<?xml version="1.0" standalone="yes"?>
<Paper uid="W97-1001">
  <Title>A Trainable Message Understanding System*</Title>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
4 Scanning New Articles
</SectionTitle>
    <Paragraph position="0"> The goal of generalizing the rules is to generate semantic networks for unseen articles. The semantic networks are built with the help of the ADD.NODE and the ADD.RELATION operations present in the RHS of the rules. The Scanning Process consists of the following steps:  \[{enterprise}\], \[seek, VG, 1, other_type\], \[{applicant}\] &gt; ADD..NODE({enterprise}), ADD_NODE({applicant}), ADD.RELATION(seek, {enterprise}, {applicant}) \[{organization}\], \[seek, VG, 1, other_type\], \[{person}\] &gt; ADD_NODE({organization}), ADD..NODE({person}), ADD.RELATION(seek, {organization}, {person})</Paragraph>
    <Paragraph position="2"> * Use the Preprocessor to identify the type (Ti) for each headword.</Paragraph>
    <Paragraph position="3"> Use the Sense Classifier (as described in Section 2.5.1 to assign the appropriate sense (Si) to each headword.</Paragraph>
    <Paragraph position="4"> Each phrase can now be uniquely represented by (W~, C~, Si, 7~). Match (14~, Ci, Si, 7~) with the LHS of a generalized rule.</Paragraph>
    <Paragraph position="5"> If the three entities \[Generalize(spi, hi)\] subsume three phrases \[(W~, C~, S~, 7~)\], within a single sentence in the article, the rule is fired and the RHS of the rule executed.</Paragraph>
    <Paragraph position="6"> If we train on IBM Corporation seeks job candidates and generate the rule as in Figure 3, Table 1 lists some sentences that can be processed as the degree of generalization.</Paragraph>
  </Section>
class="xml-element"></Paper>