Parsing String Generating Hypergraph Grammars

A string generating hypergraph grammar is a hyperedge replacement grammar where the resulting language consists of string graphs i.e. hypergraphs modeling strings. With the help of these grammars, string languages like anbncn can be modeled that can not be generated by context-free grammars for strings. They are well suited to model discontinuous constituents in natural languages, i.e. constituents that are interrupted by other constituents. For parsing context-free Chomsky grammars, the Earley parser is well known. In this paper, an Earley parser for string generating hypergraph grammars is presented, leading to a parser for natural languages that is able to handle discontinuities. 1 Discontinuous Constituents in German One (of many) problems when parsing German are discontinuous constituents [1]. Discontinuous constituents are constituents which are separated by one or more other constituents and still belong together on a semantic or syntactic level. An example for a discontinuous constituent is (1) Er hat schnell gearbeitet. He has fast worked. He (has) worked fast. The verb phrase hat gearbeitet ((has) worked) is distributed; the finite verb part, the auxiliary verb hat (has), is always in the second position in a German declarative sentence. The infinite verb part, the past participle gearbeitet (worked), is usually in the last position of a declarative sentence, only a few exceptions like relative clauses or appositions can be put after the infinite verb part. Another (more complicated) German example of discontinuous constituents is 1 The German examples are first translated word by word into English to explain the German sentence structure and then reordered into a correct English sentence. 2 The present perfect in German can be translated either in present perfect or in past tense in English.