A Graph Grammar Approach to Map Between Dependency Trees and Topological Models

Abstract

Determining the word order in free word order languages is deemed as a challenge for NLG. In this paper, we propose a simple approach in order to get the appropriate grammatically correct variants of a sentence using a dependency structure as input. We describe a linearization grammar based on a graph grammar that allows to retrieve a topological model using unordered constituent structures and precedence relations. The graph grammar formalism is totally language independent and only the grammar depends on the language. The grammar rules can be automatically acquired from a corpus that is annotated with phrase structures and dependency structures. The dependency structures annotation is retrieved by structure translation from the phrase structure annotation. We conclude with the description of a grammar and the evaluation of the formalism using a large corpus.