Modeling discontinuous constituents with hypergraph grammars

Abstract

Discontinuous constituents are a frequent problem in natural language analyses. A constituent is called discontinuous if it is interrupted by other constituents. In German they can appear with separable verb prefixes or relative clauses in the Nachfeld. They can not be captured by a context-free Chomsky grammar. A subset of hypergraph grammars are string-graph grammars where the result of a derivation must be formed like a string i.e. terminal edges are connected to two nodes and are lined up in a row. Nonterminal edges do not have to fulfill this property. In this paper it is shown that a context-free string-graph grammar (one hyperedge is replaced at a time) can be used to model discontinuous constituents in natural languages. © Springer-Verlag 2004.