<?xml version="1.0" standalone="yes"?>
<Paper uid="A00-1012">
  <Title>Preceding word</Title>
  <Section position="8" start_page="87" end_page="87" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> This paper has introduced the problem of sentence boundary detection on the text produced by an ASR system as an area of application for NLP technology.</Paragraph>
    <Paragraph position="1"> An attempt was made to determine the level of human performance which could be expected for the task. It was found that there was a noticeable difference between the observed performance for mixed and upper case text. It was found that the kappa statistic, a commonly used method for calculating inter-annotator agreement, could not be applied directly in this situation.</Paragraph>
    <Paragraph position="2"> A memory-based system for identifying sentence boundaries in ASR text was implemented. There was a noticeable difference when the same system was applied to text which included case information demonstrating that this is an important feature for the problem.</Paragraph>
    <Paragraph position="3"> This paper does not propose to offer a solution to the sentence boundary detection problem for ASR transcripts. However, our aim has been to highlight the problem as one worthy of further exploration within the field of NLP and to establish some baselines (human and algorithmic) against which further work may be compared.</Paragraph>
  </Section>
class="xml-element"></Paper>