concepts are represented in terms of points that can be located in this semantic space by determining their position on certain dimensions or factors. The degree of similarity or dissimilarity between any pair of concepts can be calculated on the basis of the distance between them along these dimensions in semantic space.

Second, it is assumed that semantic space is organized in a hierarchical manner, with more basic concepts nested within higher-order domains or classes. Thus, the concept Saddam Hussein is located within the domain of world leaders, which in turn is located within the higher class of humans. Third, it is assumed that different dimensions in semantic space are relevant at different levels in the hierarchy. Some dimensions tend to distinguish between higher-order classes or domains, such that all members of a domain would tend to be located at the same point on such a dimension. An example of such a dimension is living-nonliving. The discrepancy between two domains on such dimensions is referred to as between-domain distance. In this way, the domain of "Cars" might be at the "nonliving" end of this dimension, while the domain of "Birds" would be at the opposite pole. Other dimensions tend to distinguish between members within a given domain, and are applicable to members of many different domains. On these dimensions, concepts from several different domains may cluster to- gether, while concepts within a given domain will be spread across the entire dimension. An example of such a dimension is size. The discrepancy between concepts on such dimensions is called within-domain distance. For example, on the size dimension, "Eagle," a large member of the "Birds" domain, and "Cadillac," a large member of the "Cars" domain, might be next to one another at the "large" pole of the size dimension. However, "Hummingbird," a small member of the "Birds" domain, and "VW beetle," a small car, would be placed in semantic space at the "small" pole of the size dimension.

A study by Trick and Katz (1986) demonstrates the type of research generated by the domains-interaction model. These researchers used metaphors of the form, "A is the B of A's domain" (e.g., "Woody Allen is the Volkswagen Beetle of actors"). The terms (topics, vehicles, and their domains) of these metaphors were first rated on semantic differential scales. A factor analysis of these ratings revealed four main factors. Two of these factors were found to be domain-distinguishing. That is, within either of these factors, domain members tended to have similar loadings. For each metaphor, between-domain distance was calculated by using the differences in factor loadings for the respective domain names of the topic and vehicle on these two factors. The other two factors were found to be domain-in-

8 Ilillson and Mart in

sensitive. That is, members of a given domain had a wide range of factor loadings on these factors. For each metaphor, within-domain distance was calculated by using the differences in factor loadings of the topic and vehicle on these two factors.

The metaphors were then rated by subjects on a number of dimensions, including aptness, comprehensibility, and liking. As predicted by the domains-interaction theory, metaphors with greater between-domain distance were rated as being more comprehensible and more apt or fitting and were liked more than metaphors with lower between-domain distance. This relationship held even with the effects of within-domain distance partialled out. Similarly, metaphors with a lower within-domain distance were rated as being more comprehensible and more apt or fitting and were liked more than metaphors with greater within-domain distance. This relationship remained even with the effects of between-domain distance partialled out.

The relevance of the domains-interaction approach to humor research is readily apparent. The definition of semantic distance or discrepancy used in this approach is conceptually very similar to the way in which humor theorists have described incongruity. Methodologically, the use of semantic differential scaling to measure such discrepancies is very similar to the method used by Godkewitsch (1974) to quantify incongruity in humor. The added advantage of this approach is that it allows for the precise measurement of two different types of semantic distance, which, we suggest, may be identified with incongruity and resolution, 4 respectively. In particular, we suggest that between-domain distance, with its assessment of the degree to which topics and vehicles differ from one another at a broad level of abstraction, could be taken as a quantification of incongruity. On the other hand, within-domain distance, which assesses the degree to which topics and vehicles resemble one another despite their differences, we take as a method of quantifying the resolution of incongruity.

This approach may also be considered as corresponding with Suls' model of humor appreciation. According to Suls' (1972, 1977, 1983) incongruity-resolution model of humor appreciation, in processing a "joke" such as "Sylvester Stallone is the Trans Am of actors," the listener might first attend to the incongruity in the sentence, as a member of the actor domain who is alive and breathing has been compared to an automobile, a nonliving, nonsentient object from the domain of cars. In this example, the degree of incongruity perceived is conceptualized in the domains-interaction approach as the semantic distance between the domain of actors and

4While we have chosen to use the more familiar term, it should be noted that the concept of "resolution" is probably more accurately described as resolution potential in the present studies, because of the possible absence of incongruity in a given joke and the subsequent irrelevance of resolution for that joke.

Domains-Interaction Approach to l lumor

that of cars (between-domain distance). Because of the fact that actors and cars are far apart in semantic space, this comparison presents a certain degree of incongruity, to which the listener would respond with surprise or puzzlement (e.g., "Why is an actor being compared to a car?"). Follow- ing this surprise, the listener would search for a cognitive "rule" that would enable one to make some sense of this unexpected association. In this simple joke, such a rule might be the idea that within their respective categories, both Sylvester Stallone and a Trans Am are seen as big and powerful and possibly a little unsavory. In the domains-interaction approach, this resolution of incongruity is conceptualized as the degree to which Sylvester Stailone and a Trans Am occupy the same area of semantic space within their own domains (within-domain distance). Therefore, while Sylvester Stallone and a Trans Am can be considered to be dissimilar in that one is an actor and the other is a car, they can also be seen as similar because, within their respective categories, they share some of the same characteristics. Recent modifications to Suls' model incorporating parallel processing features (Long & Graesser, 1988) may also be captured in the domains-interaction approach, as the processes relating to the perception of incongruity and resolution likely occur in parallel. The advantage of the domains-interaction approach is that their effects can be separated empirically.

In keeping with these formulations, the present research used a domains-interaction methodology similar to that of Trick and Katz (1986), employing structurally simple "jokes" to model the effects of incongruity and resolution on humor appreciation. The "jokes" were in fact constructed in a manner identical to the metaphors used by Trick and Katz, each com- bining two concepts in the form "A is the B of A's domain." The subjects rated these jokes for how humorous they were. Subject ratings of each of the concepts and their domains on semantic differential scales allowed for the computation of between- and within-domain distance for each pair of concepts used in the jokes. Thus, the procedure was almost identical to that of Trick and Katz (1986), except that the stimuli were presented to subjects as jokes rather than metaphors, and they were rated for humor rather than aptness.

This application of the domains-interaction model to humor points to some similarities in the structure and nature of metaphors and humorous stimuli. Indeed, metaphor theorists have debated the relative importance of similarity and dissimilarity in much the way that humor theorists have discussed incongruity and resolution (cf. Tourangeau & Sternberg, 1982). Both metaphors and humorous stimuli are seen as presenting an incongruity that requires resolution through some sort of problem-solving process in order to be comprehended. At the same time, there are some differences


between metaphor and humor that lead to different predictions. In par° titular, metaphors emphasize the importance of similarities between two concepts being present in comparisons, although they seem to be enjoyed more if the concepts are drawn from quite disparate domains. In metaphor, similarity is typically seen as a necessary and sufficient condition for metaphor aptness. In contrast, while it is improved by similarities between compared concepts, humor seems to depend particularly on the discrepancy or incongruity between two concepts. Humor theorists agree that resolution to the incongruity of a joke is certainly not sufficient for humor and, according to some theorists, may not even be necessary for humor to occur. In addition, even for incongruity-resolution theorists, resolution without incongruity will not be humorous. This difference in the importance of similarity in metaphor and in humor is therefore likely a fundamental and defining difference between the two. Metaphors require similarity to be apt, while jokes benefit from similarity (resolution) only if incongruity is present and may be humorous without similarity at all.

Thus, in the present research, we predicted that ratings of the humorousness of the jokes would be positively correlated with between-domain distance (i.e., degree of incongruity), since incongruity alone may be sufficient for humor. However, in contrast to the metaphor research, we predicted no simple correlation between humor ratings and within-domain distance (resolution), since resolution alone is not sufficient for humor. Fi- nally, we also predicted an interaction involving between-domain distance and within-domain distance in predicting humor ratings, such that jokes with both high between-domain distance (incongruity) and low within-domain distance (resolution) would be rated as most humorous. This finding would demonstrate that while similarity (resolution) may add to a joke's level of humor, it does so only when incongruity is present in the joke.

STUDY 1

Method

Subjects

Forty-five University of Western Ontario students participated in this study in partial fulfillment of their introductory psychology course requirements. Data from five subjects were dropped from the analyses because of extensive missing data. Of the 40 subjects remaining, 23 were male and 17 female.

Domains-Interaction Approach to tlumor 11

Materials

Each subject completed a scaling booklet and a joke rating booklet . Scaling Booklets. Scaling booklets were used to compute the semant ic

distances between each pair of concepts in the jokes. Semantic differential ratings were used to accomplish the scaling, as in the studies by Tourangeau and Sternberg (1981) and by Trick and Katz (1986). The nouns to be used in the jokes were selected from six domains: actors, world leaders, birds, cars, foods, and newspapers and magazines. The first four of these domains were used in the Trick and Katz (1986) study, while the last two were selected f rom a study by Kelly and Keil (1987). Five members f rom each of the actors and world leaders domains, and four members from each of the other four domains were used. These 26 nouns and the six domain names were rated in the scaling booklets; they are listed in Table I.

Table 1. Nouns Used in the Jokes and Their Respective Domains

Noun Domain

John Candy Actors Tom Cruise Actors Sylvester Stallone Actors Bill Cosby Actors Woody Alien Actors

Mikhail Gorbachev World George Bush World Manuel Noriega World Pope John Paul II World Brian Mulroney World

Falcon Birds Crow Birds Sparrow Birds Buzzard Birds

Rolls-Royce Cars Volkswagen Beetle Cars Pontiac Tram Am Cars Studebaker Station Wagon Cars

Fast food hamburger Foods Steak and potatoes Foods Quiche Foods Spinach Foods

The Globe and Mail National Enquirer Sports Illustrated People Magazine

leaders leaders leaders leaders leaders

Newspapers and magazines Newspapers and magazines Newspapers and magazines Newspapers and magazines


Table Ii. Semantic Differential Scales Used to Rate Joke Terms

Lively 1 2 3 4 5 6 7 8 9 Not lively Not thinking 1 2 3 4 5 6 7 8 9 Thinking

Predatory 1 2 3 4 5 6 7 8 9 Not predatory Ugly 1 2 3 4 5 6 7 8 9 Beautiful Tiny 1 2 3 4 5 6 7 8 9 Immense

Passive 1 2 3 4 5 6 7 8 9 Violent Humanlike 1 2 3 4 5 6 7 g 9 Not humanlike

Likable 1 2 3 4 5 6 7 8 9 Not likable Fragile 1 2 3 4 5 6 7 8 9 Tough

Reserved 1 2 3 4 5 6 7 8 9 Expressive Heavy 1 2 3 4 5 6 7 8 9 Light Classy 1 2 3 4 5 6 7 8 9 Not classy

Not cute 1 2 3 4 5 6 7 8 9 Cute Powerful 1 2 3 4 5 6 7 8 9 Powerless

Small 1 2 3 4 5 6 7 8 9 Large Fast 1 2 3 4 5 6 7 8 9 Slow Bad 1 2 3 4 5 6 7 8 9 Good

At the top of each page in the scaling booklet was 1 of the 32 nouns or domain names, followed by 19 9-point semantic differential rating scales. These scales were very similar to those used by Trick and Katz (1986), with some minor modifications designed to improve clarity and reliability of ratings: These scales had been selected to reflect dimensions relevant to similarities and differences between the concepts that were to be rated. Seventeen of these scales were to be used in the factor analysis to determine semantic distances; these are presented in Table II. Two additional scales (Familiar/Unfamiliar, and Funny/Not Funny) were included so that these dimensions could be partialled out of subsequent analyses.

The 26 nouns to be used in the jokes appeared first in the scaling booklets, followed by the 6 domain-names. The 26 nouns were presented to the subjects in one of four random orders. The order of the scales for semantic differential ratings did not vary across stimuli or subjects.

Joke Rating Booklets. The joke rating booklets contained jokes made from combinations of a noun from either the actors or world leaders domain followed by one of the remaining 25 nouns from the 6 domains. The jokes were of the form A is the B of Domain of A (e.g., "John Candy is the fast food hamburger of actors"). Each of the 10 concepts from the actors

5Trick and Katz (1986) used a semantic differential rating of "Animate-Inanimate" that was found to discriminate between stimuli less effectively than the other scales used in their study. In order to improve stimulus discrimination, we substituted the scale "Reserved-Expressive" for the previous, less effective scale.

Domains-Interaction Approach to l tumor 13

and world leaders domains was used in combination with each of the remaining 25 concepts, for a total of 250 jokes. Thus, the first term in each joke was always a human, while the second term could be either human or nonhuman. To control for possible context and order effects, four versions of the joke rating booklet were made, each with a different random- ized ordering of the 250 jokes, and these were randomly assigned to the subjects. Each joke in the booklets was followed by a 9-point scale assessing a Not Humorous/Humorous dimension.

Procedure

The subjects were briefly informed of the general nature of the study and of the procedure they were to follow. They were allowed 2 hr to complete both booklets and were encouraged to take a short break if they so desired. Subjects were tested in small groups, with an average group size of eight. To control for task order effects, the sequence in which the two booklets were completed was counterbalanced across subjects.

Results and Discuss ion

Computing Semantic Distance Between Joke Terms

The first step in computing the semantic distances between the pairs of nouns used in the jokes was to perform a factor analysis on the semantic differential ratings. This analysis was conducted by means of PARAFAC (Harshman & Lundy, 1984), a three-way factor analysis technique. By accounting for subject variances in factor structures, three-way factor analysis produces a unique factor solution and, thus, solves the problem of inde- terminacy of factor rotation. The PARAFAC model gives one of the sim- plest three-way extrapolations of two-way factor analysis, and it is particularly well suited to multidimensional scaling (Harshman & De Sarbo, 1984; Trick & Katz, 1986). PARAFAC produced a four-factor (scree plot selected) solution with orthogonality forced on the stimulus mode (R = .579, RSQ = .335, stress = .816). The solution replicated in nine of nine random starting positions and was, thus, demonstrated to be highly reliable. Details concerning the nature of three-way factor analysis and the proce- dures involved in a PARAFAC analysis are presented by Harshman and Lundy (1984).

Of the four factors, two were found to be domain distinguishing. In other words, members of the same domains tended to have similar loadings on these two factors, while the domain-names themselves remained rela-

14 tlillson and Martin

tively distant from one another. The first of these, Factor II, was dominated by the Humanlike scale, with actors and world leaders loading at one end of the dimension and foods and cars at the other end. The other, Factor IV, had high loadings for the Beauty and Speed scales, with actors and cars loading at one end and foods and world leaders at the other. The between-domain distance for each joke was therefore calculated by finding the difference between factor loadings (averaged across subjects) on these two factors for the respective domain names of the two nouns in the joke. The Euclidean distance formula was employed, as follows: Distance (,4, B) = SQRT {[(loading of A - loading of B) 2 on factor II] + [(loading of A

- loading of B) 2 on factor IV]}. Factor I (predominately an Evaluation factor) and Factor III (domi-

nated by the Size scales) proved to be domain insensitive factors. In other words, these two factors tapped dimensions that cut across the domain boundaries, and measured characteristics that different members of the same domains might have to different degrees. For example, while "John Candy" and "Woody Allen" are both actors, they loaded very differently on the within-domain factor of Size. Therefore, within-domain distance for each joke was calculated by finding the distance between factor loadings (averaged across subjects) on these two factors for the two nouns in the joke, again using the Euclidean distance formula.

Familiarity and Funniness of Terms. In the scaling booklets, the subjects rated both the familiarity and the funniness of each noun used in the jokes. These ratings were to be used as control variables and were not employed in the factor analyses described above. Since differences in humor ratings of jokes may be partially related to the familiarity of the terms, it was thought important to control for this variable in subsequent analyses. Such a control procedure allowed for a more direct assessment of the effects of incongruity and resolution in humor appreciation. The mean familiarity rating for the nouns used was 2.57 (where 1 --- familiar, 9 = unfamiliar), indicating that subjects were reasonably familiar with these terms overall.

We also thought it was important to control for rated funniness of the nouns used in the jokes, as the nouns may have varying degrees of humorous connotations. For example, the comedian John Candy might be considered inherently more funny than Pope John Paul. Differences in humor ratings of jokes may be related in part to these sorts of stimulus characteristics of the nouns employed in the jokes. Since we were interested in examining humor ratings only in terms of semantic distances between concepts, we decided to control for rated funniness of the nouns in subsequent analyses. The mean funniness rating of the nouns was 5.47 (1 =

Domains-lnteraction Approach to Humor 15

funny, 9 -- not funny), indicating that, overall, these terms were seen as moderately funny.

Humor Ratings o f the Jokes. Each subject rated each of the 250 jokes on a 9-point humor scale. The average rating of joke humorousness was 3.41 ( S D - 0.583; 1 ffi not humorous, 9 ffi humorous), indicating a fairly low level of humor in these jokes overall.

Semantic Distance as a Predictor of Humor Ratings. A significant correlation, computed across the 250 jokes, was found between mean humor ratings and between-domain distances (r = .50, p < .001). 6 Thus, as predicted, jokes were rated as more humorous if there was greater between- domain semantic distance between their concepts, which we conceptualized as a measure of incongruity. In contrast, the correlation between mean humor ratings and within-domain distances was nonsignificant (r = -.01). Thus, as expected, within-domain semantic distance, which we conceptualized as a measure of incongruity resolution, was not by itself related to humor ratings. In addition, between- and within-domain distance themselves were not significantly correlated (r = -.012), indicating that the domains-interaction approach taken in this study does not suffer from the confounding of incongruity and resolution that has occurred in past humor research (Schultz, 1972, 1974; Wicker et al., 1981).

In order to examine the interaction, as well as to partial out familiarity and funniness of the joke concepts, a hierarchical multiple regression analysis was computed across the 250 jokes. Mean ratings of humor (averaged across subjects) was the dependent measure in this analysis. A familiarity and funniness score was computed for each joke by summing the means of each of these ratings for both of the nouns used in the joke. These two variables were first entered into the regression equation as a block. This block revealed no significant relationship with the ratings of humor [RSO change = .014; F(2,247) change = 1.80, p = .17]. The stimulus characteristics of familiarity and funniness therefore did not explain a significant amount of the variance in the rated humorousness of the jokes.

Within-domain distance was entered into the equation next. As seen in the simple correlation above, this variable did not produce a significant

6The level of significance of statistical calculations in these studies was assessed using the number of jokes (i.e., 250) as the basis for calculating degrees of freedom. However, it is important to note that these jokes are not statistically independent, given that they are based on combinations of only 26 concepts. Therefore 250 provides a relatively liberal estimate of degrees of freedom. Unfortunately, there is no clear method for determining appropriate degrees of freedom here. Nevertheless, when the data were analyzed using the number of subjects in each study (40 and 38) as an overly conservative estimate of degrees of freedom (Trick & Katz, 1986), all predicted relationships remained significant, with the exception of the between-domain by within-domain interactions for the complete set of jokes, which were marginally significant (p < .10).


change in variance accounted for [RSQ change = .001; F(1,246) change = 0.24, p = .62]. In addition, a 95% confidence interval for the within-domain distance beta term included 0 (-.05 < ~ < .09), indicating that within- domain distance did not exert a simple effect on ratings of humor appreciation. Between-domain distance was the next variable entered into the equation, revealing a significant effect [RSQ change = .25; F(1,245) change = 85.03, p < .0001]. Thus, the relationship seen in the simple correlation held even with the effects of familiarity, funniness, and within-domain distance removed from the analysis.

Finally, the interaction between within- and between-domain distance was examined by entering the product of these two variables into the equation. In support of our predictions, a significant interaction was revealed [RSQ change = .016; F(1,244) change = 5.547, p = .019]. The direction of this interaction was examined by computing the predicted humor ratings for jokes with high and low between- and within-domain distances (approximately 1 SD above and below the mean on each of these variables), using the beta weights obtained in the regression equation. As predicted, the jokes with the highest mean ratings of humor were those with a high between-domain distance (high incongruity) and a low within-domain distance (high resolution).

Despite the significance of the interaction involving between- and within-domain distance, the interaction term was quite weak, accounting for less than 2% of the variance in humor ratings. We hypothesized that this may have been at least partly due to the large number of jokes with mod- erate levels of within- and between-domain distance. If the effects occur most strongly at the high or low ends of the ranges of semantic distances, as suggested by (3odkewitsch's (1974) findings, then perhaps an interaction effect might be somewhat suppressed by using the entire data set.

To test this hypothesis, an extreme-groups analysis was conducted, using only the jokes in the top and bottom third of the distributions for both between- and within-domain distance. Jokes with intermediate values on either of these two dimensions were dropped, resulting in a total of 89 jokes to be used in the analysis. Jokes with low values on a given dimension were coded as 1, while those with high values were coded as 2. In this way, this selection produced the equivalent of a 2 x 2 ANOVA design, with two levels for each of between- and within-domain distance. A hierarchical multiple regression analysis was again performed using humor ratings as the dependent variable.

The results of this analysis revealed no significant effect of the Fa- miliarity and Funniness ratings [RSQ change = .0007; F(2,86) change = 0.03, p < .10] and a nonsignificant main effect for within-domain distance [RSQ change = .0128; F(1,85) change = 1.105, p > .10]. The main effect of between-domain distance was enhanced in this extreme group analysis

Domalns-lnteraction Approach to Humor 17

[RSQ change = .3359; F(1,84) change = 43.377, p < .00001], as was the interaction between within- and between-domain distance [RSQ change = .0412; F(1,83) = 5.606, p = .020]. Thus, using jokes that were particularly high or low in semantic distances, the interaction between within- and between-domain distances accounted for about 4% of the variance.

The cell means of humor ratings for jokes with high versus low within- and between-domain semantic distances in this extreme-groups analysis are shown in Table III. A Duncan multiple-range test showed that the mean humor ratings for jokes with high between-domain distance (regardless of within-domain distance) were significantly higher than those with low between-domain distance (p < .01). Thus, jokes with incongruity were rated as more humorous than those without incongruity. In addition, among jokes with high between-domain distance, those with low within-domain distance were rated as significantly more humorous than were those with high within-domain distant (p < .05). Thus, jokes with both incongruity and resolution were rated as more humorous than all other jokes.

An example of a joke that received a high mean humor rating is "Woody Allen is the quiche of actors." The semantic distance between the domain of Woody Allen and that of quiche (actors and foods, respectively) is quite large (high incongruity). At the same time, Woody Allen and quiche have similar loadings on the two factors that are not domain distinguishing, indicating that they tend to occupy a similar position within their respective domains (high resolution). In contrast, "Woody Allen is the Pontiac Trans Am of actors" was not as funny. Although the domains of Woody Allen and Pontiac Trans Am (actors and cars, respectively) are quite distant (high incongruity), their within-domain distance is also quite large, so that there is less resolution of the incongruity.

The results of this study provided support for the domains-interaction approach as a model of cognitive aspects of humor. As predicted, humor ratings of simple jokes were significantly related to their between-domain semantic distances (incongruity) but not to their within-domain distances (resolution). In addition, an interaction of these two variables was observed, with

Table III. Study 1: Cell Means of Humor Ratings of Jokes with Within-Domain and Between-Domain Distances in the Highest and

Lowest Thirds of the Distribution

Between-domain distance

Low High

Low within-domain distance 2.69 3.79 High within-domain distance 3.06 3.59


jokes rated the most humorous being those with high between-domain and low within-domain distances (both incongruity and resolution). These results are independent of the familiarity and inherent funniness of the individual topics and vehicles, as these effects were partialled out of the analyses.

STUDY 2

The second study was designed as a replication of Study 1, to ascer- tain the reliability and generalizability of the results. While the results of Study 1 were encouraging, the interaction effect was quite weak, and a replication of these findings would increase confidence in the usefulness of this methodology as a model of cognitive aspects of humor. Also in the previous study, the semantie differential ratings of the concepts used in the jokes and the humor ratings of the jokes were all obtained from the same group of subjects. In order to determine the generalizability of these findings, Study 2 made use of the semantic distances computed from the semantic differential data provided by the subjects in Study 1. A new sample of subjects was used to obtain humor ratings of the jokes. From a theoretical perspective, a replication of the results of Study 1 would indicate that the arrangement of these concepts in semantic space is somewhat uni- versal, at least within the population of undergraduates from which our samples were drawn. From a more practical viewpoint, this would indicate that the same factor loadings could be used in future research without needing to obtain new semantic differential data in every study. Metaphor research using the domains-interaction approach has demonstrated similar generalizability (Tourangeau & Sternberg, 1982).

Subjects

Forty-four University of Western Ontario students participated in the study for partial fulfillment of their introductory psychology course requirements. Data from six subjects were dropped from the analysis because of extensive missing data. Of the 38 subjects remaining, 20 were male and 18 female.

Procedure

The procedure of the second study was identical to that of the first, except that the subjects completed only the joke rating booklet, and not the scaling booklet. The same joke rating booklets were used as in the first

DomalnsAnteraction Approach to Humor 19

study, with four different random orders of the jokes distributed among the subjects. The subjects were tested in groups, with an average of eight in each group.

Results and Discussion

The analyses for this study used the factor solution from the previous study in the calculation of between- and within-domain distances for each joke. The mean Familiarity and Funniness ratings of the joke nouns obtained from the previous study were also used as control variables in the present study. As in the first study, the subjects in Study 2 found the jokes, overall, to be minimally humorous (M = 3.525, SD = 0.535; 1 = not humorous, 9 = humorous).

Semantic Distance as a Predictor of Humor Ratings

A significant relationship was again found between mean humor rating and between-domain distance, computed across the 250 jokes (r = .36; p < .001). The simple correlation between humor ratings and within-domain distance was again nonsignificant (r = -.09).

As in the previous study, a hierarchical multiple regression analysis was computed across the 250 jokes, with mean humor rating as the dependent variable. The control variables of Funniness and Familiarity of concepts were first entered as a single block. This block had a significant effect on ratings of humor [RSQ change = .024; F(2,247) change = 3.085, p < .05]. Thus, in this study the rated humorousness of the jokes was related to the familiarity and funniness of the nouns used in the jokes, as rated by the subjects in the preceding study.

Within-domain distance was entered into the equation next. As in the first study, this variable did not have a significant effect [RSQ change = 0.001; F(1,246) change = 0.383, ns]. In addition, a 95% confidence interval for the within-domain distance beta term included 0 (-.09 < 13 < .05), indicating that the simple effect of within-domain distance did not explain a significant amount of the variance in ratings of humor appreciation. In contrast, between-domain distance showed a significant increase in variance accounted for [RSQ change = .15; F(1,245) change = 42.871, p < .0001].

Finally, the interaction term, computed as the product of between- and within-domain distance, was also significant [RSQ change = .014; F(1, 244) change = 4.074, p < .05]. An examination of predicted humor ratings for jokes with high and low within- and between-domain distances again

20 Hllison and Martin

showed that the jokes with a high between-domain distance and a low within-domain distance were rated as most humorous. Thus, the results of this study were very similar to those of Study 1 and were in the predicted directions.

As in the first study, the interaction term was quite weak, again accounting for less than 2% of the variance in humor ratings. For this reason, an extreme groups analysis was again performed using a hierarchical multiple regression procedure with humor ratings as the dependent variable.

The results of this analysis revealed a marginally significant effect for the Familiarity and Funniness ratings [RSQ change = .053, F(2,86) change = 2.413, p < .10] and a nonsignificant main effect for within-domain distance [RSQ change = .02; F(1,85) change = 2.119, p > .10]. The main effect of between-domain distance was enhanced in this extreme group analysis [RSQ change = .225; F(1,84) change = 27.091, p < .00001], as was the interaction between within- and between-domain distance [RSQ change = .055; F(1,83) = 7.089, p = .009]. Thus, using jokes that were particularly high or low in semantic distances, the interaction between within- and between-domain distances accounted for about 5.5% of the variance.

The cell means are shown in Table IV. A Duncan multiple-range test of these means revealed that the mean for jokes with high between-domain distance and low within-domain distance (M = 3.89) was significantly higher than any of the other three cell means (p < .01). Thus, as predicted, jokes with a high incongruity (high between-domain distance) and resolution (low within-domain distance) were rated as most humorous. In addition, the mean for jokes with a high between-domain distance and a high within-domain distance (M = 3.54) was significantly higher than the two groups of jokes with a low between-domain distance (M = 3.01 and 3.25 for low and high within-domain distance, respectively; p < .01). Thus, jokes with incongruity but without resolution are more humorous than those without incongruity.

The second study replicated the findings of the first, using a different group of subjects to rate the jokes than those subjects whose ratings were

Table IV. Study 2: Cell Means of Humor Ratings of Jokes with Within-Domain and Between-Domain Distances in the Highest and



Low High


Domains-Interaction Approach to Humor 21

used to compute the semantic distance measures. These results provide support for the replicability of the obtained relationships, as well as the generalizability of the semantic differential factor structures to the ratings of a new subject group.

STUDY 3

The previous two studies provided support for the application of the domains-interaction approach as a model of incongruity and resolution in humor. Although this approach was originally developed in research on metaphors, the predictions made about humor are somewhat different from those in the metaphor area. In particular, previous metaphor research has supported the prediction of significant relationships between aptness ratings and both within-domain and between-domain distances, and no significant interaction (e.g., Trick & Katz, 1986). In contrast, these two studies have provided support for our prediction that ratings of humor would be related to between-domain distance but not within-domain distance and that the interaction would be significant. Thus, these two studies, when compared with previous metaphor findings, indicate that within-domain and between- domain distances have differential effects on aptness ratings of metaphors as compared to humor ratings of jokes.

However, since the "jokes" used in these studies are identical to the metaphors used in previous research, there may still be some doubt regarding the distinctiveness of the humor ratings in comparison with ratings of aptness of these (mildly humorous) metaphors. We therefore decided to conduct a third study in which subjects were presented with the same 250 sentences that were used as "jokes" in the first two studies. However, instead of presenting them as jokes to be rated for humorousness, they were presented as metaphors to be rated for aptness. We pre- d ic ted that the results would replicate the main effects for bo th within-domain and between-domain distance found in previous metaphor research, and not the pattern of results shown in our two previous humor studies. Thus, we expected that there would not be a significant interaction between between-domain and within-domain distance in predicting aptness ratings, as such an effect is not anticipated in metaphor studies using the domains-interaction approach. Results in keeping with our predictions could be taken as further evidence that we are studying something distinct from the metaphor phenomena reported in previous research.


Method

Subjects

Forty University of Western Ontario students participated in the study for partial fulfillment of their introductory psychology course requirements. Data from 1 subject were dropped from the analyses because of extensive missing data. Of the 39 subjects remaining, 23 were male and 19 female.

Procedure

The procedure of the third study was identical to that of the second, except that this study was presented to the subjects as research on metaphors, the 250 sentences were presented as metaphors instead of as jokes, and they were rated on a 9-point scale for aptness rather than humorousness. The same topic-vehicle pairs were used as in the first and second studies, with four different random orders of the metaphors distributed among the subjects. The subjects were tested in groups, with an average of eight in each group.

Results and Discussion

As in the second study, the analyses for this study used the factor solution from the first study to calculate between- and within-domain distances for each metaphor. The mean Familiarity and Funniness ratings of the nouns obtained from the previous study were also used as control variables in the present study. As in past metaphor research of this type, the subjects in Study 3 found the metaphors to be minimally apt overall (M = 2.83, SD ffi 0.73; 1 ffi not apt, 9 = apt).

Semantic Distance as a Predictor of Aptness Ratings

A significant correlation, computed across the 250 metaphors, was found between mean aptness ratings and between-domain distances (r ffi .46, p < .001). Thus, in keeping with the domains-interaction approach to metaphor, metaphors were rated as being more apt if there was greater between-domain semantic distance between their concepts. Likewise the correlation between mean aptness ratings and within-domain distances was significant (r = -.25, p < .001). Thus, as in previous research, metaphors with less within-domain distance were rated as more apt. Importantly, this

Domains-Interaction Approach to l lumor 23

finding contrasts with the findings of the previous two studies in which within-domain distance was not significantly related to humorousness ratings of the same stimuli.

A hierarchical regression analysis was performed, with aptness ratings as the dependent measure. The control variables of Funniness and Famili- arity, entered first as a block, did not have a significant effect on ratings of aptness [RSQ change = .004; F(2,247) change = .495, p = .61]. Within- domain distance was entered into the equation next. In keeping with our predictions and in contrast to the results from the two previous humor studies, within-domain distance exerted a significant main effect on ratings of metaphor aptness [RSQ change = 0.067; F(1,246) change = 17.68, p = 0.0001]. Also as expected, between-domain distance showed a significant effect [RSQ change = .212; F(1,245) change = 72.327, p < .0001]. Finally, the interaction term, computed as the product of between- and within-domain distance, was not significant [RSQ change = .0006; F(1,244) change = .210,p > .5]. This is again in keeping with our predictions and in contrast to the results of the first two studies where humorousness served as the dependent measure.

For the sake of completeness, we also performed an extreme groups analysis identical to those in the two previous studies except with aptness as the dependent measure. The results of this analysis revealed no significant effect for the Familiarity and Funniness ratings [RSQ change = .004; F(2,86) change = .182, p < .10]. The extreme group selection strengthened the main effects for within-domain distance [RSQ change = .11; F(1,85) change = 11.026,p < .001] and for between-domain distance [RSQ change = .276; F(1,84) change = 38.316, p < .00001]. However, as expected the interaction remained nonsignificant [RSQ change = .0004; F(1,83) = .05611, p > .5].

The cell means of aptness ratings for metaphors with high versus low within- and between-domain semantic distances in this extreme-groups analysis are shown in Table V. A Duncan multiple-range test of these cell means revealed that metaphors with a high between-domain distance were rated as more apt regardless of whether they had a high or a low within- domain distance (p < .01). Metaphors with a high between-domain and a low within-domain distance were also rated as being more apt than those with a low between-domain and a high within-domain distance (p < .01).

The results of this third study conformed to our expectations. When the 250 sentences were presented as metaphors, a pattern of results con- sistent with previous metaphor research appeared in the aptness ratings. Both within-domain and between-domain distance were shown to have main effects, while these semantic distance terms did not interact in their effects on ratings of metaphor aptness. This pattern contrasted with the


Table V. Study 3: Cell Means of Aptness Ratings of Metaphors with Within-Domain and Between-Domain Distances in the Highest and



Low High


humorousness ratings of the first two studies, in which only between-domain distance showed a main effect, and the two types of semantic distance interacted in their effects on ratings of humor appreciation. Thus, the findings of this third study provide a validation of the use of the domains-interaction approach in the study of humor comprehension and appreciation as distinct from research on metaphors.

GENERAL DISCUSSION

The purpose of these studies was to attempt to model incongruity and resolution in basic humor processing using the methodology of the domains-interaction approach to metaphor. The results of the studies were quite encouraging. They demonstrated that the rated humorousness of simple "jokes," of the form "Noun A is the Noun B of A's Domain," can be at least partially predicted from the semantic distances between the nouns used in the jokes. Of particular interest in this approach is the fact that two different types of semantic distance can be computed, providing a means of quantifying both incongruity and resolution independently of one another. In addition, this approach provides a clean assessment of incongruity and resolution effects in humor appreciation by allowing for the statistical removal of extraneous effects such as topic and vehicle funniness and familiarity. The third study also demonstrated that these simple jokes are processed differently than metaphors of the same structure.

The between-domain distance for each joke was computed by taking the Euclidean distance between the factor loadings of the higher-order domains or categories of the two nouns on factors that distinguished between different domains. We hypothesized that this distance could be used in the model as a quantification of incongruity, since it relates to broad categorical discrepancies between concepts. As predicted, between-domain distance was significantly correlated with humor ratings of the jokes. Thus, jokes

Domains-Interaction Approach to tlumor 25

like "John Candy is the fast-food hamburger of actors," where concepts from two very different domains are compared (actors and food in this example), were rated as being more humorous than jokes where concepts from more similar domains (like actors and world leaders) are compared.

Within-domain distance was computed for each joke by taking the distance between the factor loadings of the two nouns on factors that distinguished between concepts within domains. Within-domain distance relates to the degree to which concepts share positions similar to each other within their respective domains. This distance was used as a measure of incongruity resolution, since it concerns the degree to which the two nouns share certain characteristics in common regardless of the discrepancy between their higher-order classes. As predicted, the within-domain distance did not produce a significant simple correlation with humor ratings of jokes. Resolution without incongruity is not expected to be humorous.

However, also as predicted, the two different types of semantic distance interacted with one another in predicting humor ratings, such that the jokes with a high between-domain distance (i.e., high incongruity) and a low within-domain distance (i.e., high resolution) were rated as most humorous. Admittedly, this interaction was quite weak, particularly when all 250 jokes were used in the analyses. However, when extreme groups analyses were done on the jokes with particularly high or low semantic distances, the interaction was stronger. These analyses also suggested that, although incongruity plus resolution makes for the most humorous jokes, incongruity without resolution is still more humorous than absence of incongruity.

A third stud)/showed that this pattern of results does not occur when the sentences are presented as metaphors rather than as jokes and are rated for aptness instead of humorousness. This finding suggests that even with these simple stimuli, some conceptual differences between humor appreciation and metaphor aptness can be represented empirically. It should be noted here that it was not the goal of this study to evaluate the appropri- ateness of the domains-interaction model for research on metaphor. In- deed, there is considerable debate among researchers in the metaphor area concerning appropriate models, and research in this area has recently taken new directions (el. Tourangeau & Rips, 1991). Ultimately, the usefulness of this approach in modeling incongruity and resolution in humor does not depend on its applicability in the metaphor field.

In sum, the results of these studies indicate that the domains-interaction approach may be a useful model of at least some of the cognitive aspects of humor. Unlike previous research using authentic, naturalistic jokes (e.g., Schultz, 1972; Wicker et al., 1981), this approach allows for very precise definition, quantification, and manipulation of incongruity and resolution. Although some past research has also made use of artificial stimuli


to quantify incongruity in a precise manner (e.g., Godkewitsch, 1974; Ner- hardt, 1970), these approaches have not permitted an equally precise quantification of the resolution of incongruity.

One interesting aspect of this method is that the factors relating to incongruity and resolution are presented simultaneously, although measured independently. Incongruity can be present at the same time as resolution, since they are measured on separate dimensions. Forabosco (1992), in discussing the manner in which incongruity and resolution interact in humor, has pointed out that incongruity must not be completely resolved for something to be humorous. Instead, the resolution must be only partially successful, leaving a certain amount of incongruity remaining. Thus, this simultaneous coexistence of incongruity and resolution, which seems to be necessary for humor, is also present in the model provided by the domains-interaction approach.

This does not mean, of course, that the domains-interaction model accounts for all factors relevant to humor. Indeed, the stimuli used in these studies were seen as being only minimally humorous by our subjects, although the humor ratings did vary systematically as predicted. Clearly, in most real-life humor, a large number of other factors contribute to whether a stimulus is perceived as humorous or not, including emotional and motivational themes present in the stimulus, personality differences, modes of presentation, contextual factors, and so on. While the domains-interaction approach is able to quantify one particular form of incongruity and resolution, two factors that contribute to humor appreciation, it is not a comprehensive model of humor.

In addition, we do not suggest that semantic distances between concepts are the only way that incongruity and resolution can occur in jokes or other humorous stimuli. In naturalistic humor, there may be many more complex ways in which incongruity results from discrepancies between cognitive representations of expectations and stimuli. For example, consider the following joke: "Doctor, come at once! Our baby has swallowed a foun- tain pen! .... I'll be right over. What are you doing in the meantime? .... Using a pencil." (Suls, 1983, p. 45). Here the incongruity arises from the fact that the response in the punch line does not conform to the usual "script" for a distressed parent responding to the doctor's query. However, this incongruity is resolved when we recognize the ambiguity in the doctor's question "What are you doing in the meantime?" and interpret it as "What are you using as a substitute for the pen?" Thus, in this example, incongruity and resolution arise from factors other than simple semantic distances between two concepts. Thus, we view the approach taken in this research as one potential model of the cognitive aspects of humor, and not as an exhaustive definition of incongruity and resolution in humor.

Domains-lnteraction Approach to Humor 27

The "jokes" used in the present research are certainly less complex than most naturalistic forms of humor. In particular, the incongruity represented by semantic distance between two nouns is likely less than the incongruity present in most real jokes. Similarly, the cognitive effort required to find a way to equate the two nouns on some within-domain factors is considerably less than the effort required to resolve the incongruity in most jokes like the one given above. This is likely one reason why the rated humor of these jokes was so low, since jokes that are too easy to resolve are less humorous than more complex and difficult ones, up to some optimal level of complexity (Zigler, Levine, & Gould, 1967; McGhee, 1979). In addition, the limited complexity of the jokes used in the present studies may account in part for why the interaction between within- and between- domain distance was so small. In jokes such as these, where the incongruity is relatively small, resolution may not be as important as it would be in jokes where there is a greater degree of incongruity. As a result of this reduced complexity, the importance of the interaction between incongruity and resolution might be also reduced.

Study 2 revealed that the semantic distances computed from semantic differential ratings obtained from one group of subjects could be used to predict humor ratings from another group of subjects. This finding provides some support for the generalizability of this research. Further research is needed, however, to determine whether these findings can be replicated using other verbal and even nonverbal stimuli presented in different formats. For example, combinations of visual stimuli might be employed in a similar manner . Also, the precise m e a s u r e m e n t and manipulation of variables offered by this general approach may make it useful for examining additional cognitive processes in humor, as well as other factors such as the emotional connotations of stimuli and various individual difference variables. For example, aspects of superiority or disparagement theories (of. Zillmann, 1983) might be incorporated into this basic model.

Despite the fact that the specific methodology used in the two reported studies might be limited to simple and artificial "jokes," the concepts of incongruity and resolution that were operationalized with this approach are generally considered to be common to most humor. Therefore, while the methodology of the domains-interaction approach probably cannot be applied to the study of very complex forms of humor, we suggest that the theoretical conclusions derived from this basic research approach may be applied to further our understanding of more sophisticated forms of humor. In this way, it may be possible to develop and test more comprehensive models of humor comprehension and appreciation.

28 Hiilson and Martin

R E F E R E N C E S

Forabosco, O. (1992). Cognitive aspects of the humor process: The concept of incongruity. HUMOR: International Journal of Humor Research, 5, 45-68.

Oodkewitsch, M. (1974). Correlates of humor: Verbal and non-verbal aesthetic reactions as functions of semantic distance within adjective-noun pairs. In D. E. Berlyne (Ed.), Studies in the new experimental aesthetics (pp. 279-304). Washington, DC: Hemisphere.

Harshman, R. A., & De Sarbo, W. S. (1984). An application of PARAFAC to a small sample problem, demonstrating preprocessing, orthogonality constraints, and split-half diagnostic techniques. In H. 0 . Law, C. W. Snyder, J. A. Hattie, & R. P. McDonald (Eds.) Research methods for muhimode data anatysis (pp. 602-642). New York: Praeger.

Harshman, R. A., & Lundy, M. E. (1984). The PARAFAC model for three-way factor analysis and multidimensional scaling. In H. O. Law, C. W. Snyder, J. A. Hattie, & R. P. McDonald (Eds.) Research methods for multimode data analysis. (pp. 122-215). New York: Praeger.

Kelly, M. H., & Keil, F. C. (1987). Metaphor comprehension and knowledge of semantic domains. Metaphor and Symbolic Activity, 2, 33-51.

Long, D. L., & Graesser, A. C. (1988). Wit and humor in discourse processing. Discourse Processes, 11, 35-60.

McGhee, P. E. (1979). Humor: Its origin and development. San Francisco: W. H. Freeman. Nerhardt, G. (1970). Humor and inclination to laugh: Emotional reactions to stimuli of

different divergence from a range of expectancy. Scandinavian Journal of Psychology, 11, 185-195.

Nerhardt, G. (1976). Incongruity and funniness: Towards a new descriptive model. In A. J. Chapman & H. C. Foot (Eds.), Humour and laughter: Theory, research and applications (pp. 55-62). London: John Wiley & Sons.

Nerhardt, G. (1977). Operationalization of incongruity in humor research: A critique and suggestions. In A. J. Chapman & H. C. Foot (Eds.), It's a funny thing, humour (pp. 47-51). Oxford: Pergamon Press.

Osgood, C. E., Suci, O. J., & Tannenbaum, P. H. (1957). The measurement of meaning. Urbana, IL: University of Illinois Press.

Pien, D., & Rothbart, M. K. (1977). Measuring effects of incongruity and resolution in children's humor. In A. J. Chapman & H. C. Foot (Eds.), It's a funny thing, humour (pp. 211-213). Oxford: Pergamon Press.

Rothbart, M. K. (1976). Incongruity, problem-solving and laughter. In A. J. Chapman & H. C. Foot (Eds.), Humour and laughter: Theory, research and applications (pp. 37-54). London: John Wiley and Sons.

Schultz, T. R. (1972). The role of incongruity and resolution in children's appreciation of cartoon humor. Journal of Experimental Child Psychology, 13, 456-477.

Schultz, T. R. (1974). Development of the appreciation of riddles. Child Development, 45, 100-105.

Schultz, T. R., & Horibe, F. (1974). Development of the appreciation of verbal jokes. Developmental Psychology, 10, 13-20.

Suls, J. M. (1972). A two-stage model for the appreciation of jokes and cartoons: An information-processing analysis. In J. H. Goldstein & P. E. McGhee (Eds.) The psychology of humor: Theoretical perspectives and empirical issues (pp. 81-100). New York: Academic Press.

Suls, 3. M. (1983). Cognitive processes in humor appreciation. In P. E. McGhee & 3. H. Goldstein (Eds.) Handbook of humor research, WoL 1. Basic Issues (pp. 39-57). New York: Springer-Verlag.

Tourangeau, R., & Rips, L. (1991). Interpreting and evaluating metaphors. Journal of Memory and Language, 30, 452-472.

Tourangeau, R., & Sternberg, R. J. (1981). Aptness in metaphor. Cognitive Psychology, 13, 27-55.

Domains-Interaction Approach to Humor 29

Tourangeau, R., & Sternberg, R. J. (1982). Understanding and appreciating metaphors. Cognition, 11, 203-244.

Trick, L., & Katz, A. N. (1986). The domain interaction approach to metaphor processing: Relating individual differences and metaphor characteristics. Metaphor and Symbolic Activity, 1, 203-244.

Wicker, F. W., Thorelli, I. M., Barron, W. L., & Ponder, M. R. (1981). Relationships among affective and cognitive factors in humor. Journal of Research in Personality, 15, 359-370.

Wilson, C. P. (1979). Jokes: Form, content, use and function. London: Academic Press. Zigler, E., Levine, J., & Gould, L. (1967). Cognitive challenge as a factor in children's humor

appreciation. Journal of Personality and Social Psychology, 6, 332-336. Zillmann, D. (1983). Disparagement humor. In P. E. McGhee & J. H. Goldstein (Eds.),

Handbook of humor research, VoL 1. Basic issues (pp. 85-107). New York: Springer-Verlag.

What's so funny about that?: The domains-interaction approach as a model of incongruity and resolution in humor

Documents