<?xml version="1.0" standalone="yes"?>
<Paper uid="N04-3003">
  <Title>Limited-Domain Speech-to-Speech Translation between English and Pashto</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> This demonstration will present a prototype system for bidirectional speech-to-speech translation within a limited semantic domain, that of first encounters between a patient and a medical professional. A major goal of the work is to explore techniques that are appropriate for languages that are not of great commercial interest and that consequently are lacking in data and resources of many kinds. The system has been developed for American English and Pashto, one of the major languages of Afghanistan, which presents a variety of challenges for both data-intensive and knowledge-based approaches.</Paragraph>
    <Paragraph position="1"> It should be noted that the system must be viewed as one component in a real-time transaction between two cooperating humans. The ultimate goal of those humans is to exchange information by whatever means is effective: not, necessarily, to rely exclusively on the system's output, but to use it in combination with other, non-speech modalities of conveying meaning and with ordinary world knowledge.</Paragraph>
    <Paragraph position="2"> The final system is intended to run on a handheld device, such as a PDA, with its attendant memory and speed limitations and restriction to integer-only computation. Most components of the demonstration system run in a PocketPC emulation environment on a Windows laptop, with a few in the full Windows environment. All aspects of the prototype system are undergoing active development.</Paragraph>
  </Section>
class="xml-element"></Paper>