<?xml version="1.0" standalone="yes"?>
<Paper uid="A00-2018">
  <Title>A Maximum-Entropy-Inspired Parser *</Title>
  <Section position="7" start_page="137" end_page="138" type="concl">
    <SectionTitle>
6 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have presented a lexicalized Markov grammar parsing model that achieves (using the now standard training/testing/development sections of the Penn treebank) an average precision/recall of 91.1% on sentences of length &lt; 40 and 89.5% on sentences of length &lt; 100.</Paragraph>
    <Paragraph position="1"> This corresponds to an error reduction of 13% over the best previously published single parser results on this test set, those of Collins \[9\].</Paragraph>
    <Paragraph position="2"> That the previous three best parsers on this test \[5,9,17\] all perform within a percentage point of each other, despite quite different basic mechanisms, led some researchers to wonder if there might be some maximum level of parsing performance that could be obtained using the treebank for training, and to conjecture that perhaps we were at it. The results reported here disprove this conjecture. The results of \[13\] achieved by combining the aforementioned three-best parsers also suggest that the limit on tree-bank trained parsers is much higher than previously thought. Indeed, it may be that adding this new parser to the mix may yield still higher results.</Paragraph>
    <Paragraph position="3"> From our perspective, perhaps the two most important numbers to come out of this research are the overall error reduction of 13% over the results in \[9\] and the intermediateresult improvement of nearly 2% on labeled precision/recall due to the simple idea of guessing the bead's pre-terminal before guessing the head. Neither of these results were anticipated at the start of this research.</Paragraph>
    <Paragraph position="4"> As noted above, the main methodological innovation presented here is our &amp;quot;maximumentropy-inspired&amp;quot; model for conditioning and smoothing. Two aspects of this model deserve some comment. The first is the slight, but important, improvement achieved by using this model over conventional deleted interpolation, as indicated in Figure 2. We expect that as  we experiment with other, more semantic conditioning information, the importance of this aspect of the model will increase.</Paragraph>
    <Paragraph position="5"> More important in our eyes, though, is the flexibility of the maximum-entropy-inspired model. Though in some respects not quite as flexible as true maximum entropy, it is much simpler and, in our estimation, has benefits when it comes to smoothing. Ultimately it is this flexibility that let us try the various conditioning events, to move on to a Markov grammar approach, and to try several Markov grammars of different orders, without significant programming. Indeed, we initiated this line of work in an attempt to create a parser that would be flexible enough to allow modifications for parsing down to more semantic levels of detail. It is to this project that our future parsing work will be devoted.</Paragraph>
  </Section>
class="xml-element"></Paper>