<?xml version="1.0" standalone="yes"?>
<Paper uid="H86-1009">
  <Title>Model-based Analysis of Messages about Equipment</Title>
  <Section position="7" start_page="77" end_page="78" type="concl">
    <SectionTitle>
6. Conclusion
</SectionTitle>
    <Paragraph position="0"> We have described a new text-processing system, PROTEUS, for analyzing messages about equipment failure. We have focussed on its equipment model and the role of this model in the process of interpreting of noun phrases. This process is part of semantic analysis but also plays a role in syntactic analysis and discourse analysis.</Paragraph>
    <Paragraph position="1"> In addition to the elaboration of the existing components, substantial work will be required in at least two areas before we can hope to obtain a robust text processing system.</Paragraph>
    <Paragraph position="2"> First, we are developing a discourse component to identify temporal and plausible causal links between sentences. This information is needed not only for some of the applications (e~g., message summarization) but also to resolve some of the syntactic and semantic ambiguities in the messages. Second, we will need to move from a pass/fail strategy for enforcing our constraints to a best-fit strategy. Because of imperfections in the input, and the  inevitable omissions in a model as complex as ours, we must expect that many messages will violate one or another constraint; by employing a rich set of constraints, however, and selecting the analysis which violates the fewest constraints, we beleive that we will be able to identify the intended reading for most sentences.</Paragraph>
    <Paragraph position="3"> The initial motivation for the system has been the conversion of a stream of messages to a data base for subsequent querying, summarization, and trend analysis. However, the use of a detailed equipment model, similar to that employed in simulation and diagnostic systems, suggests that it may be equally useful as an interface for such systems. A diagnostic system, for example, would then be able to accept initial observations in the form of a brief textual summary rather than force the user to go through an elaborate questionnaire; this may be a substantial advantage for broad-coverage diagnostic systems, which must be able to accept a wide variety of different symptoms.</Paragraph>
  </Section>
class="xml-element"></Paper>