<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-3088">
  <Title>Parsing Long English Sentences with Pattern Rules</Title>
  <Section position="4" start_page="410" end_page="411" type="metho">
    <SectionTitle>
4. Examples
</SectionTitle>
    <Paragraph position="0"> Example I. The following sentence matches the pattern (%vl ving %v2, (closure (ving %v3,)) C ving %v4) where &amp;quot;c&amp;quot; is a conjunction.</Paragraph>
    <Paragraph position="1"> The air-use plans may be used as the basic framework for achieving the desired air quality by such means as limiting the emissions from individual sources, limiting the emissions from sources in certain areas, or disallowing new pollutant sources in overburdened areas.</Paragraph>
    <Paragraph position="2"> Example 2. The following sentence matches the pattern (In order INF, SDEC). The syntactic tree is the combination of the trees of INF and SDEC.</Paragraph>
    <Paragraph position="3"> In order to evaluate or rank land use plans in terms of air quality, it is necessary for planners to be able to project emission density using only planning variables, because detailed source characteristics are not available at the time alternative plans are being developed and evaluated.</Paragraph>
    <Paragraph position="4"> Example 3. The pattern (SDEC, because SDEC) can be used to parse the following sentence. The corresponding Chinese sentence can be obtained from the combination of the two Chinese segments for the two SDECs.</Paragraph>
    <Paragraph position="5"> Epa's preferred decision is to approve and support funding for the proposed alternative, because this is the most cost-effective way of achieving federal and state water-quality goals, improving the quality of the rio grande, and protecting prime a~ricultural lands.</Paragraph>
    <Paragraph position="6"> Example 4.</Paragraph>
    <Paragraph position="7"> For some special sentence patterns, they have specific chinese translation. The following is a pattern rule.</Paragraph>
    <Paragraph position="8"> (Not that SDEC but that SDEC)  The results from parsing the environmental protection corpus mentioned in i. are as follows.</Paragraph>
    <Paragraph position="9">  For ERSO-ECMT, the rate of correct matching is about one third of the total long sentence. It can be improved by organizing more rules to cover a large range of sentences, but for some long sentences without apparent features, there is no pattern rules for them.</Paragraph>
  </Section>
  <Section position="5" start_page="411" end_page="411" type="metho">
    <SectionTitle>
6. Discussion
</SectionTitle>
    <Paragraph position="0"> In fact, the problems arising from parsing long English sentence are the combined effects of some problems, which could not be treated quite well, such as prepositional phrase attachment problems\[6\], compound noun phrases. They are all about the problems of the semantic relations among words or segments in sentences. At present the ways for encoding and using massive semantic information for a practical machine translation system of some domains, computer science and environmental protection, for example, are not clear. Basically the approach here is not to solve all the problems, but to break the sentence into some segments within manageable size and then to parse them separately.</Paragraph>
    <Paragraph position="1"> In this way, since each segment is shorter than the original sentence, failing to construct a correct parsing result usually. affects only a shorter range of words, but it wastes time in parsing some pattern which is found not appropriate for the input sentence eventually. In order not to do pattern matching so much, which is time-consuming, some patterns can be put under the word, which is the leading word of the pattern, of the dicionary, and some under the augmented context free grammar rules of appropriate nonterminals to guide the parser before the parsing mechanism is initiated. The parser will check to see if the sentence is a particular pattern by looking at the pattern rule encountered in the dictionary or in the grammar rules in course of parsing and then try to parse the other parts of th~ sentence with the pattern rule found if any.</Paragraph>
  </Section>
  <Section position="6" start_page="411" end_page="411" type="metho">
    <SectionTitle>
7~ Conclusion
</SectionTitle>
    <Paragraph position="0"> For short sentence, a good MT system can usually generate readable translation, whereas for long sentence the translation is usually not satisfactory. A practical method is presented for parsing long English sentence. It bases on some patterns of long English sentences. The patterns can be inserted in the lexicon or the augmented context free grammar to guide the parser.</Paragraph>
  </Section>
  <Section position="7" start_page="411" end_page="412" type="metho">
    <SectionTitle>
8. Acknowledgement
</SectionTitle>
    <Paragraph position="0"> The paper is a partial result of the project No. 3131500 conducted by the ITRI  under sponsorship of the Minister of Economic</Paragraph>
  </Section>
class="xml-element"></Paper>