analogical reasoning, psychology of

106

Analogical Reasoning, Psychology of

Analogical Reasoning,Psychology ofDedre Gentner, Northwestern University, Evanston, Illinois, USA

CONTENTSIntroductionMapping and useFactors that influence analogical mapping and useRetrieval of analogs

Analogical reasoning is a kind of reasoning thatapplies between specific exemplars or cases, inwhich what is known about one exemplar is usedto infer new information about another exemplar .The basic intuition behind analogical reasoning isthat when there are substantial parallels acrossdifferent situations, there are likely to be furtherparallels.

INTRODUCTION

Analogical thinking is ubiquitous in human cog-nition. First, analogies are used in explainingnew concepts. Domains such as electricity or mo-lecular motion, which cannot directly be perceived,are often taught by analogy to familiar concretedomains such as water flow or billiard-ball colli-sions. Within cognitive science, mental processesare likened to computer programs (e .g. neural net-works; parallel or serial processes), or to searchingwithin a space (e .g . mental distance; close or farassociates) . Such analogies can then serve as mentalmoc 's to support reasoning in new domains . An-othe, use of analogy is in making predictionswithin domains. When the stock market plungedin 2001 after the attack on the World Trade Center,many newswriters made an analogy to the 1929Wall Street crash, and argued on this basis thatthe market would be higher after a few years(or that, because key causal conditions are differ-ent, the reverse would occur) . Analogy is also im-portant in creativity and scientific discovery, asdiscussed later . Finally, analogy is used in commu-nication and persuasion . For example, environ-mentalists have compared the earth to EasterIsland, where overpopulation and exploitation ofthe island's once-rich ecology led first to massiveloss of species, and then to famine and societalcollapse. Such persuasive analogies are meant to

Analogical learningComputational modelsThe future

invite new inferences : for example, that continuedpopulation growth will lead to irreversible eco-logical decline .

Analogical processing involves several subpro-cesses. First, given a current topic, analogical re-trieval is the process of being reminded of a pastsituation from long-term memory. Once two casesare present in working memory (either because ofan analogical retrieval or simply through encoun-tering two cases together), analogical mapping canoccur. We will begin by discussing the mappingprocesses .

MAPPING AND USE

History

Important early work on analogical mapping cameout of philosophy, notably Hesse's analysis of ana-logical models in science . Early psychological re-search on analogy focused on simple four-termanalogies of the form a :b : :c :d . In the 1970s and1980s, artificial intelligence researchers introduceda new level of representational complexity andcomputational specificity . Patrick Winston ex-plored computational algorithms for analogicalmatching and inference, and Jaime Carbonell mod-eled the transfer of solution methods from oneproblem to another. This kind of work inspiredpsychologists to lay out detailed process modelsof how analogies are represented and processed .The ensuing period has seen intense computationaland psychological research, theory revision, and anexpansion of the phenomena studied. The field ofanalogy continues to be characterized by extremelyfruitful interchange between computational modelsand psychological research . (See Analogy-making,Computational Models of)

Intermediate article


107

Analogical MappingAnalogical mapping is the core process in analogy .In a typical instance of analogical mapping, a famil-iar situation - the base or source description - ismatched with a less familiar situation - the targetdescription. The familiar situation suggests ways ofviewing the newer situation as well as further in-ferences about it . Analogical mapping requiresaligning the two situations - that is, finding thecorrespondences between the two representations- and projecting inferences from the base to thetarget. Then the reasoner must evaluate the ana-logical match and its inferences . Two further pro-cesses that can occur are re-representation of one orboth analogs to improve the match, and abstractionof the structure common to both analogs .

Structure-mapping theory (Gentner, 1983) aims tocapture the psychological processes that carry outanalogical mapping . According to this theory, thecomparison process involves finding an alignmentbetween the base and target representations thatreveals common relational structure . On the basisof this alignment, further inferences are projectedfrom base to target. People prefer to find an align-ment that is structurally consistent : that is, thereshould be a one-to-one correspondence between elem-ents in the base and elements in the target, and thearguments of corresponding predicates must alsocorrespond (parallel connectivity) . For example, inthe analogy below, Timmy in (a) could be put incorrespondence with Timmy in (b) (on the basis ofa local entity match) or with Fluffy in (b) (on thebasis of matching relational roles) . People appear toentertain both possibilities during processing, butto settle on one or the other by the end of theprocess .

(a) Lassie rescued Timmy .(b) Timmy rescued Fluffy .

Another important early theory was Holyoak's(1985) pragmatic mapping theory, which focused onthe use of analogy in problem-solving and held thatanalogical mapping processes are oriented towardsattaining goals (such as solutions to problems) .According to pragmatic mapping theory, it is goalrelevance that determines what is selected in ana-logy. Holyoak and Thagard (1989) later combinedthis pragmatic focus with structural factors in theirmulti-constraint approach to analogy .

Analogical inference projection is a crucial partof the mapping process. Once an alignment isachieved, further inferences can be made by pro-jecting information from the base (or source)domain into the target domain . For example, in

the above analogy, suppose we knew more aboutevent (a), such as :

(a) Lassie rescued Timmy because she loveshim. She has beatiful brown eyes .

(b) Timmy rescued Fluffy .In this case, the likely inference in (b) is that Timmyrescued Fluffy because he loves Fluffy . This abilityto invite new inferences is central to analogy's rolein reasoning. Importantly, analogical inference israther selective . For example, we are unlikely tomake the inference here that Timmy has browneyes (or that he has four legs, even if we alsoknow this to be true of Lassie) .

This illustrates the selection problem in theories ofanalogical inference. If people projectedd everythingknown about the base into the target, analogywould be useless in reasoning . Fortunately, peopledo not do this. Thus a central aim of theories ofanalogy is to characterize this selection process. Atleast three factors have been proposed .Holyoak and his colleagues have emphasized

goal relevance : the inferences projected are thosethat fit with the reasoner's current goals in prob-lem-solving .

A second factor, proposed in structure-mappingtheory, is relational connectivity - more specific-ally, systematicity : a preference for projecting frommatching systems of relations connected by higher-order relations such as cause, rather than projectinglocal matches. In many cases, goal relevance andsystematicity will make the same predictions, be-cause problem-solving goals often involve a focuson causal systems .

A third factor in selecting inferences, proposedby Keane, is adaptability: the ease with which apossible inference can be modified to fit the target .

There is evidence for all three of these criteria .Spellman and Holyoak (1996) showed that whentwo possible mappings are available for a givenanalogy, people will select the mapping whose in-ferences are relevant to their goals . Evidence forsystematicity comes from the finding that whenpeople read analogous passages and make infer-ences from one to the other, they are more likelyto import a fact from the base to the target when it iscausally connected to other matching predicates(Clement and Gentner, 1991; Markman, 1997) .Finally, Keane (1996) found evidence that thedegree of adaptability predicts which inferencesare made from an analogy .

There remain many open questions . For example,according to the structure-mapping account, manydifferent higher-order relations can provide infer-ential selection - including causal relations, deontic

108


relations such as permission and obligation, andspatial relations such as symmetry and transitiveincrease. The challenge then is to delineate the setof higher-order relations that can serve this pur-pose. Another open question is the time course ofthese constraints . For example, do goals havespecial priority during the analogical mapping pro-cess, or do the effects of goals occur through influ-encing the initial representations of the two analogs(before the mapping process) or through selectingamong multiple possible interpretations (after themapping process)?

EvaluationOnce the common alignment and the candidateinferences have been discovered, the analogy isevaluated . Evaluating an analogy involves at leastthree kinds of judgment : (1) structural soundness :whether the alignment and the projected inferencesare structurally consistent; (2) factual correctness :whether the projected inferences are false, true, orindeterminate in the target; and (3) relevance :whether the analogical inferences are relevant tothe current goals . In practice, the relative import-ance of these factors varies quite a bit. In domainswhere little is known, or where there is disagree-ment about the facts - for example, in politics -goal relevance may be more important than factualcorrectness .

AbstractionIn analogical abstraction, the common system thatrepresents the interpretation of an analogy is ex-tracted and stored . This kind of schema abstractionhelps to promote transfer to new exemplars . Whenpeople are asked to compare two analogous pas-sages, they are better able to later retrieve and usetheir common structure (given a relationally simi-lar probe) than are people who were given only oneof the stories (Gick and Holyoak, 1983) . Furtherstudies have shown that actively comparing twoanalogous passages leads to better subsequent re-trieval than reading the two passages separately.These findings are consistent with the claim thatanalogical alignment promotes the common struc-ture and makes it more available for later use .

Analogy in Real-world ReasoningAnalogy is often used in common-sense reasoningto provide plausible inferences . It must be notedthat analogy is not a deductive process. There is noguarantee that the inferences from a given analogy

will be true in the target, even if the analogy iscarried out perfectly and all of the relevant state-ments are true in the base . However, the set ofimplicit constraints described above make analogya relatively 'tight' form of inductive reasoning . Thismay be why analogy is heavily used in arenas suchas law, where clear reasoning is important butformal principles are often not sufficient to decideissues .

The lack of deductive certainty in analogicalreasoning has a positive side . It means that analogycan suggest genuinely new hypotheses, whosetruth could not be deduced from current know-ledge. One arena in which this kind of analogicalinferencing has been extensively studied is scien-tific reasoning and discovery . Nancy Nersessianhas examined the role of analogy and othermodel-based reasoning processes in the thoughtprocesses of Faraday and Maxwell . Paul Thagardhas discussed analogy as a contributor to concep-tual revolutions in science . Kevin Dunbar has ob-served scientists in working microbiologylaboratories and has found that analogy plays alarge role in the discovery process.

Analogy appears to be very important in chil-dren's thinking, as Halford, Goswami, and othershave argued. Children often use analogies fromknown domains as a way to fill in gaps in theirknowledge of other domains. For example, Inagakiand Hatano (1987) asked five-year-old childrenhypothetical questions like 'What would happenif a rabbit were continually given more water?'The children often answered by using an analogyto humans : for example, 'I would get sick if I keptdrinking water, and I think the rabbit would too.'Interestingly, children's answers tended to be moreaccurate when they used such analogies than whenthey did not. Children were most likely to useanalogies to humans when the target was some-what similar to humans, suggesting that for chil-dren (as for adults) similar analogs are more likelyto be retrieved and are easier to align with thetarget than dissimilar analogs .

FACTORS THAT INFLUENCEANALOGICAL MAPPING AND USEPeople's fluency in carrying out analogical map-pings is influenced by three broad kinds of factors .First are factors internal to the analogical mappingitself, such as systematicity - whether the commonrelational system possesses higher-order connect-ive structure - and transparency - the degree towhich corresponding elements are similar . Thesecond category includes characteristics of the

reasoner, such as age and expertise. The third in-cludes task factors such as processing load, timepressure, and context .

Transparency and systematicity have been foundto be important in analogical problem-solving . Thetransparency of the mapping between two analo-gous algebra problems - that is, the similarity be-tween corresponding objects - is a good predictorof people's ability to notice and apply solutionsfrom one problem to the other. For example, Ross(1989) taught people algebra problems and latergave them new problems that followed the sameprinciples . People were better able to map the solu-tion from a prior problem to a current problemwhen the corresponding objects were highly simi-lar between the two problems: for example, 'Howmany golf balls per golfer' --> 'How many tennisballs per tennis player' . They performed worst inthe cross-mapped condition, in which similar objectsappeared in different roles across the two prob-lems: for example, 'How many golf balls per golfer'

'How many tennis players per tennis ball' .The intrinsic factors of transparency and sys-

tematicity interact with characteristics of the rea-soner, notably age and experience . Gentner andToupin (1986) gave children a simple story andasked them to re-enact the story with new charac-ters. Both six- and nine-year-olds performed farbetter when the corresponding characters werehighly similar between the two stories than whenthey were different, and they performed worstwhen similar characters played different rolesacross the two stories (the cross-mapped condition) .Thus both age groups were sensitive to the trans-parency of the correspondences. In addition, olderchildren (but not younger children) benefited fromsvstematicity - that is, from hearing a summary-' atement that provided an overarching social orcausal moral.

The developmental change found here is an in-stance of the relational shift : a shift from focusing onobject matches to focusing on relational matches .Some researchers have suggested that this shift isdriven by gains in knowledge (Gentner and Ratter-mann, 1991), while others propose that it resultsfrom a developmental increase in processing cap-acity (Halford, 1993) .

The third class of factors affecting analogicalprocessing concerns task variables such as timepressure, processing load, and immediate context .One generalization that emerges from several stud-ies is that making relational matches requires moretime and processing resources than making object-attribute matches . For example, Goldstone andMedin (1994) found that when people are forced


109

to terminate processing early, they are stronglyinfluenced by local attribute matches (such as Awith A in the example below), even in caseswhere they would choose a relational match (suchasA with P) if given sufficient time :

A above M and P above A

Adult performance in mapping tasks is also in-fluenced by immediately preceding experiences.For example, in the one-shot mapping task (Markmanand Gentner, 1993) subjects are shown a pair ofcross-mapped pictures, such as a robot repairing acar and a man repairing a robot . The experimenterpoints to the robot in the first picture and the sub-ject indicates which object in the second picture'goes with' it . Subjects often choose the objectmatch (e.g. the other robot) . However, if they havepreviously rated the similarity of the pair, they arelikely to choose the relational correspondence (therepairman) . These findings suggest that carryingout a similarity comparison induces a structuralalignment .Kubose, Holyoak, and Hummel used this one-

shot mapping task to show that processing loadinfluences analogical processing . The experimenterpointed to the cross-mapped object in the first pic-ture (the robot) and subjects were instructed topoint to the relational correspondence (the repair-man) in the second picture . Subjects made moreobject-mapping errors when given an extra pro-cessing load, such as having to count backwards .Recent work by Holyoak and his colleagues alsosuggests that damage to the prefrontal cortex isassociated with detriments in analogical tasks, al-though it is not clear whether this results fromspecific involvement of the prefrontal cortex inanalogical processing or from more general factorssuch as inhibitory control .

SummaryResearch on analogical mapping has revealed a setof basic phenomena that characterize human ana-logical processing (see Table 1) . A striking featureof analogical mapping is the importance of system-atic, structurally connected representations . Com-monalities that are interconnected into a relationalsystem are considered to be more central to a com-parison than are those that are not. Connectedsystems are easier to map to a new domain thanare unconnected sets, leading to better transfer inanalogy and problem-solving . Systematicity alsogoverns inferences: inferences are projected frominterconnected systems in the base to fill outcorresponding structure in the target . Even the

110


Table 1 . Basic phenomena of analogy (adapted from Gentner and Markman, 1995, 1997 ; see also Hummel andHolyoak, 1997)

1 Relational similarity2 Structural consistency3 Systematicity4 Candidate inferences5 Alignable differences

6 Interactive mapping

7 Multiple interpretations8 Cross-mapping

differences associated with a similarity comparisonare influenced by systematicity: the differences thatare psychologically salient in a comparison arethose that are connected to the common system . Inaddition, goal relevance may have effects over andabove the effects of connected relational structure .

RETRIEVAL OF ANALOGSSo far, we have discussed the processing of ananalogy once both analogs are present . When weturn to the issue of what leads people to think ofanalogies, we see a very different pattern of results .People often fail to retrieve potentially usefulanalogs, even when there is an excellent structuralmatch, and even when they clearly have retainedthe material in memory. For example, Gick andHolyoak (1983) gave subjects a thought problem :how to cure an inoperable tumor without using astrong beam of radiation that would kill the sur-rounding flesh . Only about 10 percent of the par-ticipants came up with the ideal solution, which isto converge on the tumor with several weak beamsof radiation. If given a prior analogous story inwhich soldiers converged on a fort, three times asmany people (about 30 percent) produced conver-gence solutions . Yet the majority of participantsstill failed to think of the convergence solution .Surprisingly, when these people were simply toldto think back to the story they had heard, the per-centage of convergence solutions again tripled, to80-90 percent. Thus, the fortress story had beenretained in memory, but it was not retrieved bythe analogous tumor problem . The implicationis clear . Even when a prior experience has beensuccessfully stored in memory, it might not be re-trieved when a person encounters a new analogoussituation where it would be useful .

When we ask what does facilitate analogical re-trieval, one major factor emerges : the similaritybetween the analogs . As noted earlier, similarity is

Analogies involve relational commonalities ; object commonalities are optional .Analogical mapping involves one-to-one correspondence and parallel connectivity .In interpreting analogy, connected systems of relations are preferred over isolated relations .Analogical inferences are generated via structural completion .Differences that are connected to the common system are rendered more salient by acomparison .

Analogy interpretation depends on both terms . The same term yields different interpretationsin different pairings .

Analogy allows multiple interpretations of a single comparison .Analogies are more difficult to process when there are competing object matches .

one of the factors that facilitates analogical map-ping; but it has a much larger effect on retrieval . Forexample, Gentner et al . (1993) gave subjects a set ofstories to remember and later showed them probestories that were either surface-similar to theirmemory item (e.g. similar objects and characters)or structurally similar (i .e . analogous, with similarhigher-order causal structure). Surface similaritywas the best predictor of whether people wouldbe reminded of the prior stories; people were twoto five times more likely to retrieve prior storieswith only surface commonalities than with onlystructural commonalities . However, their judg-ments of the goodness of the match were com-pletely different. They rated the surface-similarpairs (including their own remindings) as low ininferential value and in similarity, and preferredthe structurally similar pairs . A similar dissociationbetween reminding and use has been found inproblem-solving tasks: remindings of prior prob-lems are strongly influenced by surface similarity,even though structural similarity better predictssuccess in solving current problems (Ross, 1989) .This failure to access potentially useful analogs(unless they are highly similar to the target) is aninstance of the inert knowledge problem in education.One piece of good news is that it appears thatdomain expertise may improve matters somewhat .For example, Novick (1988) found that people withmathematics training retrieved fewer surface-simi-lar lures in a problem-solving task than did novicemathematicians. Moreover, experts were quicker toreject these false matches than were novices .

One factor that may contribute to experts' successin analogical retrieval is representational uniformity :the extent to which the relations in the memorytrace are represented similarly to those in theprobe. Clement et al . (1994) explored the effect ofrelational predicate similarity on analogical accessand mapping between stories . They found that re-trieval was more likely when the probe and target

contained synonymous terms (manifest similarity)than when they contained loosely similar predi-cates such as 'munched' and 'consumed' (latentsimilarity) . However, unlike retrieval, analogicalmapping when both situations were present wasrelatively unaffected by the latent-manifest distinc-tion. In analogical reminding, with only the currentsituation in working memory, success depends onthe degree of match of the pre-existing representa-tions; whereas during mapping, with both situ-ations present in working memory, there isopportunity for re-representation .

ANALOGICAL LEARNINGAnalogical comparison can lead to new learning inat least four ways : analogical abstraction, inferenceprojection, difference detection, and re-representa-tion. The first two we have already discussed . Inanalogical abstraction, the structure common to baseand target is noticed and extracted . Sometimes thecommon system is stored as a separate representa-tion; this is referred to as schema abstraction (Gick andHolyoak, 1983) . In inference projection, a propositionfrom the base is mapped to the target . If it is retainedas part of the target structure, then learning hasoccurred . In difference detection, carrying out a com-parison process leads people to notice certain differ-ences - namely, those connected to the commonstructure. In re-representation, two non-identicalpredicates are aligned and decomposed (or ab-stracted) to find their commonalities, resulting in are-representation that contains a common predi-cate: for example, comparing chase(dog, cat) andfollow(detective, suspect) might result in pursue(entityl, entity2) . A further kind of knowledgechange, hypothesized to take place in scientific dis-covery, is restructuring, in which the target under-goes a radical change in structure .

COMPUTATIONAL MODELSThe interplay between computational models andpsychological studies has been extremely product-ive in analogical research . Current models includeBoicho Kokinov's AMBR model, which integratesretrieval and mapping; Keane's JAM model, whichutilizes an incremental mapping algorithm; Half-ord's tensor product model; and the systems ofDoug Hofstadter and his colleagues Melanie Mitch-ell and Robert French, which integrate perceptualprocessing with analogical matching . (See Ana-logy-making, Computational Models of)

Analogical modeling has made great strides, butthere are still open questions . At present no model


111

fully captures human analogy processing . Twochallenges for analogical models are (1) the selectionproblem discussed above - namely, how to avoidindiscriminate inferencing ; and (2) the problem ofrepresentational flexibility - that is, how to achieve amatching process that does not require absoluteidentity matches . Falkenhainer et al .'s (1989) SME,which uses a local-to-global alignment and infer-ence process over structured symbolic representa-tions, meets the benchmarks in Table 1 and cancapture selective matching and inference, as wellas schema abstraction . But it is not yet sufficientlyflexible in its match process . Another leadingmodel, Hummel and Holyoak's (1997) LISAmodel, uses a combination of distributed represen-tations of concepts and structured representationsof the relational connections (necessary for achiev-ing structural consistency in mapping) . It uses aconnectionist temporal-binding algorithm andmakes its matches in a serial order partly guidedby the experimenter. LISA's use of distributedrepresentations allows for flexible matching, andunlike most models of analogy, it attempts to cap-ture working memory limitations . However, it hasyet to solve the selectivity problem in inferencing .

THE FUTUREOf the many research questions that remain, fourstand out as particularly interesting and timely .First is the role of analogy in cognitive develop-ment: how much of children's rapid learning canbe attributed to the processes of comparing anddrawing inferences between partially similar situ-ations? Second is tracing the neuropsychology ofanalogical processes: what areas of the brain areimplicated, and what is the course of processing?Third is the exploration of analogy in animal cogni-tion. Comparative research so far indicates thathumans excel in analogical ability, yet this abilityexists in certain other species as well - for example,in chimpanzees and dolphins . Cross-species com-parisons may help us decompose the cognitivecomponents of analogical ability . A final importantresearch frontier is the integration of analogy intolarger models of cognition .

References

Clement CA and Gentner D (1991) Systematicity as aselection constraint in analogical mapping . CognitiveScience 15 : 89-132 .

Clement CA, Mawby R and Giles DE (1994) The effects ofmanifest relational similarity on analog retrieval .Journal of Memory and Language 33 : 396-420.

112


Falkenhainer B, Forbus KD and Gentner D (1989) Thestructure-mapping engine: algorithm and examples .Artificial Intelligence 41 : 1-63 .

Gentner D (1983) Structure-mapping : a theoreticalframework for analogy . Cognitive Science 7(2) : 155-170 .

Gentner D and Markman AB (1995) Analogy-basedreasoning in connectionism . In : Arbib MA (ed .) TheHandbook of Brain Theory and Neural Networks, pp. 91-93 .Cambridge, MA: NUT Press .

Gentner D and Markman AB (1997) Structure-mappingin analogy and similarity . American Psychologist 52:45-56 .

Gentner D and Rattermann MJ (1991) Language and thecareer of similarity. In: Gelman SA and Byrnes JP (eds)Perspectives on Thought and Language : Interrelations inDevelopment, pp. 225-277 . London, UK: CambridgeUniversity Press .

Gentner D, Rattermann MJ and Forbus KD (1993) Theroles of similarity in transfer : separating retrievabilityfrom inferential soundness . Cognitive Psychology 25 :524-575 .

Gentner D and Toupin C (1986) Systematicity and surfacesimilarity in the development of analogy . CognitiveScience 10: 277-300 .

Gick ML and Holyoak KJ (1983) Schema induction andanalogical transfer . Cognitive Psychology 15(1) : 1-38 .

Goldstone RL and Medin DL (1994) Time course ofcomparison . Journal of Experimental Psychology: Learning,Memory and Cognition 20(1) : 29-50 .

Halford GS (1993) Children's Understanding : TheDevelopment of Mental Models . Hillsdale, NJ: LawrenceErlbaum .

Holyoak KJ (1985) The pragmatics of analogical transfer .In: Bower GH (ed .) The Psychology of Learning andMotivation : Advances in Research and Theory, vol . 19,pp. 59-87. New York, NY: Academic Press .

Holyoak KJ and Thagard PR (1989) Analogical mappingby constraint satisfaction . Cognitive Science 13(3) :295-355 .

Hummel JE and Holyoak KJ (1997) Distributedrep : ;entations of structure: a theory of analogicalaccess and mapping . Psychological Review 104(3) :427-466 .

Inagaki K and Hatano G (1987) Young children'sspontaneous personification as analogy . ChildDevelopment 58 : 1013-1020.

Keane MT (1996) On adaptation in analogy : tests ofpragmatic importance and adaptability in analogicalproblem solving. Quarterly Journal of ExperimentalPsychology 49: 1062-1085 .

Markman AB (1997) Constraints on analogical inference .Cognitive Science 21(4) : 373-418 .

Markman AB and Gentner D (1993) Structural alignmentduring similarity comparisons . Cognitive Psychology 25 :431-467 .

Novick LR (1988) Analogical transfer, problem similarity,and expertise . Journal of Experimental Psychology:Learning, Memory and Cognition 14 : 510-520 .

Ross BH (1989) Distinguishing types of superficialsimilarities: different effects on the access and use ofearlier problems . Journal of Experimental Psychology:Learning, Memory and Cognition 15 : 456-468 .

Spellman BA and Holyoak KJ (1996) Pragmatics inanalogical mapping . Cognitive Psychology 31 : 307-346 .

Further Reading

Carbonell JG (1983) Learning by analogy : formulatingand generalizing plans from past experience. In:Michalski RS, Carbonell JG and Mitchell TM (eds)Machine Learning: An Artificial Intelligence Approach,vol. 1, pp. 137-161 . Palo Alto, CA: Tioga .

Dunbar K (1995) How scientists really reason : scientificreasoning in real-world laboratories . In : Sternberg RJand Davidson JE (eds) The Nature of Insight, pp . 365-395 .Cambridge, MA : MIT Press .

French R (1995) The Subtlety of Sameness : A Theory andComputer Model of Analogy-making . Cambridge, MA :MIT Press .

Gentner D, Holyoak KJ and Kokinov BN (eds) (2001) TheAnalogical Mind: Perspectives from Cognitive Science .Cambridge, MA : MIT Press .

Goswami U (1992) Analogical Reasoning in Children .Hillsdale, NJ: Lawrence Erlbaum .

Halford GS, Wilson WH and Phillips S (1998)Processing capacity defined by relational complexity :implications for comparative, developmental, andcognitive psychology . Behavioral and Brain Sciences21:803-864 .

Hesse MB (1966) Models and Analogies in Science . NotreDame, IN: University of Notre Dame Press .

Hofstadter DR and Mitchell M (1994) The Copycatproject: a model of mental fluidity and analogy-making .In: Holyoak KJ and Barnden JA (eds) Advances inConnectionist and Neural Computation Theory, vol . 2 :Analogical Connections, pp . 31-112. Norwood, NJ:Ablex.

Holyoak KJ and Thagard PR (1995) Mental Leaps : Analogyin Creative Thought . Cambridge, MA : MIT Press .

Keane MT (1990) Incremental analogising : theory andmodel . In : Gilhooly KJ, Keane MTG, Logie RH andErdos G (eds) Lines of Thinking, vol . 1 . Chichester, UK :John Wiley.

Kokinov BN and Petrov AA (2001) Integrating memoryand reasoning in analogy-making : the AMBR model . In:Gentner D, Holyoak KJ and Kokinov BN (eds) TheAnalogical Mind: Perspectives from Cognitive Science,pp. 161-196 . Cambridge, MA : MIT Press.

Nersessian NJ (1984) Faraday to Einstein : ConstructingMeaning in Scientific Theories . Dordrecht, Netherlands :Nijhoff.

Thagard P (1992) Conceptual Revolutions . Princeton, NJ:Princeton University Press .

Winston PH (1982) Learning new principles fromprecedents and exercises . Artificial Intelligence 19 :321-350 .

analogical reasoning, psychology of

Documents