Is Snow Really Like a Shovel? Distinguishing Similarity from Thematic Relatedness

Traditionally, thematic relatedness (chicken and egg) and similarity (chicken and turkey) have been thought of as distinct phenomena, the former the result of associative processes, and the latter reflecting comparison processes . However, recent studies (Bassok & Medin, 1996 ; Wisniewski & Bassok, 1996) suggest that similarity is a result of both association and comparison . This could call for a radical redefinition of similarity as inherently fused with association . We term this view the integration account . We consider an alternative, the confusability account, under which thematic influences intrude upon assessments of similarity but are not an essential part of the similarity process . We present two experiments supporting the confusability account . The first indicates that comparison and association are independent processes . The second shows that thematic influences rise with increased cognitive load. We believe that while a redefinition of similarity is not warranted, similarity is more vulnerable to error and intrusion than is generally thought . relationships is clear when one examines their predicateargument structures . For example, according to Gentner (1988), a shovel and a spade are similar because they share relations (e.g., MOVE [shovel, here, there, stuff]/MOVE [spade, here, there, stuff]) and attributes (e .g., HANDLE [shovel ]/HANDLE [spade]) . In contrast, thematic associates like snow and shovel generally do not share attributes, and take different roles in relational predicates (e .g., MOVE [shovel, here, there, snow]) . Because snow and shovel do not share features of appearance or use, both Gentner's and Tversky's models predict low similarity . Snow and shovel are related, though ; experience informs people that snow and shovels interact, appear in the same propositions, and commonly co-occur . But this hinges upon noting relations that associate snow and shovel, rather than on comparing the two and noting common structure . Perhaps because the distinction between thematic association and similarity seems obvious in many contexts its validity has not been the focus of much research. It seems clear that to base categorizations, word extensions, and inferences on thematic associations would frequently cause error . For example, whales are thematically related to the plankton they eat and to the harpoons with which vhalers hunt them. However, to conclude that plankton or harpoons are warm-blooded like whales would be incorrect . Likewise, thematic associations play roles that similarity cannot. For example, associations allow us to detect and make predictions on the basis of covariation (Kelley, 1973 ; Novick & Cheng, 1990). The association between smoke and fire leads us to sound the alarm when we see dark clouds billowing from an apartment window . But, looking for smoke when we see fog because they are perceptually similar would be a mistake . Differences such as these have provided a strong argument for treating similarity and thematic association as separate processes . Beyond these intuitions, much of cognitive research depends on the theoretical assumption that thematic association and similarity relationships are distinct . Therefore, the finding that the two are not psychologically separable would radically alter our understanding of cognition. Yet, some recent theories suggest a strong link between similarity and association . Sloman (1996) proposes that that similarity and association are processed by a single system, with abstract rules reposing in a separate system . Further, Bassok and Medin (1996) have recently found that when judging the similarity of two sentences, participants are influenced by thematic relationships between the sentences . For example. participants judged the sentences The carpenter fixed the chair and The carpenter sat on the chair to be similar because "the carpenter sat on the chair to see if the repair would hold ." Likewise, Wisniewski and Bassok (1996) compared the similarity ratings assigned to similar pairs (milk-lemonade), thematic pairs (milk-cow) and pairs sharing both similarity and thematic relationships (milk-coffee) . Thematic relationships significantly increased similarity ratings : e .g ., milk and coffee were rated as more similar than milk and lemonade . Although previous research has shown that similarity and thematic relationships are confusable by children (e .g ., Bauer & Mandler, 1989), these experiments go further in suggesting that these confusions occur for adults as well . . These results force us to consider the possibility that traditional models oversimplify the concept of similarity . Under Medin and Bassok's (1996) view, similarity is an integration of independent processes of comparison and association . However, an even more radical position can be taken, consistent with Sloman's position . Under this view, which we term the integration account, thematic associations and comparisons are the result of a unitary process . Similarity is thus inherently intermixed with thematic relatedness . Another possibility is that the comparison process is distinct from association, and thematic influences arise as the result of thematic intrusions that interfere with similarity judgments . According to this confusability account, similarity is the result of the comparison process, but this process can be derailed by other factors . Such derailment might occur if participants have difficulty distinguishing between the mental' output that arises from accessing associations as opposed to the output of a separate, independent comparison process . Several lines of reasoning lead us to entertain the confusability account. First, Markman (1989) argues that the developmental shift towards a preference for taxonomic groupings does not indicate a loss of salience for thematic relationships, but a rise in the salience of taxonomic relationships . Likewise, Smiley and Brown (1979) argue that the shift is one of preference, rather than a radical 180 restructuring of knowledge . If adults remain sensitive to thematic associations, thematic relationships could interfere with other processes . A second line of reasoning concerns individual differences . We have found evidence of substantial variation in people's ability to distinguish and identify similarity relations . In a, screening task (further described below), participants were given a standard (e .g ., dart) and had to choose which alternative was most similar : bullseye or javelin . Eleven percent of the participants were consistently unable to distinguish similarity from thematic association (Undifferentlating) . Another 41 11c showed varying degrees of differentiation . If such confusion were universal, it would suggest a basic fusion of the processes of comparison and association, supporting the integration account. However, 48% of participants show no confusion of similarity and thematic association (Differentiating) . Our proposal is that similarity and thematic relatedness result from two separate processes . However, the output of these processes is sometimes difficult to differentiate through direct introspection, particularly when similarity is very low (the conditions in which thematic intrusions have been found in similarity tasks) . We conjecture that the ability to make the distinction reliably and explicitly is learned as part of the development of metacoznitive skills . There is precedent for the view that internal cognitive distinctions must be learned . Markman (1979) found that 811 year-old children failed to report any comprehension difficulties when reading stories that contained blatant inconsistencies (e .g ., that ants rely on smell, but have no noses and cannot smell) . Nevertheless, the children were slower to read sentences that led to strong inconsistencies . This suggests that they engaged in sonic kind of inconsistency processing but could not label it as such . Elsewhere, we see evidence for people's inability to accurately distinguish between the cognitive and emotional states produced by real and implanted memories (e .g ., Loftus, 1997), and an inability to correctly identify what they know or do not know (e.g., Koriat, 1993). Additionally, there is evidence of developmental shifts in the ability to reflect upon the products of cognition (e .g ., Flavell, Green & Flavell, 1990 ; Kitchener & King . 1994). Going further, there is evidence for historical shifts in the use of comparison and association . Medieval alchemy treated similarity and thematic relatedness as interchangeable to a greater degree than in modern science (Gentner & Jeziorski, 1993) . In short, we suggest that reliably distinguishing the sensation of relatedness via commonalities from the sensation of relatedness via thematic associations may require a degree of metacognitive sophistication . On the confusability account, the findings of Bassok & Medin (1996) and Wisniewski & Bassok (1996) result from people's failure to explicitly distinguish the results of an associative retrieval process and the results of a comparison process . But if these processes are indeed separable, it should be possible to find a task that draws only on comparison . In Experiment 1, we do this by utilizing a word learning task,