In formal language theory, a cone is a set of formal languages that has some desirable closure properties enjoyed by some well-known sets of languages, in particular by the families of regular languages, context-free languages and the recursive languages.[1] The concept of a cone is a more abstract notion that subsumes all of these families.
More precisely, a cone is a non-empty family of languages such that, for any over some alphabet ,
The family of all regular languages is contained in any cone.
If one restricts the definition to homomorphisms that do not introduce the empty word then one speaks of a faithful cone; the inverse homomorphisms are not restricted. Within the Chomsky hierarchy, the regular languages, the context-free languages, and the recursively enumerable languages are all cones, whereas the context sensitive languages and the recursive languages are only faithful cones.
The terminology cone has a French origin. In the American oriented literature one usually speaks of a full trio. The trio corresponds to the faithful cone.
Contents |
A finite state transducer is a finite state automaton that has both input and output. It defines a transduction , mapping a language over the input alphabet into another language over the output alphabet. Each of the cone operations (homomorphism, inverse homomorphism, intersection with a regular language) can be implemented using a finite state transducer. And, since finite state transducers are closed under composition, every sequence of cone operations can be performed by a finite state transducer.
Conversely, every finite state transduction can be decomposed into cone operations. In fact, there exists a normal form for this decomposition, which is commonly known as Nivat's Theorem:[2] Namely, each such T can be effectively decomposed as , where are homomorphisms, and is a regular language depending only on .
Altogether, this means that a family of languages is a cone if it is closed under finite state transductions. This is a very powerful set of operations. For instance one easily writes a (nondeterministic) finite state transducer with alphabet that removes every second in words of even length (and does not change words otherwise). Since the context-free languages form a cone, they are closed under this exotic operation.