<?xml version="1.0" standalone="yes"?>
<Paper uid="W97-0909">
  <Title>NLP and Industry: Transfer and Reuse of Technologies*</Title>
  <Section position="3" start_page="0" end_page="57" type="intro">
    <SectionTitle>
2 Example: Mass Change of Text
</SectionTitle>
    <Paragraph position="0"> in that the procedure must search for candidate structured paraphrase sets (while abstracting away from surface noise) and then apply generation rules which are context dependent on the structures. The only alternative solution to this problem is to inspect and change the texts manually, a solution which is errorfraught and expensive. The original problem can be mitigated, however, by controlling the syntax and semantics of the text prior to populating the database by usingan authoring tool (for example, Boeing's &amp;quot;simplified English&amp;quot; system \[17\]).</Paragraph>
    <Paragraph position="1"> In the Boeing Company, millions of manufacturing operations texts exist in legacy databases. These texts are used by an on-line planning system to stage the manufacturing of aircraft. Because there are many manufacturing process threads, with varying degrees of changeability, and many analysts and other personnel who contribute to the collection of these texts, the databases are in constant flux and contain significant noise.</Paragraph>
    <Paragraph position="2"> When a sequence of operations must be modified, as when high volatile ozone-depleting organic compounds need to be replaced by those having low volatility, then all relevant texts must be retrieved, interpreted to understand whether they match the relevant conditions of the mass change, and then modified according to specified rules. Such a textual modification process requires robust normalization, complex pattern recognition, syntactic parsing, and semantic understanding of domain reference. Furthermore, inference is required to generate new texts based upon arbitrary change criteria.</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
2.1 The Problem
</SectionTitle>
      <Paragraph position="0"> On-line legacy databases are used daily by industry.</Paragraph>
      <Paragraph position="1"> Some of these databases consist of large amounts of relatively unconstrained texts constituting manufacturing plans and procedures, for example. These textual databases require periodic &amp;quot;mass changes&amp;quot; to correct errors and update procedures. &amp;quot;Mass change&amp;quot; means more than a simple &amp;quot;global search and replace&amp;quot; of text,</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="57" type="sub_section">
      <SectionTitle>
2.2 The Application
</SectionTitle>
      <Paragraph position="0"> Using formal language and NLP components, we customized.a procedure to effect the mass change of on-line textual databases for the circumscribed domain of chemical treatment, prime, and finish operations. These operations (represented as texts) are performed on the shop floor in a precisely determined sequence, dependent on the aircraft design requirements and the * This paper has benefited greatly from discussions with Gary Coen of the ACT center at Boeing, Philadelphia.  part under construction. The finish and rinse operations include applications of anodizers, primers, overcoats, and topcoats of a variety of compounds, thicknesses, and numbers of coats, to a range of treated or untreated parts of diverse material composition, and describe the manner in which the parts must be manipulated. The texts refer to these materials and processes directly (i.e., they name the materials and processes), indirectly (i.e., they name documents and standards which refer to the materials and processes), and in manners which combine direct and indirect reference. Various types of temporal and spatial information are present in the texts, including duration of finish application and drying time, and the location of areas to be finished or protected. Also present in the text are references to other documents, color codes, and miscellanous additional operations.</Paragraph>
      <Paragraph position="1"> Though circumscribed, the semantics of this domain is richly structured.</Paragraph>
      <Paragraph position="2"> Examples of some simple plan texts from this domain are displayed below (excluding database key information):  (1) PRIME (1) COAT ZOINC CHROMATE PRIMER PER VFI.1 (2) TOUCH UP REWORK AREA ONLY APPLY (1)</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>