<?xml version="1.0" standalone="yes"?>
<Paper uid="C88-1040">
  <Title>Robust parsing of severely corrupted spoken utterances</Title>
  <Section position="7" start_page="199" end_page="199" type="concl">
    <SectionTitle>
6 (Jonclnsmns and links with
</SectionTitle>
    <Paragraph position="0"> current research Experimer, ts show that the presence of jolly slots solvable as described above, beside permitting to successfully analy~e a much greater quota of word lattices, also speeds up parsing preventing it from being misled by false jollies. This well ::ompensatee for the growth of the inferential ~tctivity dlte to the relaxed temporM constraints in the Dis contMning ~holes'. As a consequence it is possible ~o use KS having chains of two or even three adjacent jolly slots without compromising excessively the global performai,,:es. This is a novel improvement over systems that, to our knowledge, only admit one single skippable word and use a more rigid linguistic knowledge representation \[Tomita 87\] or recognize any configuration of missing words but do not distinguish cases in which the information content of an absent word cart be ignored \[Goerz 83\]. An attracting feature of the present parsing technique is th;~t the KS activities are modularized into a set of operators. Consequently, it remains open to 'local' improvement~, on single operators as well as to overall heuristic adjustments on the score-guided control strategy. As aa exampi~, the response of the predicate JOLLY-TYPE of the oper~tor JVER3FY may be rendered more 'intelligent' by exploiting further information, such as estimates of the expected word length~ that has not been kept into consideration in the present implementation.</Paragraph>
    <Paragraph position="1"> A diff,!rent philosophy arising in very recent speech understanding research developments entrusts the problem of solving troublesome portions of the utterance (including those were jollies were not found) to a deeper ,%eoustlcal analysis guided by linguistic expectation \[Niedermair 87\]. Our approach is not in conflict, but rather, complementary to it. We believe that cornbining the two approaches would lead to a research area that should turn very fruitful in producing robust speech parsing.</Paragraph>
    <Paragraph position="2"> The authors wi~h to ezpress their gratitude to their colleague, the late Dr. SuBs, for Id8 contribution to the develotnnent of the system.</Paragraph>
  </Section>
class="xml-element"></Paper>