<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-0603">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics How and Where do People Fail with Time: Temporal Reference Mapping Annotation by Chinese and English Bilinguals Yang Ye SS</Title>
  <Section position="4" start_page="13" end_page="13" type="relat">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> There are two basic types of temporal location relationships. The first one is the ternary classification of past, present and future. The second one is the binary classification of &amp;quot;BEFORE&amp;quot; versus &amp;quot;AFTER&amp;quot;. These two types of temporal relationships are intrinsically related but each stands as a separate issue and is dealt with in different works.</Paragraph>
    <Paragraph position="1"> While the &amp;quot;BEFORE&amp;quot; versus &amp;quot;AFTER&amp;quot; relationship can easily be transferred across a language pair, the ternary tense taxonomy is often very hard to transfer from one language to another.</Paragraph>
    <Paragraph position="2"> (Wilson, et al., 1997) describes a multilingual approach to annotating temporal information, which involves flagging a temporal expression in the document and identifying the time value that the expression designates. Their work reports an inter-annotator reliability F-measure of 0.79 and 0.86 respectively for English corpora.</Paragraph>
    <Paragraph position="3"> (Katz, et al., 2001) describes a simple and general technique for the annotation of temporal relation information based on binary interval relation types: precedence and inclusion. Their annotation scheme could benefit a range of NLP applications and is easy to carry out.</Paragraph>
    <Paragraph position="4"> (Pustejovsky et al., 2004) reports an annotation scheme, the TimeML metadata, for the markup of events and their anchoring in documents. The annotation schema of TimeML is very fine-grained with a wide coverage of different event types, dependencies between events and times, as well as &amp;quot;LINK&amp;quot; tags which encode the various relations existing between the temporal elements of a document. The challenge of human labeling of links among eventualities was discussed at great length in their paper. Automatic &amp;quot;time-stamping&amp;quot; was attempted on a small sample of text in an earlier work of (Mani, 2003). The result was not particularly promising. It showed the need for a larger quantity of training data as well as more predictive features, especially on the discourse level. At the word level, the semantic representation of tenses could be approached in various ways depending on different applications. So far, their work has gone the furthest towards establishing a broad and open standard metadata mark-up language for natural language texts.</Paragraph>
    <Paragraph position="5"> (Setzer, et al., 2004) presents a method of evaluating temporal order relation annotations and an approach to facilitate the creation of a gold standard by introducing the notion of temporal closure, which can be deduced from any annotations through using a set of inference rules.</Paragraph>
    <Paragraph position="6"> From the above works, it can be seen that the effort in temporal information annotation has thus far been dominated by annotating temporal relations that hold entities such as events or times explicitly mentioned in the text. Cross-linguistic tense and aspect annotation has so far gone unstudied. null</Paragraph>
  </Section>
class="xml-element"></Paper>