<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-0133">
  <Title>Maximum Entropy Word Segmentation of Chinese Text</Title>
  <Section position="6" start_page="187" end_page="187" type="concl">
    <SectionTitle>
4 Conclusions
</SectionTitle>
    <Paragraph position="0"> Using a maximum entropy approach based on a modification of the system described by Low, Ng, and Guo (2005), our system was able to achieve a respectable level of accuracy when evaluated on the corpora of the word segmentation task of the Third International Chinese Language Processing Bakeoff. Implementing the Viterbi decoding algorithm was very beneficial for F scores and OOV recall rates. However, it should be investigated whether the rest of the added features, especially the outcome-dependent ones, are useful in general or if they were only beneficial for the 2005 test data due to some pattern in that data, after which they were modeled.</Paragraph>
  </Section>
class="xml-element"></Paper>