<?xml version="1.0" standalone="yes"?>
<Paper uid="E06-1009">
  <Title>Information Presentation in Spoken Dialogue Systems</Title>
  <Section position="3" start_page="65" end_page="66" type="intro">
    <SectionTitle>
2 Previous Work in Information
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="65" end_page="66" type="sub_section">
      <SectionTitle>
Presentation
2.1 Tailoring to a User Model
</SectionTitle>
      <Paragraph position="0"> Previous work in natural language generation showed how a multi-attribute decision-theoretic model of user preferences could be used to determine the attributes that are most relevant to mention when generating recommendations tailored to a particular user (Carenini and Moore, 2001). In the MATCH system, Walker et al. (2004) applied this approach to information presentation in SDS, and extended it to generate summaries and comparisons among options, thus showing how the model can be used to determine which options to mention, as well as the attributes that the user will find most relevant to choosing among them. Evaluation showed that tailoring recommendations and comparisons to the user increases argument effectiveness and improves user satisfaction (Stent et al., 2002).</Paragraph>
      <Paragraph position="1"> MATCH included content planning algorithms to determine what options and attributes to mention, but used a simple template based approach to realization. In the FLIGHTS system, Moore et al. (2004) focussed on organizing and expressing the descriptions of the selected options and attributes, in ways that are both easy to understand and memorable. For example, Figure 2 shows a description of options that is tailored to a user who prefers flying business class, on direct flights, and on KLM, in that order. In FLIGHTS, coherence and naturalness of descriptions were increased by reasoning about information structure (Steedman,  2000)tocontrolintonation,usingreferringexpressions that highlight attributes relevant to the user (e.g., &amp;quot;the cheapest flight&amp;quot; vs. &amp;quot;a KLM flight&amp;quot; ), and signalling discourse relations (e.g., contrast) with appropriate intonational and discourse cues.</Paragraph>
      <Paragraph position="2"> S: You can fly business class on KLM, arriving at four twentyp.m., butyou'dneedtoconnectinLondon. There is a direct flight on BMI, arriving at four ten p.m., but it has no availability in business class.</Paragraph>
      <Paragraph position="3">  This prior work demonstrated that the user model-based approach can concisely present a relatively small number of options, pointing out the ways in which those options satisfy user preferences. It is an appropriate strategy for SDS when there are a small number of options to present, either because the number of options is limited or because users can supply sufficient constraints to winnow down a large set before querying the data-base of options.</Paragraph>
      <Paragraph position="4"> However, thereareseverallimitationstothisapproach. First, it does not scale up to presenting a large number of options. When there are hundreds of options to consider (e.g., when choosing among consumer products, hotels, or restaurants) there may be many options that are close in score.</Paragraph>
      <Paragraph position="5"> In addition, users may not be able to provide constraints until they hear more information about the space of options. This brings up a second problem with the UM-based approach, namely that it does not provide the user with an overview of the option space, because options scoring below a specified threshold are not mentioned. This is related to the third problem, which is that users might miss out on options they would have chosen if they had heard about them. These last two problems may reduce user confidence in the system, if users have the perception that the system is not telling them about all of the available options. This may ultimately lead to a decrease in user satisfaction.</Paragraph>
    </Section>
    <Section position="2" start_page="66" end_page="66" type="sub_section">
      <SectionTitle>
2.2 Stepwise Refinement through Clustering
and Summarization
</SectionTitle>
      <Paragraph position="0"> Polifroni et al. (2003) developed an approach that structures large datasets for summarization and successive refinement. It supports the user in narrowing in on a suitable option by clustering the options in the database, summarizing the option clusters and suggesting further constraints. In this approach, the prompts presented to the user, and the order in which they appear are determined at run-time based on an algorithm that computes the most useful set of attributes, as dictated by the current data subset. For large data sets, attributes that partition the data into the minimal number of clusters are chosen, so that a concise summary can be presented to the user to refine, see Figure 3.</Paragraph>
      <Paragraph position="1"> M: What restaurants do you have? S: I have found 983 restaurants. Most of them are located in Boston and Cambridge. There are 32 choices for cuisine. I also have information about price range.</Paragraph>
      <Paragraph position="2"> M: Okay tell me about the ones in Boston.</Paragraph>
      <Paragraph position="3"> S: I have found 401 restaurants in Boston. There are 29 choices for cuisine.</Paragraph>
      <Paragraph position="4"> M: Do you have any that serve seafood? S: I have found 19 seafood restaurants. They are predominantly in Back Bay, the North End, South Boston and the South End.</Paragraph>
      <Paragraph position="5"> M: In the South End.</Paragraph>
      <Paragraph position="6"> S: I have found 3 restaurants. Grill Fish on Columbus Avenue, McCormick &amp; Schmick's on Columbus Avenue and Joseph's on High Street.</Paragraph>
      <Paragraph position="7"> Figure3: Dialoguebetweensimulator(M)andPolifroni system (S) Polifroni et al.'s approach was extended by Chung (2004), who proposed a constraint relaxation strategy for coping with queries that are too restrictive to be satisfied by any option. Qu and Beale (2003) had previously addressed the problem of responding to user queries with several constraints and used linguistic cues to determine which constraints had to be relaxed. Our discussion and evaluation of the SR approach is based on Chung's version.</Paragraph>
      <Paragraph position="8"> Although the SR approach provides a solution to the problem of presenting information when there are large numbers of options in a way that is suitable for SDS, it has several limitations. First, there may be long paths in the dialogue structure. Because the system does not know about the user's preferences, the option clusters may contain many irrelevant entities which must be filtered out successively with each refinement step. In addition, the difficulty of summarizing options typically increases with their number, because values are more likely to be very diverse, to the point that a summary about them gets uninformative (&amp;quot;I found flights on 9 airlines.&amp;quot;).</Paragraph>
      <Paragraph position="9"> A second problem with the SR approach is that exploration of tradeoffs is difficult when there is no optimal option. If at least one option satisfies all requirements, this option can be found efficiently with the SR strategy. But the system does not point out alternative tradeoffs if no &amp;quot;optimal&amp;quot; option exists. For example, in the flight booking domain, suppose the user wants a flight that is cheap and direct, but there are only expensive directandcheapindirectflights. IntheSRapproach, as described by Polifroni, the user has to ask for cheap flights and direct flights separately and thus has to explore different refinement paths.</Paragraph>
      <Paragraph position="10"> Finally, the attribute that suggests the next user constraint may be suboptimal. The procedure for computing the attribute to use in suggesting the next restriction to the user is based on the considerations for efficient summarization, that is, the attribute that will partition the data set into the smallest number of clusters. If the attribute that is best for summarization is not of interest to this particular user, dialogue duration is unnecessarily increased, and the user may be less satisfied with the system, as the results of our evaluation suggest (see section 5.2).</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>