<?xml version="1.0" standalone="yes"?>
<Paper uid="W01-1309">
  <Title>From Temporal Expressions to Temporal Information: Semantic Tagging of News Messages</Title>
  <Section position="6" start_page="0" end_page="0" type="evalu">
    <SectionTitle>
derivation
6 Conclusion and outlook
</SectionTitle>
    <Paragraph position="0"> We presented a semantic tagging system that automatically tags the occurrence of temporal expressions such as 3. June, on Monday and last month for German news messages. In addition, a semantics for most of the temporal expressions was defined so that temporal inferences were drawn regarding dates and events described. A more complex set of temporal expressions as extracted by recent systems (e.g. (Mani and Wilson, 2000)) was tagged. Our definition of temporal expressions also includes PPs capturing temporal relations. The system achieved an overall precision rate of 84.49 which is likely to go up as soon as the semantic definition of all temporal expressions will be completed.</Paragraph>
    <Paragraph position="1"> Our system also covers indexical and vague temporal expressions. Temporal reasoning and pragmatic inferences drawn on the basis of these expression is the focus of on-going and future work.</Paragraph>
    <Paragraph position="2"> The system we described in the present paper is intended to become a part of an experimental multi-document summarisation system currently under development. Our studies focus on financial news messages obtained from on-line information services in Germany. The task the system has to solve is the production of summaries of the most recent -- and especially, most referred to -- topics. Our experience in this domain shows that there is one topic which leads to five to twenty news messages almost every day. These news messages are mostly unrelated, and they often only focus on the last one or two hours. Thus a bare collection of such messages is nearly useless for a reader who wants to be informed at the end of the day. For a user of an on-line information service summarisations of several articles on the same hot topics would have an enormous advantage compared to unsummarised collections of news messages.</Paragraph>
    <Paragraph position="3"> The processing of temporal expressions plays a major role in building up these summaries, because temporal information is ubiquitous in this class of news. In addition, developing stories are reported via a stream of in-coming news messages. Producing coherent news depends heavily on the correct extraction of temporal information expressed by these messages.12</Paragraph>
  </Section>
class="xml-element"></Paper>