<?xml version="1.0" standalone="yes"?>
<Paper uid="M91-1004">
  <Title>MUC-3 LINGUISTIC PHENOMENA TEST EXPERIMEN T</Title>
  <Section position="12" start_page="43" end_page="44" type="concl">
    <SectionTitle>
CONCLUSION S
</SectionTitle>
    <Paragraph position="0"> In summary, the systems scored differently on the appositives than they di d on the overall testing suggesting that the testing may be isolating the phenomeno n of apposition . The systems scored similarly on the slots filled from phrase s containing appositives and sentences containing appositives suggesting that information from sentences could be used to test phenomena . Because the processing o f apposition can affect the processing of the entire sentence, the parallel results i n these scores further suggests that the phenomenon of apposition is being isolated .</Paragraph>
    <Paragraph position="1"> The systems scored markedly higher on the simpler appositives as opposed to th e more complex ones . These results are perhaps the strongest evidence that it is pos sible to isolate the phenomenon of apposition by scoring slot fills . The system s scored differently on the postposed and preposed appositives . It would be interesting to look at the methods employed by each system with respect to these classes o f appositives. It was predicted that neither class would be clearly easier . The fact that this prediction was correct provides strong support for the claim that apposition i s being isolated . The systems scored lower on their responses to the messages where simple sentences were substituted for appositives . The effect on the scores, althoug h unexpected, still supports the isolatability of apposition . In some of the more well defined trends, the anomalies noticed are often for the lower scoring systems .</Paragraph>
    <Paragraph position="2"> However, the systems are scoring highly enough overall at this stage of development for the phenomena scores to be meaningful . In conclusion, there are strong indications that the phenomenon of apposition has been isolated by the testing and tha t performance on apposition can be scored using the MUC-3 scoring system .</Paragraph>
    <Paragraph position="3"> Further Researc h Further work in phenomena testing should now be focused on carefull y developing a representative selection of phenomena tests for the messages . The evaluation of data extraction systems can be enhanced by determining performanc e of the systems on these linguistic phenomena . Phenomena testing should be done a t various linguistic levels including the word level, phrase level, sentence level , intersententially, and the level of discourse .</Paragraph>
    <Paragraph position="4"> Testing according to the linguisti c characteristics of the messages would encourage the data extraction systems t o improve capabilities applicable to other domains .</Paragraph>
  </Section>
class="xml-element"></Paper>