Abstract

Analogies ;are powerful ways to understand how things work in a new domain. We think this is because analogies enable people to construct a structure-mapping that carries across the way the components in a system interact. This allows people to create new mental models that they can then run to generate predictions about what should happen in various situations in the real world. This paper shows how analogies can be used to construct models of evaporation and how two subjects used such models to reason about evaporation. As Lakoff and Johnson (1980) have documented, our language is full of metaphor and analogy. People discuss conversation as a physical transfer: (e .g ., "Let's see if I can get this across to you" (Reddy 1979). They analogize marriage to a manufactured object : (e .g ., "They had a basic solid foundation in their marriages that could be shaped into something good" (Quinn this volume). They speak of anger as a hot liquid in a container (Lakoff & Kovecses this volume) ; and they describe their home thermostat as analogous to the accelerator on a car (Kempton this volume) ` Why are analogies so common? What exactly are they doing for us?'? We believe people use them to create generative mental models, models } they can use to arrive at new inferences. In this paper, we first discuss the general notion of a generative mental model, using three examples of artificial intelligence models of qualitative physics ; second, we lay out the analogy hypothesis of the paper, which we illustrate in terms of the component analogies that enter into mental models of evaporation ; and finally, we describe how two subjects used these analogies in reasoning about,' evaporation. The notion of running a generative model can be illustrated by an example from Waltz (1981). People hearing "The dachshund bit the mailman on the nose" spontaneously imagine scenarios such as the dachshund standing on a ledge, or the mailman bending down to pet the dachshund. Similarly , if you try to answer the question, "How far can you throw a potato chip?" your thought processes may have the feel of a mental simulation. Examples such as these suggest that simulation and generative inference are integral to language understanding (Waltz 1981). However, such