<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-3103">
  <Title>The Language of Bioscience: Facts, Speculations, and Statements in Between</Title>
  <Section position="4" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Related work
</SectionTitle>
    <Paragraph position="0"> We know of no work specifically on speculative speech in the context of text processing of bioscience literature. However, some work on information extraction from bioscience literature has dealt with speculative speech. For example, (Friedman et al., 1994) discusses uncertainty and hedging in radiology reports and their system assigns one of five levels of certainty to extracted findings.</Paragraph>
    <Paragraph position="1"> Text processing systems in general have focused &amp;quot;factual&amp;quot; language. However, a growing number of researchers have started work on other aspects of language such as expressing opinions, style of writing, etc. For example a human language technology workshop will be held this Spring entitled &amp;quot;Exploring Attitude and Affect in Text: Theories and Applications.&amp;quot; (Qu et al., 2004). Previous work along these lines includes (Wilson and Wiebe, 2003). This research focuses on newswire texts and other texts on the topic of politics and current events.</Paragraph>
    <Paragraph position="2"> There has been recent work on classifying sentences from MEDLINE abstracts for the categories such as object, background, conclusions (McKnight and Srinivasan, 2003). In addition, early work, (Liddy, 1988) built text grammars for empirical research abstracts categorized and assigned structure concerning rhetorical roles of the sentences. However, none of this work addresses the speculative vs.</Paragraph>
    <Paragraph position="3"> definite distinction we are interested in.</Paragraph>
    <Paragraph position="4"> There has also been some work on constructing test sets for knowledge discovery. Several researchers have used the discoveries by Swanson and Smalheiser to test their own algorithms. The two problems most commonly used in replication studies (e.g., (Weeber et al., 2001)) are their discovery of a link between Raynauds disease and fish oils (Swanson, 1986) and their discovery of several links between migraine and magnesium (Swanson, 1988). The most comprehensive replication to date is (Srinivasan, 2004) which employs eight Swanson and Smalheiser discoveries as a test bed.</Paragraph>
    <Paragraph position="5"> In the remainder of the paper, we describe a manual annotation experiment we performed, give preliminary results on our attempts to automatically annotate sentences as containing speculative fragments, and conclude with comments on possible future work.</Paragraph>
  </Section>
class="xml-element"></Paper>