<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1414">
  <Title>Generic Querying of Relational Databases using Natural Language Generation Techniques</Title>
  <Section position="7" start_page="100" end_page="100" type="concl">
    <SectionTitle>
5 Conclusions &amp; Further work
</SectionTitle>
    <Paragraph position="0"> Our method presents three main advantages over other natural language interfaces to databases:  1. It is easily customizable for new domains and databases.</Paragraph>
    <Paragraph position="1"> 2. It eliminates errors in parsing and query-to-SQL translation.</Paragraph>
    <Paragraph position="2"> 3. It makes clear to the user the full range of possible queries that can be posed to any given  database.</Paragraph>
    <Paragraph position="3"> From a user's point of view, one could argue that our method is less natural to use than one that allows unconstrained (or less constrained) natural language input. It could also be said that while syntactically correct, the queries as presented to the user may not be as fluent as human-authored questions. These possible disadvantages are, in our opinion, outweighed by the clarity of the query composition process, since the user is fully in control of the semantic content of the query she composes; they are unambiguous to both the user and the back-end system.</Paragraph>
    <Paragraph position="4"> We are currently extending this work to cover more complex queries that require inferences and ones that contain elements linked through temporal relations. We will also refine the query layout procedures to allow complex queries to be presented in a more intuitive way. Additionally, we are about to begin work on automating the the construction of the semantic graph. We expect that some of the semantic and syntactic information that, at the moment, has to be manually entered in the description of the semantic graph can be inferred automatically from the database content.</Paragraph>
  </Section>
class="xml-element"></Paper>