M ill!and ComonStns W) 8OfInjaimg C Carbonefl and Sloven MitonFSMarchi 19WSDEPARTMENTCOMPUTER SCIENCE DepEETAULEte* Calrneie MellonU.83 08160*CNU-CS-8 3- 11.0Metaphorand Common-Sense ReasoningJaime G Carbonell and Steven MintonCarnegie-Mellon UniversityPittsburgh, PA 152135SMarch 1983Abstract>Inferenccs based on metaphors appear to play a major role in human common sense reasoning. Thispaper identifies and analyzes general inference patterns based upon ur Jerlying metaphors, inparticular the pervasive balance principle. Strategies for metapnor comprehension are explored, andanalogical mapping structures are proposed as a means of representing metaphorical relationshipsbetween domains. In addition, a framewo~rk for a computational model embodying principles ofmetaphorical common sense reasoning is dsusd*k1This research was spor-ored in part by the Office of Nava! Research (ONP) under grant numbers N00014-79-C-O5u1 andN00014-82-C-50767.Metaphor and Common Sense ReasoningTable of Contents1. Introduction I2. Experiential Reasoning vs Formal Systems 13. Patterns of Metaphorical Inference 33.1. The Balance Principle 33.2. The Physical Metaphor Hypothesis 44. The Role of Metaphor in Common-Sense Reasoning 55. Metaphorical Inference and The Mapping Problem a5.1. Knowledge Acquisition via Analogical Mappings 75.2. Salience and Novel Metaphors 85.3. A Classification Based on Processing Requirements 106. Representing Metaphors: The LIKE Relation 117. Generalizing Mapping Structures 158. Towards a Computational Model cf Metaphorical Inference 179. Conclusions 1910. Bibliography 21M AIIMetaphor and Common Sense ReasoningMetaphor and Common Sense ReasoningJaime G. Carbonell and Steven Minton1. IntroductionThe theory that metaphor dominates large aspects of human thinking, as well playing a significantrole in linguistic commur.cation, has been argued with considerable force [26, 24, 8, 5]. However, thevalidity of such a theory is a matter of continuing debate that appears neither to dissuade itsproponents nor convince its detractors. Being among the proponents, we propose to develop acomputational reasoning system for performing metaphorical inferences. If such a system exhibitscognitively plausible common sense reasoning capabilities, it will demonstrate, at the very least, theutility of metaphorical inference in noodeling significant aspects of naive human reasoning. Thispaper reviews our initial steps towards the development of a computational model of metaphor-basedreasoning.2. Experiential Reasoning vs Formal SystemsHumans reason and learn from experience to a degree that no formal system, Al model, orphilosophical theory has yet been able to explain. The statement that the human mind is (or contains)the sum total of its experiences is in itself rather vacuous. A more precise formulation of experience-bas.d reasoning must be structured in terms of coordinated answers to the following questions: Howare experiences brought to bear in understanding new situations? How is long tern memory modifiedand indexed? How are inference patterns acquired in a particular domain and adapted to apply innovel situations? How does a person "see the light" when a previously incomprehensible problem isviewed from a new perspective? How are the vast majority of irrelevant or inappropriate experiencesand inference patterns filtered out in the understanding process? Answering all these "how"questions requires a process model capable of organizing large amounts of knowledge andmapping relevant aspects of past experience to new situations. Some meaningful starts have beenmade towards large-scale episodic.based memory organization [32,33, 34, 28, 25]and towardsepisodic-based analogical reasoning [9, 12, 7]. Bearing these questions in mind, we examine theissue of common sense reasoning in knowledge-rich mundane domains.Our central hypothesis is:Experiential reasoning hypothesis: Reasoning in mundane, experience-rich recurrentsituations is qualitatively different from formal, deductive reasoning evident in moreabstract, experimentally contrived, or otherwise non-recurrent situations (such as somemathematical or puzzle-solving domains).2. Carbonell and MintonIn the statement of our hypothesis we do not mean to exclude experience-rich metaphorical inferencefrom scientific or mathematical thought. Rather, we claim that formal deductive inference is definitelynot the dominant process in mundane reasoning. In essence, the experiential reasoning hypothesisstates that structuring new information according to relevant past experience is an important aspectof human comprehension -- perhaps more important than other aspects studied thus far in muchgreater depth.Common-sense experience-rich reasoning consists of recalling appropriate pest experiences andinference patterns, whereas solving abstract problems divorced frnm real-world experience requiresknowledge-poor search processes more typical of past and present Al problem solving systems.Since computer programs perform much better in simple, elegant, abstract domains than in "scruffy"experience-rich human domains, it is evident that a fundamental reasoning mechanism is lackingfrom the Al repertoire. The issue is not merely that Al systems lack experience in mundane humanscenarios .- they would be unable to benefit from such experience if it were encoded in theirknowledge base. We postulate that the missing reasoning method is based on the transfer of proveninference patterns and experiential knowledge across domains. This is not to say that humans areincapable of more formal reasoning, but rather that such reasoning is seldom necessary, and whenapplied it requires a more concerted cognitive effort than mundane metaphorical inference.There is evidence that human expertise, far beyond what we would label common sense reasoning,draws upon past experi6nce and underlying analogies. For instance, the master chess player is not abetter deductive engine than his novice counterpart. Rather, as Chase and Simon [14J have shown,he commands a vast repertoire of chess-board patterns and associated strategies that comprise hispast experience. And, when encountering a new chessboard situation he uses the relevant patterns(which may only partially match the current position) to index the appropriate knowledge. Mechanicsproblems in physics are often solved by creating a simple mental model -- an analog of the realsituation -- that preserves the significant properties. The model,