<artificial intelligence, programming, language> A declarative, relational style of programming based on first-order logic. The original logic programming language was Prolog. The concept is based on Horn clauses.
The programmer writes a "database" of "facts", e.g.
wet(water).("water is wet") and "rules", e.g.
mortal(X) :- human(X).("X is mortal is implied by X is human"). Facts and rules are collectively known as "clauses".
The user supplies a "goal" which the system attempts to prove using "resolution" or "backward chaining". This involves matching the current goal against each fact or the left hand side of each rule using "unification". If the goal matches a fact, the goal succeeds; if it matches a rule then the process recurses, taking each sub-goal on the right hand side of the rule as the current goal. If all sub-goals succeed then the rule succeeds.
Each time a possible clause is chosen, a "choice point" is created on a stack. If subsequent resolution fails then control eventually returns to the choice point and subsequent clauses are tried. This is known as "backtracking".
Clauses may contain logic variables which take on any value necessary to make the fact or the left hand side of the rule match a goal. Unification binds these variables to the corresponding subterms of the goal. Such bindings are associated with the choice point at which the clause was chosen and are undone when backtracking reaches that choice point.
The user is informed of the success or failure of his first goal and if it succeeds and contains variables he is told what values of those variables caused it to succeed. He can then ask for alternative solutions.