<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-3201">
  <Title>A Combined Phonetic-Phonological Approach to Estimating Cross- Language Phoneme Similarity in an ASR Environment</Title>
  <Section position="7" start_page="11" end_page="11" type="evalu">
    <SectionTitle>
5.4 Results
</SectionTitle>
    <Paragraph position="0"> Each recognition task includes about 3000 utterances of digit strings, command words, and sentences. The word accuracy results in Table 4 include the native baseline performance, i.e. the performance of the native monolingual, context-independent models from each target language, as well as the acoustics-based and feature-based performances. These results show that the performance of models selected by the CPP phoneme distance approach is equivalent overall to that of models selected by acoustic distance.</Paragraph>
    <Paragraph position="1">  The performance of models selected by the CPP approach nearly matches the performance of the native models for Latin American Spanish and surpasses those for Italian. This approach performs better than the acoustic distance approach for Latin American Spanish, Italian, and Japanese and not as well for Danish and European Portuguese.</Paragraph>
  </Section>
class="xml-element"></Paper>