<?xml version="1.0" standalone="yes"?>
<Paper uid="N04-1028">
  <Title>Non-Native Users in the Let's Go!! Spoken Dialogue System: Dealing with Linguistic Mismatch</Title>
  <Section position="6" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Conclusion and Future Directions
</SectionTitle>
    <Paragraph position="0"> In this paper, we described the Let's Go!! bus information system, a dialogue system targetted at non-native speakers of English. In order to investigate ways to improve the communication between non-native users and the system, we recorded calls from both native and non-native speakers and analyzed their linguistic properties. We found that besides the problem of acoustic mismatch that results from the differences in accent and pronunciation habits, linguistic mismatch is also significant and degrades the performance of the language model and the natural language understanding module. We are exploring two solutions to reduce the linguistic gap between native and non-native users. First we studied the impact of taking into account non-native data to model the user's language and second we designed a mechanism to generate confirmation prompts that both match the user's input and a set of predefined target utterances, so as to help the user acquire idiomatic expressions related to the task.</Paragraph>
    <Paragraph position="1"> Real-world systems like Let's Go!! are in constant evolution because the data that is collected from users calling the system is used to refine the acoustic and linguistic models of the system. In the near future, our priority is to collect more data to improve the acoustic models of the system and address the specific issues related to a general non-native population, which does not share a common native language. We will also work on integrating the confirmation prompt generation method proposed in this work with state-of-the-art confidence annotation methods.</Paragraph>
  </Section>
class="xml-element"></Paper>