<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1414">
  <Title>Generic Querying of Relational Databases using Natural Language Generation Techniques</Title>
  <Section position="3" start_page="0" end_page="95" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Natural Language interfaces to databases (hereafter NLIDBs ) have long held an appeal to both the databases and NLP communities. However, difficulties associated with text processing, semantic encoding, translation to database querying languages and, above all, portability, have meant that, despite recent advances in the field, NLIDBs are still more a research topic than a commercial solution. null Broadly, research in NLIDBs has focused on addressing the following fundamental, inter- null The extent of NLIDB research is such that it is beyond the scope of this paper to reference a comprehensive list of projects in this area. For reviews on various NLIDBs , the reader is referred to (Androutsopoulos et al., 1995). * interpretation of the input query, including parsing and semantic disambiguation, semantic interpretation and transformation of the query to an intermediary logical form (Hendrix et al., 1978; Zhang et al., 1999; Tang and Mooney, 2001; Popescu et al., 2003; Kate et al., 2005); * translation to a database query language (Lowden et al., 1991; Androutsopoulos, 1992); * portability (Templeton and Burger, 1983; Kaplan, 1984; Hafner and Godden, 1985; Androutsopoulos et al., 1993; Popescu et al., 2003) In order to recover from errors in any either of these steps, most advanced NLIDB systems will also incorporate some sort of cooperative user feedback module that will inform the user of the inability of the system to construct their query and ask for clarification.</Paragraph>
    <Paragraph position="1"> We report here on a generic method we have developed to automatically infer the set of possible queries that can apply to a given database, and an interface that allows users to pose these questions in natural language but without the previously mentioned drawbacks of most NLIDBs . Our work is substantially different from previous research in that it does not require the user to input free text queries, but it assists the user in composing query through a natural language-like interface. Consequently, the necessity for syntactic parsing and semantic interpretation is eliminated. Also, since users are in control of the meaning of the query they compose, ambiguity is not an issue. Our work builds primarily on two directions of research: conceptual authoring of queries via  WYSIWYM interfaces, as described in section 2, and NLIDB portability research. From the perspective of the query composing technique, our system resembles early menu-based techniques, such as Mueckstein (1985), NL-Menu (Tennant et al., 1983) and its more recent re-development Lingo-Logic (Thompson et al., 2005). This resemblance is however only superficial. Our query editing interface employs natural language generation techniques for rendering queries in fluent language; it also allows the editing of the semantic content of a query rather than its surface form, which allows seamless translation to SQL .</Paragraph>
    <Paragraph position="2"> As in (Zhang et al., 1999), our system makes use of a semantic graph as a mean of representing the database model. However, whilst Zhang et al (1999) use the Semantic Graph as a resource for providing and interpreting keywords in the input query, we use this information as the main means of automatically generating query frames.</Paragraph>
  </Section>
class="xml-element"></Paper>