<?xml version="1.0" standalone="yes"?>
<Paper uid="P96-1050">
  <Title>A Synopsis of Learning to Recognize Names Across Languages</Title>
  <Section position="4" start_page="357" end_page="358" type="evalu">
    <SectionTitle>
3 Experiment
</SectionTitle>
    <Paragraph position="0"> The system was first built for English and then ported to Spanish and Japanese. For English, the training text consisted of 50 messages obtained from the English Joint Ventures (E/V) domain MUC-5 corpus of the US Advanced Research Projects Agency (ARPA). This data was hand-tagged with the locations of companies, persons, locations, dates, and &amp;quot;other&amp;quot;. The test set consisted of 10 new messages from the same corpus.</Paragraph>
    <Paragraph position="1"> Experimental results were obtained by applying the generated trees to test texts. Proper names which are voted into more than one class are handled by choosing the highest priority class. Priorities are determined based on the independent accuracy of each tree. The metrics used were recall (R), precision (P), and an averaging measure, P&amp;R, defined as:</Paragraph>
    <Paragraph position="3"> Obtained results for English compare to the English resuits of Rau (1992) and McDonald (1993). The  weighted average of P&amp;R for companies, persons, locations, and dates is 94.0% (see Table 2). The date grammar is rather small in comparison to other name classes, hence the performance for dates was perfect. Locations, by contrast, exhibited the lowest performance. This can be attributed mainly to: (I) locations are commonly associated with commas, which can create ambiguities with delimitation, and (2) locations made up a small percentage of all names in the training set, which could have resulted in overfitting of the built tree to the training data.</Paragraph>
    <Paragraph position="4"> Three experiments were conducted for Spanish.</Paragraph>
    <Paragraph position="5"> First, the English trees, generated from the feature set optimized for English, are applied to the Spanish text (E-E-S). In the second experiment, new Spanishspecific trees are generated from the feature set optimized for English and applied to the Spanish test text (S-E-S). The third experiment proceeds like the second, except that minor adjustments and additions are made to the feature set with the goal of improving performance (S-S-S).</Paragraph>
    <Paragraph position="6"> The additional resources required for the first Spanish experiment (E-E-S) are a Spanish POS tagger (Farwell et aL, 1994) and also the translated feature set (including POS) optimally derived for English. The second and third Spanish experiments (S-E-S, S-S-S) require in addition pre-tagged Spanish training text using the same tags as for English.</Paragraph>
    <Paragraph position="7"> The additional features derived for S-S-S are shown in Table 1 (FN/LN=given/family name, NNP=proper noun, DE=&amp;quot;de&amp;quot;). Only a few new features allows for significant performance improvement.</Paragraph>
    <Paragraph position="8">  The same three experiments are being conducted for Japanese. The first two, E-E-J and J-E-J, have been completed; J-J-J is in progress. Table 2 summarizes performance results and compares them to other work.</Paragraph>
  </Section>
class="xml-element"></Paper>