<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-2008">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics Towards Conversational QA: Automatic Identification of Problematic Situations and User Intent [?]</Title>
  <Section position="4" start_page="57" end_page="57" type="relat">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> Open domain question answering (QA) systems are designed to automatically locate answers from large collections of documents to users' natural language questions. In the past few years, automated question answering techniques have advanced tremendously, partly motivated by a series of evaluations conducted at the Text Retrieval Conference (TREC) (Voorhees, 2001; Voorhees, 2004). To better facilitate user information needs, recent trends in QA research have shifted towards complex, context-based, and interactive question answering (Voorhees, 2001; Small et al., 2003; Harabagiu et al., 2005). For example, NIST initiated a special task on context question answering in TREC 10 (Voorhees, 2001), which later became a regular task in TREC 2004 (Voorhees, 2004) and 2005. The motivation is that users tend to ask a sequence of related questions rather than isolated single questions to satisfy their information needs.</Paragraph>
    <Paragraph position="1"> Therefore, the context QA task was designed to investigate the system capability to track context through a series of questions. Based on context QA, some work has been done to identify clarification relations between questions (Boni and Manandhar, 2003). However context QA is different from interactive QA in that context questions are specified ahead of time rather than incrementally as in an interactive setting.</Paragraph>
    <Paragraph position="2"> Interactive QA has been applied to process complex questions. For analytical and non-factual questions, it is hard to anticipate answers. Clarification dialogues can be applied to negotiate with users about the intent of their questions (Small et al., 2003). Recently, an architecture for interactive question answering has been proposed based on a notion of predictive questioning (Harabagiu et al., 2005). The idea is that, given a complex question, the system can automatically identify a set of potential follow-up questions from a large collection of question-answer pairs. The empirical results have shown the system with predictive questioning is more efficient and effective for users to accomplish information seeking tasks in a particular domain (Harabagiu et al., 2005).</Paragraph>
    <Paragraph position="3"> The work reported in this paper addresses a different aspect of interactive question answering.</Paragraph>
    <Paragraph position="4"> Both issues raised earlier (Section 1) are inspired by earlier work on intelligent conversational systems. Automated identification of user intent has played an important role in conversational systems. Tremendous amounts of work has focused on this aspect (Stolcke et al., 2000). To improve dialog performance, much effort has also been put on techniques to automatically detect errors during interaction. It has shown that during human machine dialog, there are sufficient cues for machines to automatically identify error conditions (Levow, 1998; Litman et al., 1999; Hirschberg et al., 2001; Walker et al., 2002). The awareness of erroneous situations can help systems make intelligent decisions about how to best guide human partners through the conversation and accomplish the tasks.</Paragraph>
    <Paragraph position="5"> Motivated by these earlier studies, the goal of this paper is to investigate whether these two issues can be applied in question answering to facilitate intelligent conversational QA.</Paragraph>
  </Section>
class="xml-element"></Paper>