<?xml version="1.0" standalone="yes"?>
<Paper uid="P86-1029">
  <Title>DONNELLAN'S DISTINCTION AND A COMPUTATIONAL MODEL OF REFERENCE</Title>
  <Section position="10" start_page="188" end_page="189" type="concl">
    <SectionTitle>
IMPLEMENTATION
</SectionTitle>
    <Paragraph position="0"> This paper is part of an extensive analysis of the referential/attributive distinction, which I use in the construction of a general model of reference \[13\]. My ultimate research objective is to provide s computational version of the reference model, then to incorporate it into a general plan-based account of definite and indefinite noun phrases. An experimental program that implements individuating ~ets has already been written.</Paragraph>
    <Paragraph position="1"> Called BERTRAND, this program interprets a small subset of English statements, and stores the information in its database, which it then uses to answer questions. Individuating sets are represented by an equivalence relation that holds among referring expressions: two referring expressions, R1 and R2, belong to the same individuating set if, according to the information interpreted so far, RI and R 2 denote the same object. In construeting individuating sets, BERTRAND uses a combination of logical and pragmatic strategies. The logical strategy exploits the fact that the relation &amp;quot;denote the same object&amp;quot; is symmetric, transitive, and closed under substitution. Thus, it  can be concluded that two referring expressions, RI and Rz, denote the same object (belong to the same individuating set) in one of the following ways: 5  1. Directly, when the statement &amp;quot;Rt is Rz ~ (or &amp;quot;R2 is RI ~) has been asserted.</Paragraph>
    <Paragraph position="2"> 2. Recursively using transitivity -- i.e., when, for a referring expression Rs, it can be shown that Rl and Rs, as well as Rs and Rz, belong to the same individuating set.</Paragraph>
    <Paragraph position="3"> 3. Recursively using substitution -- i.e., when Rl and Rz are  identical, except that Rl contains a referring expression subRl exactly where Rz contains a referring expression subRz, and 8ubRl and subR2 belong to the same individuating set.</Paragraph>
    <Paragraph position="4"> Note that, in the logical strategy, it is tacitly assumed that the relation of denoting the same object always holds between two identical tokens of referring expressions. This is obviously too strong an assumption for any realistic discourse: for example, two utterances of &amp;quot;The man&amp;quot; may very well denote two different people. On the other hand, the logical strategy fails to capture cases in which it is implied (although never actually asserted) that two distinct referring expressions denote the same thing. For example, &amp;quot;I met Marvin Maxwell yesterday. The man is utterly insane! ~ To compensate for these weaknesses, BERTRAND uses a strategy based on Grosz's notion of ffocus stack&amp;quot; \[8,10\]. In conceptual terms (and without going into details), it works as follows: a stack of individuating sets, representing objects that are &amp;quot;in focus,&amp;quot; is maintained throughout the &amp;quot;conversation.&amp;quot; When a new referring expression is interpreted, it is transformed into an open sentence D(z) with a single free variable z. s An individuating set I is said to subsume an open sentence S if S can be derived from I. The first individuating set in the focus stack to subsume D(z) represents the object denoted by the new referring expression. This solves the aforementioned problems: two occurrences of the same referring expression are considered as denoting the same object only if both are subsumed by the same individuating set in the focus stack, and two distinct referring expressions may still be considered as denoting the same object even though the logical strategy failed to show this, provided that both are subsumed by the same individuating set.</Paragraph>
    <Paragraph position="5"> Once the concept of an individuating set has been implemented, referring intentions can be represented as intentions to activate appropriate subsets of individuating sets. For example, the intention to use a conversationally relevant description can be represented as the plan to activate a subset of an individuating set that contains the term associated with the description. This is the topic of a current joint research effort with D. Appelt \[2\] to investigate the interaction that takes place between individuating sets and Appelt's four types of SWhat belongs to an individuating set, of course, is not a referring expression but the logical structure associated with it. For the sake of simplicity, however, I do not make this distinction here.</Paragraph>
    <Paragraph position="6">  where Xi is an &amp;quot;internal symbol&amp;quot; associated with Clty(y)&amp;By(y,Xi) , and )(j is associated with Bay(z).</Paragraph>
    <Paragraph position="7"> concept activation actions \[1\]. The next stage in the development of BERTRAND -- the implementation of referring intentions -- will be based on this research. In the final stage, individuating sets and referring intentions will be used to generate actual referring expressions.</Paragraph>
  </Section>
class="xml-element"></Paper>