<?xml version="1.0" standalone="yes"?>
<Paper uid="C88-2132">
  <Title>Island Parsing and Bidirectional Charts</Title>
  <Section position="9" start_page="638" end_page="638" type="concl">
    <SectionTitle>
BOSS WANTS AN IMMEDIATE CALL TO MILAN.
</SectionTitle>
    <Paragraph position="0"> For clarity's sake, we shall consider vertices univocally detected and lexical interpretations unambiguous. Of course we ~hall consider words of class a (islands) and of class b.</Paragraph>
    <Paragraph position="1"> Starting from MILAN we get to produce an inactive edge with cat =PP, between vertices 7 and 9, and an active edge i with eat = S, relative to rule 1, once more between vertices !7 and 9, with fromposition-- 3.</Paragraph>
    <Paragraph position="2"> When the word BOSS is analyzed we get to produce an active edge with cat=S, relative to rule I , and with toposition = 2.</Paragraph>
    <Paragraph position="3">  A top-down process is needed after that because a phrase occurs without any islands in it. Through that process a noun phrase is recognized between vertices 4 and 7, so that we have the situation shown in Figure 1.</Paragraph>
    <Paragraph position="4"> At this point, by virtue of the A-I composition rule an active edge with cat=S can be inserted into the chart between vertices 1 and 7 as shown in Figure 2.</Paragraph>
    <Paragraph position="5"> The last step consists in introducing an inactive edge with&amp;quot; Cat = S into the chart between vertices 1 and 9, by virtue of the A-A composition rule, as shown in Figure 3. This yields a succesful recognition of the sentence.</Paragraph>
    <Paragraph position="6"> Conclusions A mechanism that extends the chart algorithm with bidirectionality has been introduced. This step is a major one, since a monodirectional chart would not be able to 64J~ base its processing selectively on easily identifiable fragments (the so called islands) and derive hypotheses about the other parts of the input string. Instead, with the mechanism proposed here, for any place where the easily identifiable fragments occur in the sentence, the process will extend to both the left and the right of the islands, until possibly completely missing fragments are reached. At that point, by virtue of the fact that both a left and a right context were found, heuristics can be introduced that predict the nature of the missing fragments.</Paragraph>
    <Paragraph position="7"> The described mechanism is particularly advantageous when dealing with complex sentences, because it is an inherently nondeterministie mechanism, capable of dealing with the complex local ambiguity typical of natural language. An important aspect is that the mechanism is completely independent of the particular linguistic theory adopted. In technical terms, the * linguistic representation is reflected only in the particular functional description, and in its particular operations, which are added to the edges of the chart and will provide the necessary information for constraining the process and allowing better predictiions.</Paragraph>
    <Paragraph position="8"> The use of bidirectional charts seems to be particularly suitable fiPr speech recognition, but also for processing other forms of ill-formed input; lastly, it seems particularly suited even for processing well formed strings when combined with a head-driven linguistic theory, i.e. a theory that privileges particular elements inside constituents \[see for instance Stock 1986\].</Paragraph>
  </Section>
class="xml-element"></Paper>