<?xml version="1.0" standalone="yes"?>
<Paper uid="P00-1010">
  <Title>Robust Temporal Processing of News</Title>
  <Section position="7" start_page="0" end_page="0" type="relat">
    <SectionTitle>
6 Related Work
</SectionTitle>
    <Paragraph position="0"> The most relevant prior work is (Wiebe et al. 98), who dealt with meeting scheduling dialogs (see also (Alexandersson et al. 97), (Busemann et al. 97)), where the goal is to schedule a time for the meeting. The temporal references in meeting scheduling are somewhat more constrained than in news, where (e.g., in a historical news piece on toxic dumping) dates and times may be relatively unconstrained. In addition, their model requires the maintenance of a focus stack. They obtained roughly .91 Precision and .80 Recall on one test set, and .87 Precision and .68 Recall on another.</Paragraph>
    <Paragraph position="1"> However, they adjust the reference time during processing, which is something that we have not yet addressed.</Paragraph>
    <Paragraph position="2"> More recently, (Setzer and Gaizauskas 2000) have independently developed an annotation scheme which represents both time values and more fine-grained inter-event and event-time temporal relations. Although our work is much more limited in scope, and doesn't exploit the internal structure of events, their annotation scheme may be leveraged in evaluating aspects of our work.</Paragraph>
    <Paragraph position="3"> The MUC-7 task (MUC-7 98) did not require VALs, but did test TIMEX recognition accuracy. Our 98 F-measure on NYT can be compared for just TIMEX with MUC-7 (MUC-7 1998) results on similar news stories, where the best performance was .99 Precision and .88 Recall. (The MUC task required recognizing a wider variety of TIMEXs, including event-dependent ones.</Paragraph>
    <Paragraph position="4"> However, at least 30% of the dates and times in the MUC test were fixed-format ones occurring in document headers, trailers, and copyright notices. ) Finally, there is a large body of work, e.g., (Moens and Steedman 1988), (Passoneau 1988), (Webber 1988), (Hwang 1992), (Song and Cohen 1991), that has focused on a computational analysis of tense and aspect. While the work on event chronologies is based on some of the notions developed in that body of work, we hope to further exploit insights from previous work.</Paragraph>
    <Paragraph position="5"> Conclusion We have developed a temporal annotation specification, and an algorithm for resolving a class of time expressions found in news.</Paragraph>
    <Paragraph position="6"> The algorithm, which is relatively knowledge-poor, uses a mix of hand-crafted and machine-learnt rules and obtains reasonable results.</Paragraph>
    <Paragraph position="7"> In the future, we expect to improve the integration of various modules, including tracking the temporal focus in the time resolver, and interaction between the eventorder and the event-aligner. We also hope to handle a wider class of time expressions, as well as further improve our extraction and evaluation of event chronologies. In the long run, this could include representing event-time and inter-event relations expressed by  In the last step after years of preparation, the countries &lt;lex eindex=&amp;quot;9&amp;quot; precedes=&amp;quot;10|&amp;quot; TIME=&amp;quot;19981231&amp;quot;&gt;locked&lt;/lex&gt; in the exchange rates of their individual currencies to the euro, thereby &lt;lex eindex=&amp;quot;10&amp;quot; TIME=&amp;quot;19981231&amp;quot;&gt;setting&lt;/lex&gt; the value at which the euro will begin &lt;lex eindex=&amp;quot;11&amp;quot; TIME=&amp;quot;19990104&amp;quot;&gt;trading&lt;/lex&gt; when financial markets open around the world on &lt;TIMEX VAL=&amp;quot;19990104&amp;quot;&gt;Monday&lt;/TIMEX&gt;.......</Paragraph>
  </Section>
class="xml-element"></Paper>