Core Concepts
Finite automata can be used to efficiently recognize graph languages by applying a modified powerset construction to turn nondeterministic automata into deterministic ones without the need for backtracking.
Abstract
The paper discusses the use of finite automata for the efficient recognition of graph languages. It starts by defining graphs with front and rear interfaces, and how they can be composed using a graph composition operation similar to string concatenation.
The authors then introduce finite automata over a typed alphabet of graph symbols, where each symbol represents a basic graph. They show that these automata can be used to recognize graph languages, but that nondeterminism in the automata can lead to the need for backtracking during recognition, which should be avoided for efficiency.
The main technical contribution of the paper is an extension of the classical powerset construction for finite automata, which can be used to turn a nondeterministic automaton over graph symbols into a deterministic one. The authors prove that the resulting automaton is equivalent to the original one in terms of the graph languages they recognize.
However, they also show that the powerset construction does not always result in a deterministic automaton, due to the presence of blank transitions that can lead to nondeterminism even in a deterministic automaton. To address this, the authors provide two sufficient conditions under which the powerset automaton can be made deterministic without the need for backtracking during recognition.