Automated Multi-purpose Text Processing 2 Previous and Current Research 3 a Theory of Syntactic Style 2

Multiple versions of a document may need to be produced for diierent purposes and so may require quite diierent stylistic structures. This paper describes how we are building stylistic control into both natural language parsers (Pundit) and generators (Pen-man) to handle variations in hospital patient educational materials. Stylistic knowledge is formally represented in terms of a multi-level stylistic grammar which deenes a systematic correspondance between low-level syntactic structures and high-level stylistic eeects. 1 The importance of style in natural language processing The importance of dealing with stylistic aspects of language in computational systems is undeniable. People communicate a great deal of information through stylistic nuances, and a knowledge of how these subtleties innuence meaning is part of a full understanding of language. Systems that could analyze the eeects of style on communication would provide information about the implicit meaning that is contained in a text. And generation systems that could control style would produce text that intentionally conveys a speciic communicative eeect. Both stylistic analysis and generation could be used in applications , such as text critiquing, second-language instruction , and machine translation, for which understanding the eeects of how something is said is as important as understanding what is said. Ultimately , computational stylistics should be a part of any system that attempts to deal with`real-world' language. But very few natural language understanding systems have attempted to deal with issues of style, 1 1 By style, we do not mean literary style, but and those that do have generally taken a simplistic and heuristic approach. Stylistic analysis has not yet developed the systematic and rigorous methods of syntactic analysis and semantic interpretation. Part of the reason is obvious: understanding style is hard. Stylistic eeects are diicult to articulate and even more diicult to deene. In the past several years, we have worked to address the problems of syntactic style, understanding how particular syntactic structures can convey corresponding stylistic eeects. We are currently involved in a collaborative research project with Dietmar RR osner's group at the Forschungsinstitut f ur anwen-dungsorientierte Wissensverarbeitung (Ulm, Baden-W urttemberg) on \Multi-purpose text generation from knowledge bases." The two participating sub-projects are both concerned with the generation of multiple versions of natural language text from a single knowledge base. In their TechDoc project, RR osner's group is addressing issues of multilingual generation in the production of technical manuals. In our HealthDoc project, we are developing systems …