<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-2930">
  <Title>First Workshop on Treebanks and Linguistic Theories</Title>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
1 Nivre's Parser
</SectionTitle>
    <Paragraph position="0"> Nivre (2003) proposed a dependency parser that creates a projective and acyclic graph. The parser is an extension to the shift-reduce algorithm. As with the regular shift-reduce, it uses a stack S and a list of input words W. However, instead of finding constituents, it builds a set of arcs G representing the graph of dependencies.</Paragraph>
    <Paragraph position="1"> Nivre's parser uses two operations in addition to shift and reduce: left-arc and right-arc. Given a sequence of words, possibly annotated with their part of speech, parsing simply consists in applying a sequence of operations: left-arc (la), right-arc (ra), reduce (re), and shift (sh) to the input sequence.</Paragraph>
  </Section>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
2 Parsing an Annotated Corpus
</SectionTitle>
    <Paragraph position="0"> The algorithm to parse an annotated corpus is straightforward from Nivre's parser and enables us to obtain, for any projective sentence, a sequence of actions taken in the set {la,ra,re,sh} that parses it. At a given step of the parsing process, let TOP be the top of the stack andFIRST, the first token of the input list, and arc, the relation holding between a head and a dependent.</Paragraph>
    <Paragraph position="1">  1. if arc(TOP,FIRST) [?] G, then ra; 2. else if arc(FIRST,TOP) [?] G, then la; 3. else if [?]k [?] Stack,arc(FIRST,k) [?] G or arc(k,FIRST) [?] G, then re; 4. else sh.</Paragraph>
    <Paragraph position="2">  Using the first sentence of the Swedish corpus as input (Table 1), this algorithm produces the sequence of 24 actions: sh, sh, la, ra, re, la, sh, sh, sh, la, la, ra, ra, sh, la, re, ra, ra, ra, re, re, re, re, and ra (Table 2).</Paragraph>
  </Section>
  <Section position="5" start_page="0" end_page="207" type="metho">
    <SectionTitle>
3 Adapting Nivre's Algorithm to
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="206" type="sub_section">
      <SectionTitle>
Machine-Learning
3.1 Overview
</SectionTitle>
      <Paragraph position="0"> We used support vector machines to predict the parse action sequence and a two step procedure to  skapet och familjen ar en gammal institution, som funnits sedan 1800-talet 'Marriage and family are an old institution that has been around from the 19th  produce the graph. We first ran the classifier to select unlabeled actions, la, ra, sh, re. We then ran a second classifier to assign a function to ra and la parse actions.</Paragraph>
      <Paragraph position="1"> We used the LIBSVM implementation of the SVM learning algorithm (Chang and Lin, 2001). We used the Gaussian kernel throughout. Optimal values for the parameters (C and g) were found using a grid search. The first predicted action is not always possible, given the parser's constraints. We trained the model using probability estimates to select the next possible action.</Paragraph>
    </Section>
    <Section position="2" start_page="206" end_page="206" type="sub_section">
      <SectionTitle>
3.2 Feature Set
</SectionTitle>
      <Paragraph position="0"> We used the following set of features for the classifiers: null  * Word and POS of TOP and FIRST * Word and POS of the second node on the stack * Word and POS of the second node in the input list * POS of the third and fourth nodes in the input list  leftmost child of FIRST to FIRST, if any For the POS, we used the Coarse POS, the Fine POS,and all the features (encoded as boolean flags). We did not use the lemma.</Paragraph>
      <Paragraph position="1">  och familjen ar en gammal institution, som funnits sedan 1800-talet.</Paragraph>
      <Paragraph position="2"> Ac. Top word First word Rel.</Paragraph>
      <Paragraph position="3"> sh nil Aktenskapet sh Aktenskapet och la och familjen ++ ra Aktenskapet familjen CC re familjen ar la Aktenskapet ar SS sh nil ar sh ar en sh en gammal la gammal institution AT la en institution DT ra ar institution SP ra institution , IK sh , som la som funnits SS re , funnits ra institution funnits ET ra funnits sedan TA ra sedan 1800-talet PA re 1800-talet .</Paragraph>
      <Paragraph position="4"> re sedan .</Paragraph>
      <Paragraph position="5"> re funnits .</Paragraph>
      <Paragraph position="6"> re institution .</Paragraph>
      <Paragraph position="7"> ra ar . IP 4 Extensions to Nivre's Algorithm</Paragraph>
    </Section>
    <Section position="3" start_page="206" end_page="207" type="sub_section">
      <SectionTitle>
4.1 N-best Search
</SectionTitle>
      <Paragraph position="0"> We extended Nivre's original algorithm with a beam search strategy. Foreach action, la,ra,shandre,  we computed a probability score using LIBSVM. These scores can then be used to carry out an N-best search through the set of possible sequences of actions.</Paragraph>
      <Paragraph position="1"> We measured the improvement over a best-first strategy incrementing values of N. We observed the largest difference between N = 1 and N = 2, then leveling off and we used the latter value.</Paragraph>
    </Section>
    <Section position="4" start_page="207" end_page="207" type="sub_section">
      <SectionTitle>
4.2 Bidirectionality and Voting
</SectionTitle>
      <Paragraph position="0"> Tesniere (1966) classified languages as centrifuge (head to the left) and centripetal (head to the right) in a table (page 33 of his book) that nearly exactly fits corpus evidence from the CONLL data. Nivre's parser is inherently left-right. This may not fit all the languages. Some dependencies may be easier to capture when proceeding from the reverse direction. Jin et al. (2005) is an example of it for Chinese, where the authors describe an adaptation of Nivre's parser to bidirectionality.</Paragraph>
      <Paragraph position="1"> We trained the model and ran the algorithm in both directions (left to right and right to left). We used a voting strategy based on probability scores.</Paragraph>
      <Paragraph position="2"> Each link was assigned a probability score (simply by using the probability of the la or ra actions for each link). We then summed the probability scores of the links from all four trees. Toconstruct a singlehead, rooted, and cycle-free tree, we finally applied the Chu-Liu/Edmonds optimization algorithm (Chu and Liu, 1965; Edmonds, 1967).</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>