<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1413">
  <Title>The Clarity-Brevity Trade-off in Generating Referring Expressions [?]</Title>
  <Section position="4" start_page="89" end_page="90" type="metho">
    <SectionTitle>
3 Exploring the Trade-off
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="89" end_page="89" type="sub_section">
      <SectionTitle>
3.1 Varying penalties for distractors
</SectionTitle>
      <Paragraph position="0"> Imagine the following situation. You are preparing a meal in a friend's house, and you wish to obtain, from your own kitchen, a bottle of Italian extra virgin olive oil which you know is there. The only way open to you is to phone home and ask your young child to bring it round for you. You know that also in your kitchen cupboard are some distractors: one bottle each of Spanish extra virgin olive oil, Italian non-virgin olive oil, cheap vegetable oil, linseed oil (for varnishing) and camphorated oil (medicinal). It is imperative that you do not get the linseed or camphorated oil, and preferable that you receive olive oil. A full expression, Italian extra virgin olive oil, guarantees clarity, but may overload your helper's abilities. A very short expression, oil, is risky. You might well settle for the intermediate olive oil.</Paragraph>
      <Paragraph position="1"> To model this situation, fC could take a much higher value if [[ S ]] contains a distractor which  mustnotbeselected(e.g.varnishratherthancooking oil). That is, instead of a simple linear function of the size of [[ S ]], there is a curve where the cost drops more steeply as the more undesirable distractors are excluded. For example, each object could be assigned a numerical rating of how undesirable it is, with the target having a score of zero, and the fC value for a set A could be the maximum rating of any element of A. (This would, of course, require a suitably rich domain model.) The brevity cost function fB could still be a relatively simple linear function, providingfB values donotmasktheeffectoftheshapeofthefC curve.</Paragraph>
    </Section>
    <Section position="2" start_page="89" end_page="89" type="sub_section">
      <SectionTitle>
3.2 Fuzziness of target
</SectionTitle>
      <Paragraph position="0"> Suppose Mrs X has dropped a piece of raw chicken meat on the kitchen table, and immediately removed the meat. She would now like Mr X to wipe the area clean. The meat leaves no visiblestain, soshehastoexplainwhereitwas. Inthis case, it appears that there is no such thing as a distinguishing description (i.e. a description that pins down the area precisely), although Mrs X can arbitrarily increase precision, by adding properties: - the edge of the table, - the edge of the table, on the left (etc.) The ideal description would describe the dirty area and nothing more, but a larger area will also do, if not too large. Here, the domain D is implicitly defined as all conceivable subareas of the table, the target is again one element of D, but - unlike the traditional set-up with discrete elements a description (fuzzily) defines one such area, not a disjoint collection of individual items. Our fC operates on the description S, not just on the number of distractors, so it can assess the aptness of the denotation of any potential S. However, it has to ensure that this denotation (subarea of the surface) contains the target (contaminated area), and does not contain too much beyond that. Hence, we may need to augment our clarity cost function with another argument: the target itself. In general, more complex domains may need more complicated functions.</Paragraph>
    </Section>
    <Section position="3" start_page="89" end_page="90" type="sub_section">
      <SectionTitle>
3.3 Underspecification in dialogue
</SectionTitle>
      <Paragraph position="0"> Standard GRE algorithms assume that the speaker knows what the hearer knows (Dale and Reiter, 1995). In practice, speakers can often only guess.</Paragraph>
      <Paragraph position="1"> It has been observed that speakers sometimes produce referring expressions that are only disambiguated through negotiation with the hearer, as exemplified in the following excerpt (quoted in (Hirst, 2002)).</Paragraph>
      <Paragraph position="2">  1. A: What's that weird creature over there? 2. B: In the corner? 3. A: [affirmative noise] 4. B: It's just a fern plant.</Paragraph>
      <Paragraph position="3"> 5. A: No, the one to the left of it.</Paragraph>
      <Paragraph position="4"> 6. B: That's the television aerial. It pulls out. A and B are in the same room, in an informal setting, so A can be relatively interactive in convey null ing information. Also, the situation does not appear to be highly critical, in comparison to a military officer directing gunfire, or a surgeon guiding an incision. Initially, A produces an expression which is not very detailed. It may be that he thinks this is adequate (the object is sufficiently salient that B will uniquely determine the referent), or he doesn't really know, but is willing to make an opening bid in a negotiation to reach the goal of reference. In the former case, a GRE algorithm which took account of salience (e.g. (KrahmerandTheune, 1999)), operatingwithA'smodel of B's knowledge, should produce this sort of effect. (A dialogue model might also be needed.) In the latter case, we need an algorithm which can  relax the need for complete clarity. This could be arranged by havingfC give similar scores to denotations where there are no distractors and to denotations where there are just a few distractors, with fB making a large contribution to the cost.</Paragraph>
    </Section>
    <Section position="4" start_page="90" end_page="90" type="sub_section">
      <SectionTitle>
3.4 Over-specification
</SectionTitle>
      <Paragraph position="0"> Recently, interest has been growing in 'overspecified' referring expressions, which contain more information than is required to identify their intendedreferent. Someofthisworkismainlyorexclusively experimental (Jordan and Walker, 2000; Arts, 2004), but algorithmic consequences are also being explored (Horacek, 2005; Paraboni and van Deemter, 2002; van der Sluis and Krahmer, 2005).</Paragraph>
      <Paragraph position="1"> Over-specification could also arise in a dialogue situation (comparable to that in Section 3.3) if a speaker is unclear about the hearer's knowledge, and so over-specifies (relative to his own knowledge) to increase the chances of success.</Paragraph>
      <Paragraph position="2"> This goes beyond the classical algorithms, where the main goal is total clarity, with no reason for the algorithm to add further properties to an already unambiguous expression. That is, such algorithms assume that every description S for which  |[[ S ]] |= 1 has the same level of clarity (fC value). This assumption could be relaxed. For example, the approach of (Horacek, 2005) to GRE allows degrees of uncertainty about the effectiveness of properties to affect their selection. Within such a framework, one could separately compute costs for clarity (e.g. likelihood of being understood) and brevity (which might include the complexity of expressing the properties).</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>