<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1131">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics Incremental generation of spatial referring expressions in situated dialog [?]</Title>
  <Section position="7" start_page="1046" end_page="1047" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> We have argued that an if a conversational robot functioning in dynamic partially known environments needs to generate contextually appropriate locative expressions it must be able to construct a context model that explicitly marks the spatial relations between objects in the scene. However, the construction of such a model is prone to the issue of combinatorial explosion both in terms of the number objects in the context (the location of each object in the scene must be checked against all the other objects in the scene) and number of inter-object spatial relations (as a greater number of spatial relations will require a greater number of comparisons between each pair of objects.</Paragraph>
    <Paragraph position="1"> We have presented a framework that addresses this issue by: (a) contextually defining the set of objects in the context that may function as a landmark, and (b) sequencing the order in which spatial relations are considered using a cognitively motivated hierarchy of relations. Defining the set of objects in the scene that may function as a landmark reduces the number of object pairs that a spatial relation must be computed over. Sequencing the consideration of spatial relations means that in each context model only one relation needs to be checked and in some instances the agent need not compute some of the spatial relations, as it may have succeeded in generating a distinguishing locative using a relation earlier in the sequence.</Paragraph>
    <Paragraph position="2"> A further advantage of our approach stems from the partitioning of the context into those objects that may function as a landmark and those that may not. As a result of this partitioning the algorithm avoids the issue of infinite recursion, as the partitioning of the context stops the algorithm from distinguishing a landmark using its target.</Paragraph>
    <Paragraph position="3"> We have employed the approach in a system for Human-Robot Interaction, in the setting of object  manipulation in natural scenes. For more detail, see (Kruijff et al., 2006a; Kruijff et al., 2006b).</Paragraph>
  </Section>
class="xml-element"></Paper>