A Grammatical Approach to DCG Parsing

Definite Clause Grammars can be implemented in many different ways, using alternative parsing algorithms such as SLD-resolution, chart parsing either top-down or bottom-up, and so on. We present here a formal framework in which most of these techniques can be expressed. The two issues of building a parse-tree and handling non-determinism are considered separately. A grammar is first compiled into an intermediate form that encodes the construction of the corresponding language's parse-trees. This intermediate form is a machine that belongs to a class of extended push-down automata (X-automata). The extension takes into account the unifications occurring at each node of the tree. The X-automata execution, when non-deterministic, can make use of either a backtracking or a tabulation device. X-automata provide a single formalism to encode alternative implementations of the same grammar, and in so doing, offer a way to compare parser performance. Furthermore, static analysis of X-automata can be designed independently of the implementation the automata encode. We give some experimental results comparing the efficiency of the X-automaton implementation of the most common parsing techniques.