<?xml version="1.0" standalone="yes"?>
<Paper uid="A92-1015">
  <Title>Detecting and Correcting Morpho-syntactic Errors in Real Texts</Title>
  <Section position="8" start_page="117" end_page="117" type="concl">
    <SectionTitle>
7. Conclusions
</SectionTitle>
    <Paragraph position="0"> I have shown the feasibility of building a practical grammar-based spell checker that detects and corrects the important class of morpho-syntactic errors in normal texts (i.e., texts that have not been spedally prepared before processing). The system described in this paper is the first example of such a spell checker for Dutch. It is currently being tested at a large publishing company.</Paragraph>
    <Paragraph position="1"> I have demonstrated what can be expected of the approach I have taken. Depending on the complexity of the sentences, the combination of a word-level spell checker plus a syntactic parser performs from nearly perfect to satisfactory in regard to morpho-syntactic errors. Other types of errors cannot be handled reliably with the current framework, partly due to the permissive nature of both grammar and dictionary. However, enrichment of grammar and lexicon is only possible on an ad hoc basis. It will not lead to a systematic improvement of the correction process. Moreover, it is likely to interfere with the other components. Although many details still have to be worked out, the limits of this approach become visible. The next major improvement must come from analysis beyond syntax.</Paragraph>
  </Section>
class="xml-element"></Paper>