<?xml version="1.0" standalone="yes"?>
<Paper uid="C73-1005">
  <Title>EBERHARD PAUSE A CLASS OF TRANSFOKMATIONAL RECOGNITION GRAMMARS</Title>
  <Section position="7" start_page="59" end_page="59" type="concl">
    <SectionTitle>
60 EBERHARD PAUSE
</SectionTitle>
    <Paragraph position="0"> In addition the following condition is important: the final derived tree of the input string (with respect to the generation process) could be reconstructed in the course of the analysis process. It must then be checked if some obligatory T-rule of G' is applicable to this tree. If this is not the case, the input string is a sentence of the grammar.</Paragraph>
    <Paragraph position="1"> The problems which arise out of the foregoing construction are the following: a) There are cases in which no reversible grammar G' for a given grammar G could be constructed, because, for instance, some auxiliary symbol could not be removed in the course of a sentence derivation.</Paragraph>
    <Paragraph position="2"> b) Another source of not obtaining a reversible grammar is, that there could occur subtrees of arbitrary depth which have to be considered for the exp=msion of some .term in a structural description. In this case a partially reversible grammar could be constructed. The recognition grammar then works with base rules, T-rules, and so-called &amp;quot;predictions &amp;quot; associated to the T-rules. 14 Analysis with these grammars will without doubt be much less efficient.</Paragraph>
    <Paragraph position="3"> The advantage in using reversible grammars appears to be a) that there is no additional synthesis phase necessary when the analysis step has been carried out, b) that the different possible paths which must be pursued in parsing a sentence could be reduced to a minimum, in using auxiliary symbols whenever it is possible, c) and that the recognition grammar could effectively be given' for the whole set of generated sentences. 15 14 See my dissertation (E. PAusE, 1972).</Paragraph>
    <Paragraph position="4"> 15 At page 54, Theorem (A), the number k must correctly be taken as the maximum number of base symbols occurring in a kernel tree of the set B.</Paragraph>
  </Section>
class="xml-element"></Paper>