This paper provides an introduction to many-sorted logics and motivates their use for representation and reasoning. Perhaps the most important reason to be interested in many-sorted logic is that computational efficiency can be achieved because the search space can be smaller and the length of a derivation shorter than in unsorted logic. There are many possible many-sorted logics of varying degrees of expressiveness, and the dimensions in which many-sorted logics differ are outlined and logics at various points in this space described. The relationship of manysorted logic to unsorted logic is discussed and the reason why many-sorted logics derivations may be shorter is. demonstrated. The paper concludes with a discussion of some many-sorted logic programming languages and some implementation issues.
I n t r o d u c t i o nLogic has been associated with AI since its earliest days: for example McCarthy's Advice Taker project (McCarthy, 1968) was formulated using logic as a knowledge representation language. Moreover, logic has a wide following among AI practitioners and researchers today as a glance at the proceedings of almost any AI conference will confirm. There are several ways in which logic and AI relate. Logic may be used directly as a representation language (e.g. Green, 1969; Michalski, 1983;Hayes, 1985), as an implementation language for AI tools and systems through the use of a logic programming language (e.g. Bundy et al., 1979;Bratko, 1986), logic may be applied as tool to analyse the semantics of formerly ad hoc representation systems (e.g. Hayes, 1979) or as a formalism in which to build a rational reconstruction of an existing but poorly understood AI program (e.g. Bundy, 1983b).Of course logic has also received much criticism from within the AI community over the years and continues to do so (for example see the debate started by McDermott (1987) in Vol. 3(3) of the journal Computational Intelligence). This is not the place to further this debate; however, two criticisms that have been often repeated are of importance to the present context: logic as a knowledge representation language has been criticized for being 'flat' and 'unstructured' (e.g. Minsky, 89