<?xml version="1.0" standalone="yes"?>
<Paper uid="C82-1033">
  <Title>SURFACE ANALYSIS OF QUERIES DIRECTED TOWARD A DATABASE</Title>
  <Section position="10" start_page="207" end_page="207" type="concl">
    <SectionTitle>
3.2.2 STATISTICS
</SectionTitle>
    <Paragraph position="0"> A statistical knowledge about words in queries can contain such information as  (a) the chance that a word with a certain role will appear in a given position in a query (b) the chance that a word with a certain role will appear in a specified position of a n-word pattern with the word roles in the other n-1 positions in the pattern specified (c) the chance that a word with a certain role will appear after (or before) a  specific vocabulary word This information is used reductively to resolve words that had more than one canditate role after the dictionaries are applied.</Paragraph>
    <Section position="1" start_page="207" end_page="207" type="sub_section">
      <SectionTitle>
3.3 PHRASES AND TEMPLATES
</SectionTitle>
      <Paragraph position="0"> The method to be used to identify the phrase boundaries is a non-serlal technique which uses keywords and the word roles identified by the word role identifier. This analysis is pattern driven and uses patterns developed from an extensive sample of actual DBMS queries.</Paragraph>
      <Paragraph position="1"> Progressive recognition of the use of words and word groups leads to the development of patterns which include both syntactic and semantic groups. For example, the following patterns represent query template skeletons: o &lt;ques. word&gt;&lt;existance verh&gt;&lt;desired info&gt;&lt;verb&gt; (&lt;prep&gt;&lt;reqd. sttr.&gt;) ~ WHAT&lt;desired info&gt;&lt;existance verh&gt;&lt;reqd, attr.&gt;(&lt;prep&gt;&lt;reqd, attribute&gt;) ' What happens is that the boundaries of a semantic group are delimited by syntactic units and a limited set of semantic purposes can be assigned to particular word groupn in the input query. Once the pragmatic intentionality of a group is recognized, this group can then he further analyzed to identify the specific roles of words and word groups within it.</Paragraph>
      <Paragraph position="2"> Once the initial analysis has been completed, phrase analysis mechanisms take over to transform the phrases into candidate template fragments. A template fragment contains that information in the phrase that is needed to accurately evaluate the _query. This includes identification of what is the desired informatlon~ attributes the desired information must have and actions requested of the system. After the individual phrases have been transformed into template fragments the template matching mechanism takes over.</Paragraph>
      <Paragraph position="3"> A pattern recognition approach selects the template that has the closest match between the information needed to complete the template and the information in the template fragments. A measure of fit &amp;quot;goodness&amp;quot; is developed and used to choose between competing interpretations. After the appropriate template has been selected the template matching mechanism completes the stereotyped query using information taken from the template fragments.</Paragraph>
      <Paragraph position="4"> An example of this process can be found in the authors&amp;quot; paper: &amp;quot;A Pattern Driven</Paragraph>
    </Section>
    <Section position="2" start_page="207" end_page="207" type="sub_section">
      <SectionTitle>
Analysis of Queries Directed Toward Existing Databases&amp;quot; (Mazlack~Felnauer, 1982).
3.4 SEMANTIC INFORMATION
</SectionTitle>
      <Paragraph position="0"> The content of the database is not directly referenced as an information source.</Paragraph>
      <Paragraph position="1"> The logical database schema is used as a primary semantic information source 212 L.J. MAZLACK and R.A. FEINAUER because it already exists separate from the natural language query system and does not have to be created when the natural language analyzer is implemented with a new database. One of the major problems problems with many existing natural language query systems is that they use significant information specific to the particular database they reference. By using information sources that do not have to be recreated for each new application, the amount of effort needed for new systems is reduced.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>