<?xml version="1.0" standalone="yes"?>
<Paper uid="C86-1067">
  <Title>A Kana-Kanji Translation System for Non-Segmented Input Sentences Based on Syntactic and Semantic Analysis</Title>
  <Section position="5" start_page="284" end_page="284" type="evalu">
    <SectionTitle>
4. EXPERIMENT
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="284" end_page="284" type="sub_section">
      <SectionTitle>
4.1 System Implementation
</SectionTitle>
      <Paragraph position="0"> The experimental system developed by the authors is shown in Figure 4. The system consists of three subsystems: a translation control program, a morphological analysis program and a syntactic and semantic analysis program. The total size of the system is about 35K steps in PL/I. Two large dictionaries are also developed: a word dictionary of 160,000 entries and a case</Paragraph>
    </Section>
    <Section position="2" start_page="284" end_page="284" type="sub_section">
      <SectionTitle>
4.2 Experimental Results
</SectionTitle>
      <Paragraph position="0"> Both the upper limit and the substantial level of the accuracy of the new Kana-Kanji translation system was determined experimentally. The upper limit of translation accuracy was determined using a set of benchmark texts consisting of 9 typical Japanese texts including official documents, scientific papers and legal documents. The total number of input characters in the benchmark text was about ten thousand. Program errors and data errors in dictionaries were corrected to as great an extent as possible. The accuracy of the system using the benchmark texts was 94.9%. Another set of Japanese texts was prepared with twenty thousand chracters and the translation exreriment was repeated. This time no correction of data errors was made during the experiment.</Paragraph>
      <Paragraph position="1"> The average accuracy was 90.5%, which is the current level of performance of our system.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>