<?xml version="1.0" standalone="yes"?>
<Paper uid="C92-2118">
  <Title>Exploiting Linguistic Iconism for Article Machine Translation</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> This paper is meant to give a cognltive-linguistic explanation of the process of reference. This means that we ate concerned with meaning arising from the speaker's conceptualization of reality. Different ways of referring to the s~me real situation are reflected differently on the expression side of language. We will exemplify this with the use of articles. We deal with two contrasting processes wh/.ch are reflected in NP syntax: on the one hand, this is the selection of a spedfic part of a MAss, which normally has an indefo inite extension. This process results in identification emd hence in ~oken reference (cf. Deelerck 1986:163; Croft 1985; Langacker 1987a). On the other hand we are concerned with type reference to COUNT entities (cf. ibd.), more specifically with how we can talk about the whole kind of an entity which in reality is represented by individual instances.</Paragraph>
    <Paragraph position="1"> Our ultimate aim is to exploit the cognitive principles by which reference is determined and to hnport them into Machine T~anslation (MT). Traditional MT systems so far have not been concerned with a semantic interpretation mid translation of articles. The translationai relevance of interpreting NP readings has been stressed in different places (cf. Meya 1990; Grabski 1990; gelinsky-Wibbelt 1988, 1901): Bounding by individuatioh of a MASS results in definiteness in Germanic languages (Das Wasser in der Tasse ist sehmutzig. 'The water in the cup is dirty.'). In English type reference to masses is usually expressed by the bare construction as in Wa~er is indispensable, which in German can be expressed both by the definite NP and by the bare construction as in (Das) Wasser ist nnverzichtbar (see e.g. ter Meulen 1988:390). In Spanish a D~.FINITZ NP is usually used for type reference (El espa~ol me gnsta rods que el raso. '1 like Spanish better than Russian.').</Paragraph>
    <Paragraph position="2"> Type reference to a COUNT entity by the subject NP may surface in two contrasting structures in French: Uu specta~enr veut voir quelque chose. 'A spectator wants to see something.' Le speeta~eur es~ uu ~tre humaiu. 'The spectator is a human being.' In this paper we will explain the conceptual conditions for type and token reference, which in turn establish the conditions for generating the correct surface structures in the respective target language. We interpret genericity vs. identification by default rules (cf. McDermott &amp; Doyle 1980; Reiter 1980), which should mirror cognitive processing (cf. e.g. Morreau 1988, Schubert &amp; Pelletier 1988). There seems to exist a preference order among the contextual conditions on the restriction or preservation of the unbounded extension of a MASs. This order is based on the degree of prototypicality of the respective reference function: the typicality of the NP's reference function is rendered by the strength which the modo ifiers have with respect to bounding or unbouudlng the entity's normal extension denoted by the noun.</Paragraph>
    <Paragraph position="3"> The component of default rules has been implemented in the CAT2 system and results in successful article translation. We relate our conceptual conditions to the CAT2 rules given in the annex.</Paragraph>
    <Paragraph position="4"> DeMing with different ways of reference, the conceptualization of entities will be in the foreground. Entities are denoted by nouns. We assume that words have a basic prototypical sense, which is represented in the lexicon. Senses of decreasing prototypicniity ate related to this prototypical sense by systematic metonymy rules.</Paragraph>
  </Section>
class="xml-element"></Paper>