<?xml version="1.0" standalone="yes"?>
<Paper uid="W96-0401">
  <Title>The HealthDoc Sentence Planner</Title>
  <Section position="4" start_page="0" end_page="0" type="intro">
    <SectionTitle>
(1) Alternative reference:
</SectionTitle>
    <Paragraph position="0"> In some instances, an implant wears out, loosens, or fails. If this happens, it will have to be removed.</Paragraph>
    <Paragraph position="1"> (2) Alternative lexical choice: In some instances, an implant wears out, loosens, or fails.</Paragraph>
    <Paragraph position="2"> If replacement is needed, it will have to be removed.</Paragraph>
    <Paragraph position="3">  (3) Removal of redundancy  (aggregation): In some instances, an implant wears out, loosens, or fails, and \[\] will have to be removed. In this paper we describe the sentence planner in the HealthDoc project. HealthDoc \[Di-Marco et al., 1995\] was established in early 1995 with the goal of generating customized patient-education documents. It combines existing generation technology--the sentence generator KPML \[Bateman, 1995\] 1 and its input notation sPL \[Kasper, 1989\]--and new systems, such as the sentence planner described here. The sentence planner embodies a design that we hope has some general applicability in bridging the gap between text planners and sentence generators. Its input is a specification of the desired output content (a patient document) written in Tezt Source Language (TSL), see Subsection 2.3; its output consists of one or more SPL expressions. Its general operation is to recombine, enhance, and refine TSL expressions until they are adequately specific SPL expressions.</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
1.1 Sentence Planning Tasks
</SectionTitle>
      <Paragraph position="0"> After analysis of a number of patient-education documents, including those on diabetes, cholesterol, and hormone replacement therapy, we have identified the following most important sentence planning tasks: * Fine-graln discourse structuring: Discourse relations (aST relations, for example) that conjoin clause-size pieces in the TSL are still open to considerable variation of expression, such as the inclusion of a discourse marker or the lexical or implicit communication of the discourse relation. See, for example, \[Scott and de Souza, 1990\] for treatment of ELABORATION, \[Vander Linden and Martin, 1995\] of PURPOSE, and \[Grote et al., 1995\] of CONCESSION.</Paragraph>
      <Paragraph position="1"> * Sentence grouping and sentence content determination: Individual sentences must be delimited; temporal, spatial, and causal nuances of predicates must be determined, and so on \[Meteer, 1991; Pustejovsky, 1995; Stede, 1996\]. * Clause-internal structuring: The order of clause constituents, taxis, and projectivity of propositions within the clause \[Hovy, 1992; DiMarco and Hirst, 1993; Panaget, 1994\] must be determined; within each sentence, the thematized and focused elements must be identified \[Iordanskaja, 1992\]; redundancy must be removed \[Dalianis and Hovy, 1996a; Dalianis and Hovy, 1996b\].</Paragraph>
      <Paragraph position="2">  is a development of the Penman system \[Penman Project, 1989\].</Paragraph>
      <Paragraph position="3"> choice): The particular form of coreference (including anaphora, deixis, and ellipis) and reference must be chosen in order to maintain discourse cohesion \[McDonald, 1978; Tutin and Kittredge, 1992; Dale and Reiter, 1995\].</Paragraph>
      <Paragraph position="4"> * Exophoric lexical choice: As argued in \[Nirenburg et al., 1989; Meteer, 1991; Wanner, 1994\], lexical choice other than linguistic reference should also be considered a sentence planning task, since lexical units predetermihe the syntactic structure of a clause, and since salience may be realized by lexical means.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>