<?xml version="1.0" standalone="yes"?>
<Paper uid="P98-2234">
  <Title>Some Properties of Preposition and Subordinate Conjunction Attachments*</Title>
  <Section position="8" start_page="1441" end_page="1441" type="concl">
    <SectionTitle>
7 Discussion
</SectionTitle>
    <Paragraph position="0"> This paper presents a system for attaching prepositions and subordinate conjunctions that just relies on easy-to-find constructs like noun groups to determine when it is applicable. In sample text, we find that the system is applicable for trying to attach 89% of the prepositions/subordinate conjunctions that are outside of the easy-to-find constructs and is 75.4% correct on the attachments that it tries to handle.</Paragraph>
    <Paragraph position="1"> In this sample, we also notice that these attachments very much tend to be to only one or two different spots and that the attachment problems can be divided into 6 categories. One just needs those easy-to-find constructs to determine the category of an attachment problem.</Paragraph>
    <Paragraph position="2"> The 75.4% results may seen low compared to parsing results like the 88% precision and recall in (Collins, 1997), but those parsing results include many easier-to-parse constructs. (Manning and Carpenter, 1997) presents the VNPN example phrase &amp;quot;saw the man with a telescope&amp;quot;, where attaching the preposition incorrectly can still result in 80% (4 of 5) recall, 100% precision and no crossing brackets. Of the 4 recalled constructs, 3 are easy-to-parse: 2 correspond to noun groups and 1 is the parse top level.</Paragraph>
    <Paragraph position="3"> In our experiments, we found that limiting the choice of possible attachment points to the two most likely ones improved performance.</Paragraph>
    <Paragraph position="4"> This limiting also lets us use the large training sets lt- and It2-. In addition, we found that different training data produces rules that work better in different categories. This latter result suggests trying a system architecture where each attachment category is handled by the rule set most suited for that category.</Paragraph>
    <Paragraph position="5"> In the best overall result, nearly half of the remaining errors occur in one category, ~npn, so this is the category in need of most work.</Paragraph>
    <Paragraph position="6"> Another topic to examine is how many of the remaining attachment errors actually matter.</Paragraph>
    <Paragraph position="7"> For instance, when one's interest is on finding a semantic interpretation of the sentence &amp;quot;They flash letters on a screen. &amp;quot;, whether on attaches to flash or to letters is irrelevant. Both the letters are, and the flashing occurs, on a screen~</Paragraph>
  </Section>
class="xml-element"></Paper>