<?xml version="1.0" standalone="yes"?>
<Paper uid="W96-0418">
  <Title>Matchmaking: dialogue modelling and speech generation meet*</Title>
  <Section position="6" start_page="176" end_page="176" type="evalu">
    <SectionTitle>
LOW
</SectionTitle>
    <Paragraph position="0"> If the system is confident that it has understood what the user said, it would ask only to confirm what it believes to know, hence it would choose a declarative with tone 2 (\[answering:positive/answering-toquestion:strong\]). If the confidence is somewhat lower, there are two ways of realizing a yes/no-question: tone 2a (\[interrogative:yes/no-type:informationseeking:unmarked:neutral-assessment\]) or tone 2 (\[interrogative:yes/no:request\]). Finally, if the system has not at all understood what the user said, it could indicate this by using a clarifying wh-question with tone 4 (\[interrogative:wh-type:wh-tonic:clarifying\]).</Paragraph>
    <Paragraph position="1"> Utterance F) is also in response to a user inform, but what makes this situation different from the response above is that here, there is a mismatch between what the user wanted and what the system can offer (User wanted 3 o'clock, while system can only offer 14.52).</Paragraph>
    <Paragraph position="2"> Hence the system must offer the user an alternative and the linguistic form of this utterance might differ with the &amp;quot;closeness&amp;quot; of the alternative to the original demand.</Paragraph>
    <Paragraph position="3"> If the alternative is reasonably close (In our example, there is a time difference of 8 minutes, which, for this scenario, might be considered a good alternative), we find it appropriate to generate a yes/no-question with tone 2b (\[interrogative:yes/no-type:informationseeking:unmarked:strong-assessment\]). The lack of good alternatives, however, might condition a wh-question (&amp;quot;What is your next preferred departure time?&amp;quot;) with tone 1 (\[interrogative:wh-type:whnontonic :neutral- involvement \] ) . Inform The system answers the user's question, i.e., it is \[giving/information\], and hence the speech function is statement. Statements of this type do not need any particular intonational marking, since at this point they are expected, hence we choose the features \[ declarative:stating:neutral:ndegnemphatic:ndegncontrastive\], i.e., tone la. E.g., (&amp;quot;Eine einfache Fahrt kostet 6DM&amp;quot; (=&amp;quot;The ticket costs 6DM.&amp;quot;).</Paragraph>
    <Paragraph position="4"> Promise The information knower can utter a promise when she wants to signal the information seeker that she is considering the request. For instance, &amp;quot;Ich DURCHSUCHE die datenbank.&amp;quot; (= &amp;quot;I am searching&amp;quot;). A promise move is always in response to a request move and the relevant partial structure is:</Paragraph>
    <Paragraph position="6"> The speech function is statement since the system tures we choose \[declaratwe:stating:neutral:nonemphatic:non-contrastive\], hence tone la. As indicated in the partial structure above, an assert move can follow a promise act. This is additional information that the system volunteers the user, which often take the form of a polite command, e.g., &amp;quot;Bitte warren Sie.&amp;quot;(= &amp;quot;Please wait.&amp;quot;). Linguistically~ commands are realized as imperatives and hence tone 1.</Paragraph>
    <Paragraph position="7"> Request in Withdraw A request in the context of any of the unexpected dialogue moves (e.g., withdrawrequest) mostly serves as confirmation question similar to the responding requests in inform that we discussed above. This is, however, an unexpected move on the part of the user, hence we suggest that these requests, again mapping to question on the speech functional level, are realized as yes/no-question (as opposed to declarative with tone 2, see above) i.e., &amp;quot;Do you want to quit?&amp;quot; vs. &amp;quot;You want to quit?&amp;quot;. Which tone one chooses for this type of question depends on how involved one wants the system to appear. Tone 4a indicates neutral involvement, while tone 4b signals strong involvement. The partial structure of this type of requests is as follows:  Offer In an information retrieval system, the system often offers the user a list of alternatives from which she has to choose one. If we consider the appearance of a list on the screen a metaphor for actually handing over an object, this situation corresponds to \[demanding/goods ~4 services\], i.e., the speech function is command, hence we suggest that offers are realized as imperatives with tone 1. E.g., &amp;quot;Bitte wS~hlen Sie eins.&amp;quot; (=&amp;quot;Please choose one.&amp;quot;) Summary The above discussion is summarized in Table 1. Further, from the data that we have collected so far we observe: * The dialogue move guides the selection of speech function, e.g., request corresponds to speech function question, whereas offer maps to command.</Paragraph>
    <Paragraph position="8"> * The dialogue history, or context, guides the selection of semantic choices, i.e., pure initiating moves (e.g., request) correspond to \[exchanging/initiating\], while responding initiating moves (e.g., inform(request)) correspond to \[exchanging/responding\] in a first grammar traversal and \[exchanging/initiating\] in a second.</Paragraph>
    <Paragraph position="9"> * Choices in MOOD and KEY systems can often not be made unless we have access to additional knowledge sources as, for instance, a confidence measure. is \[giving/information\], and as MOOD and KEY fea- Future empirical studies will determine whether these  alell systems, ':' indicate refinement of the previous choice. WH = wh-question, Y/N = yes/no question, A = answering, S = stating, I = imperative.</Paragraph>
  </Section>
class="xml-element"></Paper>