<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0104">
  <Title>GeoName: a system for back-transliterating pinyin place names</Title>
  <Section position="6" start_page="10" end_page="10" type="concl">
    <SectionTitle>
6. WWW
</SectionTitle>
    <Paragraph position="0"> confirmation 7. Evaluate probability; rank according to tag, name character length, probability to the desired input names. The best result is returned when all the processes are employed including checking on the bilingual List-A. Apparently many of our input names appear on this list, and it leads to simple table-lookup for the back-transliteration. This is probably not surprising because the bilingual map is not large (2'x3'), and it would only show the more well- known cities. Thus for the tag=111 run, it is seen that the correct candidates at rank 1 increase to 116 (71.6%), and if up to rank 10 candidates are included, 140 (86.4%) of the correct names are identified.</Paragraph>
    <Paragraph position="1"> Conclusion We have described GeoName, a system to back-transliterate English Pinyin geographic names to Chinese characters based on bilingual list lookup, monolingual place name character frequency, and Web confirmation. Evaluation using Pinyin city names shows that nearly 72% of the names suggested are correct at rank 1, and over 86% of correct names are included in the top ten candidates.</Paragraph>
    <Paragraph position="2"> The evaluation is small involving only 162 city names. One needs larger scale studies with more obscure names or names actually in use. The resources we employed are rather limited. We intend to improve our training data, as well as our formula for name suggestion. Bilingual resources are difficult to locate. We are exploring how to use the Web as a gigantic bilingual name list in order to improve our system further.</Paragraph>
  </Section>
class="xml-element"></Paper>