<?xml version="1.0" standalone="yes"?>
<Paper uid="A94-1032">
  <Title>Automatic Aquisition of Semantic Attributes for User Defined Words m Japanese to English Machine Translation Satoru Ikehara *, Satoshi Shirai * LAkio Yokoo *</Title>
  <Section position="4" start_page="184" end_page="184" type="evalu">
    <SectionTitle>
4. Evaluation
</SectionTitle>
    <Paragraph position="0"> The proposed method was used to determine the SAs to create user dictionaries for translating newspaper articles and software design documents shown in Tabe 3.</Paragraph>
    <Paragraph position="1"> The following 3 methods were examined.</Paragraph>
    <Paragraph position="2">  (~) Automatic Determination (Proposed Method) (~) Manual Determination (Manual Method) (~) Experimental Determination (Correct Attributes) (1) Accuracy of Noun Type (Table 2)  In the case of newspaper articles, the method's accuracy in determining the noun type was 93.5%. Manual determination achieved an accuracy rote of 94.8%. Similar results were obtained for the software specification documents.</Paragraph>
    <Paragraph position="3">  using the same texts used in the above sectiom It can be seen in table 4 that using the automatically determined SAs improved the translation quality by 6-13%. This improvement is almost the same as that achieved with manually determined SAs. The translation success rate is 2-3% lower than that achieved with the correct attributes. This is, however, satisfactory if we consider the high cost needed to obtain the correct attribute by repeatedly tuning them.</Paragraph>
    <Paragraph position="4"> Thus, automatic determination makes it possible to acquire useful sets of SAs; a task which normally requires the most labor in creating user dictionaries.</Paragraph>
  </Section>
class="xml-element"></Paper>