<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-2309">
  <Title>But What Do They Mean? An Exploration Into the Range of Cross-Turn Expectations Denied by &amp;quot;But&amp;quot;</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Motivation
</SectionTitle>
    <Paragraph position="0"> The main motivation behind modelling cross-turn relations is to get at what expectations and beliefs speakers might have upon interpreting the previous turn in the dialogue. Inferring the relations speakers perceive in cases where the related material spans speaker turns sheds light on how they interpret the previous speaker's turn, which in turn enables response generation that can specifically address these implicit relations. Here we focus on cases involving DofE, where the speaker of the &amp;quot;but&amp;quot; turn in dialogues like Ex.1 below has an expectation that beautiful people a0 marry, where a0 indicates defeasible implication. null (1) Example 1.</Paragraph>
    <Paragraph position="1"> A: Greta Garbo was the yardstick of beauty.</Paragraph>
    <Paragraph position="2"> B: But she never married.</Paragraph>
    <Paragraph position="3"> Thomas and Matheson (2003) argue that B has the expectation that beautiful people (usually) marry, and interpreting A's utterance triggers this expectation, which B knows does not hold, since he knows that Greta never married, denying the consequent of the rule. Hence he generates DofE, and depending on A's beliefs w.r.t. B's assertion that Greta never married or the inferred expectation that beautiful people marry that is being denied, she can respond accordingly. E.g., if she agrees with the assertion but disagrees with the expectation, she can respond &amp;quot;But beautiful people don't have to marry!&amp;quot; Thomas and Matheson (2003) focus on modelling DofE in Task-Oriented Dialogue (TOD). They present TOD examples like the following, (2) Example 2.</Paragraph>
    <Paragraph position="4"> A: Add the vinegar to the sauce.</Paragraph>
    <Paragraph position="5"> B1: (Yeah) But it's not tangy enough.</Paragraph>
    <Paragraph position="6"> B2: (Yeah) But we forgot to add the mushrooms.</Paragraph>
    <Paragraph position="7"> where B1 involves an expectation similar to the one above involving beautiful people marrying, namely, that adding vinegar makes things tangy, which is a general cause-effect relationship. However they argue in that paper that B2 involves satisfaction-precedence (s.p.) between adding vinegar and adding mushrooms, namely, that B expects adding mushrooms to be done before adding vinegar. They then went on to argue that TOD DofE should be distinguished from Nontask-Oriented Dialogue (NTOD) DofE, because of examples like Ex.2B2 above, where the DofE arises from the denial of an ordering of actions in B's task-plan.</Paragraph>
    <Paragraph position="8"> While we do not disagree with their claim that these s.p. DofEs in TOD (e.g., Ex.2B2) are distinct from causal cases like Ex.2B1, we disagree that these noncausal cases are unique to TOD; i.e., we argue for a unified treatment of DofE in TOD and NTOD, where, while search methods might differ (i.e., searching task-plans in TOD and private beliefs in NTOD), examples involving noncausal expectations which are denied are not unique to TOD.</Paragraph>
    <Paragraph position="9"> Consider the example below: (3) Example 3.</Paragraph>
    <Paragraph position="10"> A: Greta had a child in '43.</Paragraph>
    <Paragraph position="11"> B: But she married in '47.</Paragraph>
    <Paragraph position="12"> here we interpret B's &amp;quot;but&amp;quot; as signalling the denial of his expectation that marriage (usually) precedes having children in order to coherently interpret his response. The relation between turns (or antecedent and consequent) here is temporal ordering, and is very similar to the s.p. in the previous example (Ex.2B2). Unlike s.p., however, temporal ordering does not require the actions or states that temporally precede the later one to be achieved; i.e., the accomplishment aspect of s.p. is novel to planning, where goals are posted and accomplished, and there is a sense of agency. Temporal ordering relates actions, events, states, effects, etc, with no notion of agency involved. Prior work on DofE has not focussed much on the nature of the relation underlying the denied expectation, and we argue that this information will facilitate much more adaptive and appropriate response generation.</Paragraph>
  </Section>
class="xml-element"></Paper>