<?xml version="1.0" standalone="yes"?>
<Paper uid="C82-1021">
  <Title>A MULTILAYERED APPROACH TO THE HANDLING OF WORD FORMATION</Title>
  <Section position="9" start_page="133" end_page="133" type="evalu">
    <SectionTitle>
EVALUATION OF WORD FORMATION IN DIALOGUE CONTEXT
</SectionTitle>
    <Paragraph position="0"> Trying to evaluate entire utterances is the ultimate processing phase for accepting, reinterpreting or rejecting interpretations of word formations in HAM-ANS. Suppose the client in the hotel-reservation situation already knows about a desk in the room being offered and asks whether the desk chair is a comfortable one. Having interpreted the compound SCHREIBTISCHSTUHL (desk chair) as a chair which is conceptually located to a desk, the system would try to identify a referent with this property. According to the system's intentions (cf. JAMESON/WAHLSTER 1982) it might reject the existentially presupposed interpretation of 'desk chair', or it might find an appropriate referent and accept the interpretation. A third possibility, which is particular plausible in this communicative setting, is to take any chair in the neighbourhood of the desk and set it up as the object referred to. A similar case is the relaxation of one part of an additive compound, e.g. to agree to an object's property stated as 'dark brown', even if only 'brown' is a proper attribute according to the system's extensional knowledge and no contrary information relating to the obiect's brightness is available.</Paragraph>
    <Paragraph position="1"> It should be emphasized that the task of analyTing word formation in an ongoing dialogue is not finished when the system is able to interpret a compound or derivative in utterances or even has given a satisfactory reply. The knowledge gained through the analysis and the commitments to the interpretation chosen have to be integrated into the knowledge sources. At present two consequences are associated with a successful analysis: First, the conceptual and referential semantic networks are updated to allow for subsequent reference. Second, the knowledge sources representing the partner's assL:m!)tions about the domain of discourse are updated accordingly (cf. JAMESON/WAHLSTER 1982).</Paragraph>
    <Paragraph position="2"> The worst case conceivable appears to be misspelling in a way which allows the word 138 W. HOEPPNER to be acceptable on structural and semantic grounds but doesn't make sense in the context of the utterance and the dialogue. These cases, however, seem to be rare.</Paragraph>
  </Section>
class="xml-element"></Paper>