<?xml version="1.0" standalone="yes"?>
<Paper uid="W99-0621">
  <Title>A Learning Approach to Shallow Parsing*</Title>
  <Section position="3" start_page="168" end_page="168" type="intro">
    <SectionTitle>
I
</SectionTitle>
    <Paragraph position="0"> terior of a phrase or not, and then groups the words into phrases. The second instantiation finds the borders of phrases (beginning and end) and then pairs !them in an &amp;quot;optimal&amp;quot; way into different phrases. These problems formulations are similar to those studied in (Ramshaw and Marcus, 1995) and (Church, 1988; Argamon et al., 1998), respectively.</Paragraph>
    <Paragraph position="1"> The experimental results presented using the SNoW based approach compare favorably with previously published results, both for NPs and SV phrases. A s important, we present a few experiments that shed light on some of the issues involved in using learned predictors that interact to produce the desired inference. In particular, we exhibit the contribution of chaining: features that are generated as the output of one of the predictors contribute to the performance of another predictor that uses them as its input. Also, the comparison between the two instantiations 0f the learning paradigm - the Inside/Outside and the Open/Close - shows the advantages of the Open/Close model over the Inside/Outside, especially for the task of identifying long sequences.</Paragraph>
    <Paragraph position="2"> The contribtition of this work is in improving the state of the art in learning to perform shallow parsing tasks, developing a better understanding for how to model these tasks as learning problems and in further studying the SNoW based computational paradigm that, we believe, can be used in many other related tasks in NLP.</Paragraph>
    <Paragraph position="3"> The rest of this paper is organized as follows: The SNoW architecture is presented in Sec. 2.</Paragraph>
    <Paragraph position="4"> Sec. 3 presents the shallow parsing tasks studled and provides details on the computational approach. Sec. 4 describes the data used and the experimental approach, and Sec. 5 presents and discusses the experimental results.</Paragraph>
  </Section>
class="xml-element"></Paper>