<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-2063">
  <Title>Parsing for Grammar and Style Checking</Title>
  <Section position="3" start_page="368" end_page="369" type="ackno">
    <SectionTitle>
X3 PP
</SectionTitle>
    <Paragraph position="0"> We need some intermediate level here on which a notion like PP is already known, and precedence relations can be determined independent of the actual attachment of these PPs. This requires complex structural matching processes on the trees.</Paragraph>
    <Paragraph position="1"> Ambiguity of conjunction falls into the same class of problems: Here again, the parser will finally decide somehow, i.e. try to resolve the ambiguity e.g. of (10).</Paragraph>
    <Paragraph position="2"> 10. The data were input and output compatible null Again, the question is, how difficult this will be; and this can be expressed in terms of how many rules a conjunctional terminal node can feed (whether successfully or not). In order to know this, we have to examine the chart (as most of the rules tried will not have led to a successful parse) and mark the conjunction accordingly. null</Paragraph>
    <Section position="1" start_page="368" end_page="369" type="sub_section">
      <SectionTitle>
4.3 Subject-Object inversion
</SectionTitle>
      <Paragraph position="0"> This last example shows possible complex interactions in the area of style checking. In German, the direct object of a verb can be put before the subject, e.g. in (11), (12), (i3): 11. Den Mann hat er gesehen (the man has he seen) 12. Die Daten beschreibt das Programm (the data describes the program) 13. Daten beschreiben die Program- Die me (the data describe the programs) All these sentences are grammatical and have to be covered by the verb valency routines. Sometimes, however, the subject-object-conversion leads to unclear references (as in (13) where both NPs can be both subject and object). This is considered to complicate the process of text understanding. A style checker could flag these occurrences; but there is the following interference: If the grammar can recognize subject-objectinversion easily (as in (11)), then the reader  can do so as well, and a style checker should not flag anything. The cases which might be ambiguous for the reader, however, are ambiguous for the grammar as well; and as the parser uses certain heuristics to decide on subject and object in unclear cases, it might pick the wrong distribution and not flag anything, although it should do so in exactly this case. The result is that the checker's flagging is useless in the cases where the recognition is good, and that there is no flagging in the real important cases.</Paragraph>
      <Paragraph position="1"> This example shows that much fine tuning is necessary, to make a checking device a really useful tool and improve its value to users.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>