Similarity and Emergence in Conceptual Combination

Journal of Memory and Language 45, 21–38 (2001) doi:10.1006/jmla.2000.2772, available online at http://www.academicpress.com on Similarity and Emergence in Conc Merryl J. Wilkenfeld and Tho ity co pan si fea th par co for w few s a em ir cep n i cep of of ith cep o fac je str he str ro to yo cep pa sec e go A st is our capacity to produce and comprehend con- ceptual co arate con thoughts a binations ously unl mom,” an pand the l creative p ley, Doar Baughma Rothenbe A parti combinati properties as being c of either o 1987, 199 Kunda, M s understood to be personal electronic entries on Address c B. Ward, De College Stat mbinations that merge previously sep- cepts into units that express new nd stimulate new ideas. Because com- can describe things that were previ- abeled, such as “homepage,” “soccer d “couch potato,” they can serve to ex- anguage and have even been linked to roductivity (Donaldson, 1991; Mob- es, & Mumford, 1992; Mumford, n, Maher, Costanza, & Supinski, 1997; rg, 1979). cularly intriguing aspect of conceptual ons is that they can yield emergent , that is, properties that people identify haracteristic of a combination but not f its constituents (see, e.g., Hampton, the World Wide Web, properties that people would be unlikely to list as being characteristic of either “homes” or “pages” considered sepa- rately. Such emergent properties can be truly novel in the sense that they appear in the combi- nation without being present at all in the repre- sentation of either parent concept, or they can merely be newly salient in the sense that they are in the representation of one or both parent con- cepts but low enough in importance that people would not think to list them when considering the parents in isolation. Either type of emer- gence, however, reflects a change in the way the concepts are interpreted; even in the newly salient case, an attribute that was unimportant enough to come to mind for either constituent in Texas A&M Univers The influence of similarity on emergence in interpretations of ts wrote two definitions for each of eight similar and eight dis tures of each definition. Those features were compared with ticipants who listed the characteristic properties of the parent the combinations but not for the parent concepts separately er emergent features than dissimilar pairs, and first attempt ergent features than second attempts. Definitions of similar pa t to the other, whereas definitions of dissimilar pairs more ofte ts. Similar pairs and first attempts also had higher proportions the combinations’ parent concepts. The results are consistent w tual combination and with a “quick fix” hypothesis. Members ilitated the identification of a property of the modifier to pro uctures required little emergent modification to incorporate t uctures were less readily alignable, provided fewer projectible p incorporate them, and more often required participants to go be ts to achieve coherent interpretations. On second definitions, ondary alignments that led to less property projection more em Key Words: conceptual combination; emergent properties; cate riking characteristic of human cognition For in 21 7; Hastie, Schroeder, & Weber, 1990; iller, & Claire, 1990; Murphy, 1988). isolatio tion. D change of how A se concep orrespondence and reprint requests to Thomas partment of Psychology, Texas A&M University, ion, TX 77843. E-mail: tbw@psyc.tamu.edu. eptual Combination mas B. Ward nceptual combinations was assessed. Partici- milar word pairs and then listed the important e features collected from a different group of ncepts presented individually. Features listed ere considered emergent. Similar pairs led to t defining the combinations produced fewer s more often assigned a property of one con- dentified a thematic relation between the con- features from within the structural alignments the influence of structural alignment on con- f similar pairs, which could easily be aligned, ct onto the head noun, and their compatible projected property. Dissimilar pairs, whose perties, required more emergent modification nd the alignable properties of the parent con- rticipants may have pursued less satisfactory rgent features. © 2001 Academic Press ries; concepts. tance, “homepages” are now commonly 0749-596X/01 $35.00 Copyright © 2001 by Academic Press All rights of reproduction in any form reserved. n becomes prominent in the combina- etermining the factors that underlie such s is essential to a complete understanding people interpret combined concepts. cond and potentially related aspect of tual combination is that the same com- 22 WILKENFELD AND WARD bination can yield multiple forms of interpre- tation, constit in som tion, in concep which blend niewsk such a tional bras,” horse,” mal th horse.” The these more s ity bet tion ca tation a It also ity-bas gence one an Our from a finding the exp related stituen many combin that co spiring tion ar berg, 1 diverge emerge Con is sugg betwee from s combin being stituen (1990) person junctio is Harvard-educated and a carpenter) or a per- h t r h n f t r e n o t g A e r t e l s r r e r h t e t w p h io n including relation linking, in which the uent concepts play complementary roles e thematic relation; property interpreta- which a property is projected from one t to the other; and hybridization, in the combination is seen as a cross or between the constituents (see, e.g., Wis- i, 1997b). For instance, a combination s “zebra horse” could yield the rela- interpretation of “a horse for herding ze- the property interpretation of “a striped or the hybrid interpretation of “an ani- at is a cross between a zebra and a present research focused on both of aspects of conceptual combination and pecifically on how the degree of similar- ween the parent concepts of a combina- n influence both the form of the interpre- nd the amount of emergence that occurs. focused on the extent to which similar- ed effects on interpretations and emer- are either interrelated or distinct from other. focus on similarity stemmed initially necdotal reports as well as experimental s and theoretical proposals that lead to ectation that emergence will be inversely to the similarity of a combination’s con- ts. Anecdotally, creative people from different disciplines report that they use ations as a source of creative ideas and nceptual combination is particularly in- when the components of the combina- e dissimilar (Donaldson, 1991; Rothen- 979). The implication is that the more nt the component concepts are, the more ntly creative the outcome will be. sistent with these anecdotal reports, there estive experimental evidence for a link n similarity and emergence, particularly tudies using intersective or conjunctive ations, in which an entity is specified as simultaneously a member of two con- t categories. For instance, Kunda et al. had participants describe either a target who is a member of a “surprising” con- n of social categories (e.g., a person who son w consti Harva ter). T ties in contai ther o educa as non person simila more for co cial w Hamp gence lappin fruit). set ov of the it is re more stituen are su consti at leas tion b Bui studie simila sessed simila yield the ea tions t other catego as “h likely are “h could other are es stand ilarity pretat combi tent to o is a member of one or the other of the uent categories (e.g., a person who is d-educated or a person who is a carpen- ey found evidence for emergent proper- that the descriptions of the conjunctions ed properties that were not listed for ei- the constituent categories (e.g., Harvard- ed carpenters were sometimes described materialistic, whereas Harvard educated s or carpenters alone were not). Using procedures, Hastie et al. (1990) found mergent properties for incongruent than gruent conjunctions (e.g., Republican so- rker versus Republican bank teller), and on (1997) reported a great deal of emer- for imaginary conjunctions with no over- members (e.g., furniture which is also lthough surprisingness, congruence, and rlap are not necessarily direct indicators similarity of a conjunction’s constituents, asonable to suppose that less surprising, congruent, and more overlapping con- ts are more similar to one another than prising, incongruent and nonoverlapping uents. Consequently, the results provide t suggestive support for an inverse rela- tween similarity and emergence. ding on these earlier findings, the present used a more direct measure of concept ity (i.e., participants’ ratings) and as- the degree to which interpretations of and dissimilar noun–noun combinations mergent properties. They also extended lier findings to nonpredicating combina- at can take many forms of interpretation han joint membership in their constituent ries. That is, we used combinations, such licopter blanket,” which would be un- o be interpreted as the class of things that elicopters and also blankets” but which reasonably be interpreted in a variety of ays. Because these types combinations ecially common, it is essential to under- ow they are interpreted and how the sim- of their constituents affects their inter- n. In addition, by examining such ations it is possible to determine the ex- which the form of interpretation (e.g., re- SIMILARITY AND EMERGENCE 23 lation linking versus property interpretation) and th The binatio spectiv count Kunda ness o gage i person social gators menta reason from c otherw 1997; The are pa gories incong to reso patible we sus genera tion. C noun– need n contra betwee fluenc served It is conflic play w reason compl For ex provok means junctio combi fied as there m develo link be minim to inte proper Although a number of models can explain e t H a e l t w a u u e o i o e n p s n ti io c h r , ly o F p n s e amount of emergence are interrelated. present work also stemmed from a com- n (no pun intended) of theoretical per- es on conceptual combination. To ac- for the presence of emergent properties, et al. (1990) proposed that the surprising- f a social conjunction prods people to en- n causal reasoning to explain how a target might be a member of two incongruent categories. In a related vein, other investi- have suggested that people might develop l simulations or engage in other forms of ing to resolve apparent conflicts that arise onsidering simultaneous membership in ise discrepant categories (e.g., Hampton, Hastie et al., 1990; Thagard, 1984). se various types of reasoning mechanisms rticularly applicable to conjunctive cate- for which simultaneous membership in ruent categories would force people to try lve conflicts between potentially incom- properties of those categories. However, pect that the mechanisms also apply more lly to other forms of conceptual combina- onsequently, we anticipate that, even for noun combinations in which the entity ot be simultaneously a member of two sting categories, the degree of similarity n the constituent categories will still in- e the number of emergent properties ob- . important to note, however, that although t resolution processes may come into hen people interpret combinations, those ing processes alone cannot provide a ete account of the interpretation process. ample, although surprisingness clearly es causal reasoning, there must be some by which the surprisingness of a con- n is initially determined. In addition, for nations for which the entity is not speci- a member of both constituent categories, ust be some process by which a person ps a reasonable idea of the nature of the tween the constituents to begin with. At a um, there must be processes that can lead rpretations in the form of relation linking, ty projection, and hybridization. how p terpre 1988; vide a (e.g., full r Wisni mode tempt tion, a larity To tions, constr struct Gentn the c and d the m noun, select pariso ation but co head n are as ison a nating tions poten pretat which given by the The fact t are sim interp to mo 1997a many (i.e., d under from ward. conce tify a string ople might arrive at relation linking in- ations (Cohen & Murphy, 1984; Murphy, Shoben & Gagne, 1997), and others pro- n account of conjunctive interpretations ampton, 1987), most do not predict the nge of interpretation types. In contrast, wski’s (1997a, 1997b) dual-process posits two distinct mechanisms in an at- o account for all three types of interpreta- nd it also has implications for how simi- ill influence interpretations. ccount for property and hybrid interpreta- Wisniewski proposed a comparison and ction process that uses the principles of ral alignment (see, e.g., Markman & r, 1993) to align the representations of nstituents, identify their commonalities fferences, determine which properties of difier should be projected onto the head and instantiate some new version of the d properties in the head noun. The com- process competes with a scenario cre- rocess, which seeks to identify different mplementary roles for the modifier and oun in a thematic relation. The processes umed to run in parallel, with the compar- d construction process potentially culmi- in property interpretations or hybridiza- and the scenario creation process ally culminating in relation linking inter- ns. Which process “wins,” and hence, type of interpretation is expressed for a ombination is assumed to be influenced similarity of the component concepts. dual-process model correctly predicts the at combinations of parent concepts that ilar to one another lead to more property etations, whereas dissimilar pairings lead re relation linking (Wisniewski, 1996, 1997b). Because similar concepts share commonalities and alignable differences ifferences that exist along some common ing dimension), projecting properties ne concept onto the other is straightfor- or example, a comparison of the similar ts “guitar” and “harp” could readily iden- alignable difference in the number of and lead to the property interpretation, “a 24 WILKENFELD AND WARD six-stringed harp.” In addition, similar entities would roles, a relat differe ity of t “couch proper jected tation. ever, roles f tion li object does n horses larity p large s alignab crease for sim tions f Give fluence form o fully a interre though likelih giving direct gence Relatio erties a pretati may or To s of inte siderin similar is imp ings o shorth erties o that p viewed knowle rather exact c its representation (see, e.g., Wisniewski, 1996, . m a i e c d . it io o c , f t a n n n p p o i ly a , e a l io re m W x b a t d s o o y tend to have the capacity to fill similar which could make it difficult to construct ional scenario in which they would play nt roles. In contrast, the limited alignabil- he attributes of dissimilar concepts (e.g., ” and “skate”) makes it difficult to find ties of one that could reasonably be pro- onto the other to form a property interpre- A scenario construction process, how- might readily identify complementary or the differing objects, leading to a rela- nking interpretation (e.g., “a runnerlike for moving a couch”). High similarity ot rule out relation linking (e.g., “zebra ” that “herd zebras”), nor does low simi- reclude property interpretations (e.g., “a kate that seats three”), but variations in ility and role-filling capacity would in- the likelihood of property interpretations ilar pairs and relation linking interpreta- or dissimilar pairs. n the possibility that similarity might in- both the extent of emergence and the f the interpretation, it is essential to care- ssess the degree to which those effects are lated or separate. Our view is that, even low similarity may increase both the ood of emergence and the probability of a relation linking definition, there is no connection between the amount of emer- and the form of the given interpretation. n linking does not foster emergent prop- ny or more or less so than property inter- on. Rather, either type of interpretation may not contain emergent properties. ee how emergence can occur for all types rpretations, and to set the stage for con- g the complexity of the linkages between ity, interpretation type, and emergence, it ortant to note that people’s understand- f combined concepts are richer than the and labels that refer to the projected prop- r relations. This is due, in part, to the fact rojecting a property or relation is best as a constructive process of integrating dge into the head noun of a combination than as simply a matter of placing an opy of a known property or relation into 2000) tation or rel receiv emerg a con of one (as in imply (see, e a “gu pretat from conne shape played differ make a harp ent m copter “a bla certai throw helico of a from ing s quent resent a role of em Bec how a pretat comp less e pairs. the e would aligna associ noun, aligne That i either from factor The process of constructing the interpre- ay entail modifications to the property tion being projected or to the head noun ng it, either of which could result in nce. In addition, properties contained in ept’s representation are not independent another, so that a change in one property icated by a property interpretation) might potentially emergent changes in others g., Medin & Shoben, 1988). For instance, ar harp,” defined via the property inter- n of “a six-stringed harp,” may differ rdinary harps in several ways that are ted to its having six strings (e.g., size, manner of play, and type of music ), and the strings may well be expected to rom ordinary guitar strings in ways that hem more appropriate for the structure of (e.g., larger, thicker, and made of a differ- terial). Similarly, for a pair such as “heli- blanket,” the relation linking definition of ket to cover a helicopter” would almost ly not simply refer to an ordinary blanket over an ordinary helicopter. Rather, for a ter blanket to adequately serve the role rotective cover, it would have to differ rdinary blankets in several ways, includ- ze and material composition. Conse- a person might need to modify the rep- tion of “blanket” to allow it to play such and the modifications could take the form rgent properties. use judged similarity is an indicator of ignable concepts are, to the extent inter- ns are driven by an alignment process, hending similar pairs should result in ergence than comprehending dissimilar hen interpreting a similar combination, istence of readily alignable structures not only facilitate the identification of an le difference and the projection of the ted value from the modifier to the head he absence of major conflicts across the structures would allow a “quick fix.” , there would be little pressure to modify f the constituents or to import properties utside the alignment to achieve a satis- definition. For example, a comparison SIMILARITY AND EMERGENCE 25 process would not identify many strongly in- congru and ho vided withou combi their r stripes soning conflic erties tween a relat ity wo would occurr In c structu quick one co define more charac cepts concep alignm tion fo seats t that pr crepan nent proper tion to case o (1998) the de plies t cation tion o drastic such a modifi proper a three belts” Bec to res cause ally th of their properties, the predictions regarding a s t b t n , p ent or conflicting properties of zebras rses, so that “a striped horse” could pro- a satisfactory definition of “zebra horse” t requiring the person interpreting the nation to make major modifications to epresentations of “horse” or the zebra’s . There would be little need for the rea- processes described earlier to resolve ts and, consequently, few emergent prop- would be expected. As with the link be- similarity and interpretation type, this is ive, not absolute argument. High similar- uld not preclude emergent properties, but be exp