<?xml version="1.0" standalone="yes"?>
<Paper uid="P98-1031">
  <Title>Named Entity Scoring for Speech Input</Title>
  <Section position="7" start_page="204" end_page="204" type="concl">
    <SectionTitle>
4. Conclusion
</SectionTitle>
    <Paragraph position="0"> We have generalized the MUC text-based named entity scoring procedure to handle non-identical underlying texts. Our algorithm can also be used to score other kinds of non-embedded SGML mark-up, e.g., part-of-speech, word segmentation or noun- and verb-group. Despite its generality, the algorithm is backwardcompatible with the original MUC algorithm. The distinction made by the algorithm between extent and content allows speech understanding systems to achieve a partial score on the basis of identifying a region as containing a name, even if the recognizer is unable to correctly identify the content words. Encouraging this sort of partial correctness is important because it allows for applications that might, for example, index radio or video broadcasts using named entities, allowing a user to replay a particular region in order to listen to the corresponding content.</Paragraph>
    <Paragraph position="1"> This flexibility also makes it possible to explore information sources such as prosodics for identifying regions of interest even when it may  be difficult to achieve a completely correct transcript, e.g., due to novel words.</Paragraph>
  </Section>
class="xml-element"></Paper>