Gross-grained RST through XML metadata for multilingual document generation

Gross-grained RST through XML Metadata for Multilingual Document Generation Guillermo Barrutieta Departmento de Informática Mondragon Unibertsitatea Loramendi, 4 20500 Arrasate (Spain) gbarrutieta@eps.muni.es Joseba Abaitua Filosofía y Letras Universidad de Deusto Avda. de las Universidades, 24 48007 Bilbao (Spain) abaitua@fil.deusto.es JosuKa Díaz ESIDE Universidad de Deusto Avda. de las Universidades, 24 48007 Bilbao (Spain) josuka@eside.deusto.es Abstract We present an RST-based discourse annotation proposal used in the construction of a trial multilingual XML-tagged corpus of teaching material in Basque, English and Spanish. The corpus feeds an experimental multilingual document generation system for the web. The main contributions of this paper are an implementation of RST through XML metadata and the adoption of gross-grained RST to avoid non-isomorphism in multilingual corpora. Keywords Corpus, RST, XML Background The natural language generation community has conducted research in many aspects of multilingual generation systems. One of these aspects is the construction of multilingual corpora as the source to generation (Jurafsky et al. 2000). In this paper we present a novel attempt to model a series of teaching units into an XML-tagged multilingual corpus for its use in a document generation system. The document generation system leans on a multilingual master document (Hirst et al., 1997) that contains text snippets that are combined in an adequate way to satisfy the needs of the users (or readers in this context) dynamically depending on the moment, level of expertise, time available and some other user aspects. Figure 1: General schema of the multilingual document generation system The creation of the multilingual master document met a number questions that are listed below: - What are the parts of a technical-educational document? - What is the optimum segmentation level? Is it the clause, or the discourse unit? - What are the relations among those parts or segments? - In the domain of technical education, how can we formalize information at discourse level? - How can this information be stored formally and in a reusable way for the purpose of document generation? Rhetorical Structure Theory Most of these questions found an adequate answer within the Rhetorical Structure Theory (RST) of Mann and Thompson (1988). RST granted a sound theoretical base and some helpful points of reference. According to RST, a text may be divided into minimal text spans that have an unambiguous discourse function (nuclei and satellites) and relations between these text spans can be established. Documents in our corpus were segmented into minimal text spans that have an unambiguous discourse function; that is, nuclei which hold the core message and satellites that support the core message of the text. Nuclei retain the essential content to understand the text and satellites can be replaced or erased without changing the general meaning . RST caters also with a set of possible relations between nuclei and satellites. There is a wide-range of relations that link nuclei and satellites such as elaboration, background, justification, evidence, concession and so on. After manually annotating the discourse (Marcu 1999b) a tree is obtained. A discourse tree looks like …

Content truncated to 500 words