<?xml version="1.0" standalone="yes"?>
<Paper uid="J89-3002">
  <Title>KNOWLEDGE REPRESENTATION FOR COMMONSENSE REASONING WITH TEXT</Title>
  <Section position="24" start_page="154" end_page="154" type="ackno">
    <SectionTitle>
NOTES
</SectionTitle>
    <Paragraph position="0"> 1. Dahlgren and McDowell are the main investigators. Stabler contributed the first-order logic and problem solver.</Paragraph>
    <Paragraph position="1"> 2. Nominalized verbs are treated as verbal concepts.</Paragraph>
    <Paragraph position="2"> 3. We are exlLremely grateful to Hajime Wada who wrote the  original version of the DRT module to provide DRSs for a wide range of syntactic constructions. The present DRT module is an extension of his work. The lexical entries that form the NS data bases are the result of the careful, diligent efforts of Carol Lord, Robert Hagiwara, and Susan Mordechay. Susan Hirsh contributed the programs that generate the FOL data bases and performed other programming tasks.</Paragraph>
    <Paragraph position="3"> 4. The atoms subj, obj, obliq, pobj, which appear in the generic representations, signal which element of the sentence is to be accessed for the output response. In answering questions, the generic representations are mapped to a small set of canonical sentences whose slots are filled with elements from the input query. Consider the feature representation implies(merchandise (obj)), which is part of the generic entry for buy. If the query What is implied if a man buys a truck? is processed, since the direct object of buy in the query is truck, the response is the truck is merchandise. If the query had been What is implied if John buys a house?, the response would be the house is merchandise.</Paragraph>
    <Paragraph position="4"> 5. The analysis work was originally reported in Dahlgren and McDowell (1986b). We review that here and also report on the implementation.</Paragraph>
    <Paragraph position="5"> 6. For the most part, examples given in this paper are modified versions of actual sentences from the WSJ corpus that the KT researchers are using.</Paragraph>
    <Paragraph position="6"> 7. For readability, the remaining generic entries will be shown in English paraphrase rather than as they are coded.</Paragraph>
    <Paragraph position="7"> 8. The question arises as to how natural or likely such sentences would be in use. For example, would we be more likely to encounter a letter from all the lawyers rather than a letter from every lawyer. The standard answer, which we adopt, is that semantics is not a predictive theory. We can't tell what a person will say, but we have to be able to interpret whatever is said. We can restrict ourselves to likely expressions, but then we are putting ourselves in the position of predicting what is likely, and we might be wrong. We prefer to try to interpret what is possible, even if unlikely.</Paragraph>
    <Paragraph position="8"> 9. A full theoretical discussion of the issues involved with medals can be found in McDowell (1987).</Paragraph>
    <Paragraph position="9"> 10. The main predicate of a clause, whether it be a verb or a term that is the complement of the copula, carries a DR-theoretic event-type reference marker as its tense argument. Other predicates carry an argument which is linked to tense, but which is not a reference marker in DR-theoretic terms.</Paragraph>
    <Paragraph position="10"> 11. Actually, sentences interpreted as generics in English have a complex combination of the following features: indefinite NPs, present tense, copula, and/or inherently stative verb (but not the progressive). Not all of these features are always present in all generically interpreted sentences.</Paragraph>
    <Paragraph position="11"> 12. In ~Lhis and the following sections we report on work in progress that is not yet fully implemented.</Paragraph>
    <Paragraph position="12"> Computational Linguistics, Volume 15, Number 3, September 1989 170</Paragraph>
  </Section>
class="xml-element"></Paper>