<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1672">
  <Title>Discriminative Methods for Transliteration</Title>
  <Section position="8" start_page="613" end_page="614" type="evalu">
    <SectionTitle>
6 Experiments
</SectionTitle>
    <Paragraph position="0"> We present transliteration experiments for three language pairs. We consider transliteration from Arabic, Korean, and Russian into English. For all language pairs, we apply the same training and decoding algorithms.</Paragraph>
    <Section position="1" start_page="613" end_page="614" type="sub_section">
      <SectionTitle>
6.1 Data
</SectionTitle>
      <Paragraph position="0"> The training and testing transliteration dataset sizes are shown in Table 1. For Arabic and Russian, we created the dataset manually by keying in and translating Arabic, Russian, and English names. For Korean, we obtained a dataset of transliterated names from a Korean government website. The dataset contained mostly foreign  names transliterated into Korean. All datasets were randomly split into training and (blind) testing parts.</Paragraph>
      <Paragraph position="1">  Prior to transliteration, the Korean words of the Korean transliteration data were converted from their Hangul (syllabic) representation to Jamo (letter-based) representation to effectively reduce the alphabet size for Korean. The conversion process is completely automatic (see Unicode Standard 3.0 for details).</Paragraph>
    </Section>
    <Section position="2" start_page="614" end_page="614" type="sub_section">
      <SectionTitle>
6.2 Algorithm Details
</SectionTitle>
      <Paragraph position="0"> For language modeling, we used the list of 100,000 most frequent names downloaded from the US Census website. Our language model is a 5-gram model with interpolated Good-Turing smoothing (Gale and Sampson 1995).</Paragraph>
      <Paragraph position="1"> We used the learning-to-classify version of Voted Perceptron for training local models (Freund and Schapire 1999). We used Platt's method for converting scores produced by learned linear classifiers into probabilities (Platt 1999). We ran both local and global Voted Perceptrons for 10 iterations during training.</Paragraph>
    </Section>
    <Section position="3" start_page="614" end_page="614" type="sub_section">
      <SectionTitle>
6.3 Transliteration Results
</SectionTitle>
      <Paragraph position="0"> Our discriminative transliteration models have a number of parameters reflecting the length of strings chosen in either language as well as the relative distance between strings.</Paragraph>
      <Paragraph position="1"> While we found that choice of W(E)=W(F) = 2 always produces the best results for all of our languages, the distance d(E,F) may have different optimal values for different languages.</Paragraph>
      <Paragraph position="2"> Table 2 presents the transliteration results for all languages for different values of d. Note that the joint probabilistic model does not depend on d. The results reflect the accuracy of transliteration, that is, the proportion of times when the top English candidate produced by a transliteration model agreed with the correct English transliteration. We note that such an exact comparison may be too inflexible, for many foreign names may have more than one legitimate English spelling.</Paragraph>
      <Paragraph position="3"> In future experiments, we plan to relax the requirement and consider alternative variants of transliteration scoring (e.g., edit distance, top-N candidate scoring).</Paragraph>
    </Section>
    <Section position="4" start_page="614" end_page="614" type="sub_section">
      <SectionTitle>
Values of Relative Distance (d).
</SectionTitle>
      <Paragraph position="0"> Table 2 shows that, for all three languages, the discriminative methods convincingly outperform the joint probabilistic approach. The global discriminative approach achieves the best performance in all languages. It is interesting that different values of relative distance are optimal for different languages. For example, in Korean, the Hangul-Jamo decomposition leads to fairly redundant strings of Korean characters thereby making transliterated characters to be relatively far from each other. Therefore, Korean requires a larger relative distance bound. In Arabic and Russian, on the other hand, transliterated characters are relatively close to each other, so the distance d of 1 suffices. While for Russian such a small distance is to be expected, we are surprised by such a small relative distance for Arabic. Our intuition was that omitting short vowels in spelling names in Arabic will increase d.</Paragraph>
      <Paragraph position="1"> We have the following explanation of the low value of d for Arabic from the machine learning perspective: incrementing d implies adding a lot of extraneous features to examples, that is, increasing attribute noise. Increased attribute noise requires a corresponding increase in the number of training examples to achieve adequate performance. While for Korean the number of training examples is sufficient to cope with the attribute noise, the relatively small Arabic training sample is not. We hypothesize that with increasing the number of training examples for Arabic, the optimal value of d will also increase.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>