<?xml version="1.0" standalone="yes"?>
<Paper uid="N03-3010">
  <Title>Cooperative Model Based Language Understanding in Dialogue</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> In this paper, we propose a novel language understanding approach, Cooperative Model, for a dialogue system. It combines both Finite State Model and Statistical Learning Model for sentence interpretation.</Paragraph>
    <Paragraph position="1"> This approach is implemented in the project MRE (Mission Rehearsal Exercise). The goal of MRE is to provide an immersive learning environment in which army trainees experience the sights, sounds and circumstances they will encounter in real-world scenarios (Swartout et al., 2001). In the whole procedure, language processing part plays the role to support the communication between trainees and computers.</Paragraph>
    <Paragraph position="2"> In the language processing pipeline, audio signals are first transformed into natural language sentences by speech recognition. Sentence interpretation part is used to &amp;quot;understand&amp;quot; the sentence and extract an information case frame for future processing such as dialogue management and action planning. We adopt the Cooperative Model as the overall frame of sentence interpretation, which incorporates two mainly used language processing approaches: the Finite State Model and the Statistical Learning Model. Currently there is relatively little work on the cooperation of the two kinds of models for language understanding.</Paragraph>
    <Paragraph position="3"> The Cooperative Model has great advantages. It balances the shortcomings of each separate model. It is easy to implement the parsing algorithm and get the exact expected result for finite state model (FSM) but it's difficult and tedious to design the finite state network by hand. Also, the finite state model is not too robust and the failure of matching produces no results. On the other hand, statistical learning model (SLM) can deal with unexpected cases during designing and training by giving a set of candidate results with confidence scores. It is a must to provide some kind of rules to select results needed. However, applying it may not give a completely satisfactory performance.</Paragraph>
    <Paragraph position="4"> The rest of this paper is organized as follows: Section 2 describes the case frame as the semantic representation produced by the cooperative model. In section 3, we explain our cooperative language understanding model and discuss two different strategies of the Finite State Model and the Statistical Learning Model. We analyze the experimental results in Section 4. Section 5 concludes with on-going research and future work.</Paragraph>
  </Section>
class="xml-element"></Paper>