In this chapter we briefly described the main methods of reasoning: induction, a type of learning by making generalizations from examples; abduction, or explanation by reasoning back from a situation to the state or action that may have produced it; spatial reasoning, by forming a mental image; and deductive reasoning using classical logic.

One popular method of representing the meaning of words in a way that enables a computer to carry out reasoning is the semantic network. A semantic network describes the relationships between word concepts, by means of directed links. We introduced three special types of link: ISA, INST, and CONNECTS. The INST link relates an individual (e.g., Tweetie) to its general class (e.g., canary), and ISA links a class (e.g., canary) to its superclass (e.g., bird). Both links are transitive -- that is, they allow inferences like "If Tweetie is a canary and a canary is a kind of bird, then Tweetie is a bird". The ISA and INST links permit property inheritance, a form of inferencing whereby properties of a class or superclass are inherited by the subclass or instance. Given that Tweetie is a canary and canaries have the property `can sing', an inference can be made that Tweetie has the property `can sing'. The CONNECTS link is both transitive and symmetric. If A connects B and B connects C, then we can infer that A connects C, C connects A, B connects A, and C connects B.

The meaning of generic nodes (words like `canary' that stand for a class of objects) can be interpreted in different ways, and we introduced three interpretations: word-senses (the meaning of `canary' includes the meaning of `bird'), sets ('canary' represents the set of individuals like Tweetie and friends, and is a subset of `bird'), and prototypes (a `canary' is prototypically yellow and can sing).