<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-0217">
  <Title>Probabilistic Dialogue Modelling</Title>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
3 Anaphora resolution
</SectionTitle>
    <Paragraph position="0"> Several different factors enter into the resolution of anaphora in a dialogue. How recently a potential referent was referred to is one important factor, another is the embedding activity within which the anaphoric reference was made (e.g. the type of verb phrase in which the referent appears), a third is the intra-sentential location of the relevant noun phrases in the preceding dialogue, a fourth is the relative prominence of a potential referent in the dialogue situation, and so on. The basic idea is that condi- null tional probability distributions are generated dynamically in the Bayesian network. When we look at a distribution corresponding to a node we are interested in, then the most salient object in the context will be the one whose value has the highest probability. null We point out that an obvious deterministic way to rank different combinations of factors (in an optimality-theoretic way, for example), and choose the most salient object based on this ranking, does not seem to work - because any potential ranking of principles (e.g. &amp;quot;Recency overrides subject placement&amp;quot;) would have to be context-dependent, and this defeats the idea of a ranking. See the examples in Figures 5 and 6.</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
3.1 Examples
</SectionTitle>
      <Paragraph position="0"> Here we work with two basic activities of the WITAS robot helicopter (see (Lemon et al., 2002)) which our dialogue system interfaces to - moving and searching. The helicopter can move to various landmarks (e.g. the tower, the church) and waypoints, and can see various objects (e.g. landmarks, vehicles). Our results use the network in Figure 1. In reading the tables (the figures appear after the references), use the following key: U=user, S=system, r=red car, g=green car, w=waypoint, s= search, m=move. All examples start with an even distribution of 0.2 for all variables (all objects are equally salient) at the start of each dialogue.</Paragraph>
      <Paragraph position="1"> 3.1.1 Activity and recency We will start with what may be the simplest type of example of anaphoric reference, in Figure 2. Here it is intuitively clear that &amp;quot;it&amp;quot; is intended to pick out the green car. The contribution of &amp;quot;see it&amp;quot; is modelled as an observed even distribution over all possible referents which can be seen (i.e. a1a3a2 and a4a5a2 each get a 0.5 weight). The conditional probability table for Salience List at time a0 is then used to compute the new probability distribution over the object-activity pairs (a1a7a6a9a8a10a4a11a6a9a8a13a12a14a6a9a8a13a1a3a2a11a8a10a4a5a2 ). Here we see that the green car is the most salient after the user's second utterance (a4a11a6 ), and that this salience increases after the utterance &amp;quot;it&amp;quot;, because a4 was both the most recent NP, and is also a possible object in the context of the &amp;quot;seeing&amp;quot; activity.</Paragraph>
      <Paragraph position="2"> In the example in Figure 3, the anaphoric pronoun &amp;quot;it&amp;quot; should pick out the red car and not the waypoint, even though the waypoint was referred to more recently. Intuitively, this is because the embedding activity of looking for the red car is tied to the pronoun, and this fact overrides the most recent referent. Here, the waypoint is not a possible object in the &amp;quot;seeing&amp;quot; activity, whereas the red car has been introduced as part of that activity. Thus the pronoun &amp;quot;it&amp;quot; in the user's final utterance has the effect of raising the probabilities of all the objects which can be seen, and this in fact overrides the recency effect of the utterance of &amp;quot;waypoint&amp;quot;.</Paragraph>
      <Paragraph position="3"> An extended example (not shown) shows how activity information can outweigh recency in an interleaved fashion and then that a newly introduced referent can become the most salient. Having considered the ways in which activity and recency interact in determining salience for anaphoric resolution, we then investigated adding another determining factor in the model - the syntactic placement of the referring expression.</Paragraph>
      <Paragraph position="4">  Figure 4 shows how subject placement influences availability for anaphoric reference. Here, the sub-ject (&amp;quot;red car&amp;quot;) of the user's second utterance is intuitively the one picked out by the later anaphoric expression, and not the green car, even though &amp;quot;the green car&amp;quot; is the most recent NP. See Figure 4 for our results, using an extension of the network in Figure 1, where the &amp;quot;Activity a0 &amp;quot; node was enriched to include syntactic information about the input - specifically, what referring expressions appear in subject and object places. Note here that the red car becomes the most salient object after the user's second utterance. We model the referential import of this sentence as an input triple &amp;quot;a1a1a0 a4 &amp;quot;to the Activity a0 node - denoting: red car (subject), no activity, green car (object). The updated table for this node ensures that objects in subject place are given more weight than those in object place.</Paragraph>
      <Paragraph position="5"> In Figure 5, the subject (&amp;quot;red car&amp;quot;) of the user's second utterance is intuitively the one picked out by the later anaphoric expression, and not the green car, even though &amp;quot;the green car&amp;quot; is involved in the &amp;quot;seeing&amp;quot; activity.</Paragraph>
      <Paragraph position="6"> In Figure 6 the red car is most salient after the second utterance, but the waypoint becomes more salient, even though the red car was in subject position, because the waypoint is involved in the activity of moving, as is the pronoun &amp;quot;it&amp;quot;, and so is a better candidate for anaphoric resolution. Combined with Figure 5 this shows that no static ranking of anaphoric binding principles will cover all situations, and that a probabilistic approach is useful even as a theoretical model.</Paragraph>
      <Paragraph position="7"> Obviously this model could be made more complex with representations for direct and indirect objects, and so forth, but we leave this for future work.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>