<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1110">
  <Title>Advances in Discriminative Parsing</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Discriminative machine learning methods have improved accuracy on many NLP tasks, including POS-tagging, shallow parsing, relation extraction, and machine translation. Some advances have also been made on full syntactic constituent parsing.</Paragraph>
    <Paragraph position="1"> Successful discriminative parsers have relied on generative models to reduce training time and raise accuracy above generative baselines (Collins &amp; Roark, 2004; Henderson, 2004; Taskar et al., 2004). However, relying on information from a generative model might prevent these approaches from realizing the accuracy gains achieved by discriminative methods on other NLP tasks. Another problem is training speed: Discriminative parsers are notoriously slow to train.</Paragraph>
    <Paragraph position="2"> In the present work, we make progress towards overcoming these obstacles. We propose a flexible, end-to-end discriminative method for training parsers, demonstrating techniques that might also be useful for other structured prediction problems.</Paragraph>
    <Paragraph position="3"> The proposed method does model selection without ad-hoc smoothing or frequency-based feature cutoffs. It requires no heuristics or human effort to optimize the single important hyper-parameter.</Paragraph>
    <Paragraph position="4"> The training regime can use all available information from the entire parse history. The learning algorithm projects the hand-provided features into a compound feature space and performs incremental feature selection over this large feature space. The resulting parser achieves higher accuracy than a generative baseline, despite not using a generative model as a feature.</Paragraph>
    <Paragraph position="5"> Section 2 describes the parsing algorithm. Section 3 presents the learning method. Section 4 presents experiments with discriminative parsers built using these methods. Section 5 compares our approach to related work.</Paragraph>
  </Section>
class="xml-element"></Paper>