<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-2033">
  <Title>A t3ottom-up Generation for Principle-based Grammars Using Constraint Propagation</Title>
  <Section position="3" start_page="0" end_page="189" type="metho">
    <SectionTitle>
3 Principle-based Grammar Gen-
</SectionTitle>
    <Paragraph position="0"> eration pendent of data, it does not incur such a problem. Hasida has shown that sentence analysis and synthesis can be described following a simple program with a constraint, constitttent 3 struct( Category,X,Y); constituent(Category,X,Y). If active constraint solving techniques are applied to the problem, can this program be executed efficiently? Active constraint solving is equivalent to fold/unfold transformation \[Tud89\]. If the constraint clause is simply unfolded, then the number of clauses created will be the same as the number of lexical items. Presently this is not an efficient way 4.</Paragraph>
    <Paragraph position="1"> If passive constraint solving techniques are used, then how is data obtained? The answer is to predict the base lexical item, which is the core of a sentence, using a top-down prediction analogous to a bottom-up parsing technique.</Paragraph>
    <Paragraph position="3"/>
    <Section position="1" start_page="0" end_page="189" type="sub_section">
      <SectionTitle>
3.1 Generation Algorithm Using Con-
straint Propagation
</SectionTitle>
      <Paragraph position="0"> Natural language processing, such as sentence comprehension and production, is thought of as constraint satisfaction problems\[Has86\]. Constraint propagation techniques are very effective in these problems\[Din86\]. Constraint propagation techniques are classified into two methodologies: active constraiut, which transforms constraints into more efficient ones, and passive constraint, which is realized by the function such as freeze in Metalog. Passive constraint is similar to data-driven control, so if data does not arrive, the constraints are unsolved. As active constraint solving transforms constraints inde-This generation algorithm is sketched by the cu-PROLOG, or proiogIII notation ill figure 1. Tile predicate gen produces a sentence string from the term cat(P,F, Aa,Au, Sc, Sem,) 5, Tile term cat represents a set of features: P is tile feature for part-ofspeech, F is form such as verb inflection, Aa is adjacent node specification, Au is adjunct node specification, Sc is subcategorization information, and Sere is semantic information. The predicate prod anticipates BaseLez that is the core of a sentence(normally head verb), using part-of-speech and semantic information. BaseLez has lexical and feature information.</Paragraph>
      <Paragraph position="1"> The predicate genl gets a lexical item, and applies principles to the item and the base item until producing a sentence, getLez extracts a lexical item that is  constrained by principles 6 using semantic information. introduceFLez extracts an identity semantic item that is constrained by principles.</Paragraph>
    </Section>
    <Section position="2" start_page="189" end_page="189" type="sub_section">
      <SectionTitle>
3.2 Counterarguments to Bottom-up
Generation Deficiency
</SectionTitle>
      <Paragraph position="0"> C.onstraint propagation techniques in the previous section remedies nondeterminism of a bottom-up generation problem. An example of nondeterminism is the noun phrase generation in \[Shi89\]. If a NP occurs before a verb, different case NPs will be generated nondeterministically in figure 2.</Paragraph>
      <Paragraph position="2"> member(Daughter,Head.subcategorization).</Paragraph>
      <Paragraph position="3"> psv represents a Mother --, Daughter Head rule.</Paragraph>
      <Paragraph position="4"> The predicate member is used as a constraint, which says the Daughter node is a member of a subcategorization of tire Head node. The constraint propagation process is shown in figure 3.</Paragraph>
      <Paragraph position="5"> verb cor, st rair~s ,VPj due to its ~ubcate~orization l~t.</Paragraph>
    </Section>
  </Section>
  <Section position="4" start_page="189" end_page="189" type="metho">
    <SectionTitle>
~ verb
</SectionTitle>
    <Paragraph position="0"> Constraints can also eliminate the irrelevant phonolotgical expressions shown in \[Ca189\], and compactly express tire order-variant subcategorization list in JPSG.</Paragraph>
    <Paragraph position="1">  Suppose a Mother --~ Daughter Head rule, where</Paragraph>
    <Paragraph position="3"> and semantic function f destructively creates Mother.sem from Daughter.sere and Head.sere. That is, Daughter.sem and Head.sere cannot be predicted from Mother.sere. To get the value of Daughter.sere or Head.sere from Mother.sere, the inverse function f-1 is needed. The implicit assumption of inversability of the function\[Shi89\] is very severe, and a rather tricky feature structure must be constructed to escape the completeness problem. Therefore, it seems better to use another function in a semanticmonotonous framework instead of this one.</Paragraph>
    <Paragraph position="4"> However, we can easily modify the predicate predict to get another head using the semantic information to which the function is applied, if this inverse function is obtained.</Paragraph>
    <Paragraph position="5">  When using constraints to access fuuctlonal lexical items, an exhaustive search is not required. The insertion of Non-null constituents, such as case markers and fnnctional nouns, can be restricted using various constraints (syntactic, semantic information). For example, case markers that indicate the relationships between verbs and nouns are demanded by subcategori~ation information of the verbs. By using the constraint solving techniques, efficiency can be improved equal to or more than that of the top-down algorithm. Of Course since such occurrences(e.g, mdl constituents, or gap) cannot be restrained, this convenient mechanism cannot be used. However, this situation is the same as top-down algorithms.</Paragraph>
  </Section>
class="xml-element"></Paper>